JP4118680B2 - Protein three-dimensional structure construction method - Google Patents

Description

ここでＵｌｅｎは、配列上隣の残基およびＣｙｓ残基のペアのＣα原子間の距離に関するもので下記式（２）のように設定される。

ここでＤｉ，ｉ＋１は残基ｉと残基ｉ＋１のＣα間距離である。ＤｉＳＳはジスルフィド結合を形成するＣｙｓ残基のペア同士の距離である。Ｋ１とＫＳＳは定数でありそれぞれ２および５と設定される。
ＵａｎｇはＣα原子の結合角の関数であり下記式（３）のとおりである。

ここでθｉ（ｒａｄ）はｉ，ｉ＋１，ｉ＋２番目の残基Ｃα原子の角度である。θ０はＰＤＢのＸ線構造から（１００／１８０）・π（ｒａｄ）と設定される。Ｋａは定数であり１とする。
Ｕｐｏｓは、Ｃα原子の位置に関する関数であり、下記式（４）のとおりである。

ここで‖・‖が意味する所はノルムであり、Ｍｉは構造を基にしたアライメント上で構造的に等価な位置にあるＣα原子間の平均距離である。残基ｉについてＭｉの値が求められないとき、Ｍｉの値は１０と設定される。ここでは、Ｃα原子の平均座標であり下記式（５）のとおりである。

ここでＸｊｉはｊ番目の参照タンパク質のｉ番目の残基のＣαの原子座標である。ｗｊｉは、ｊ番目の参照タンパク質のｉ番目の残基の重みである。この重みは目的タンパク質の大体の形を決定するため重要なパラメータであるが、これはローカルスペースホモロジー（ＬＳＨ）と呼ばれる着目部位の１２Å以内の空間的近傍の局所的な値によって決定している（第４図参照）。ＬＳＨと構造がよく保存されている部位（ＳＣＲｓ：Ｓｔｒｕｃｔｕｒａｌ Ｃｏｎｓｅｒｖｅｄ Ｒｅｇｉｏｎｓ）に存在する残基のペアの比率との相関は第５図に示されているように非常に高い。これは、高いＬＳＨ値を持つときは統計的にＣα原子の位置が参照タンパク質構造と比べて１．０Å以内にあることを意味する。Ｕｖｄｗは下記式（６）のとおりである。

ここでＫｖｄｗは０．０１（Ｄｉ，ｊ＜３．２Å）と０．００１（Ｄｉ，ｊ＞３．２Å）と設定され６Åをカットオフ値とした。
Ｃα原子は式（１）に従って、シミュレーティッドアニーリング法を用いて最適化される。この最適化の段階でＣα原子の摂動は１．０Å以内になるように設定する。またこのアニーリングの段階は全てのＣα原子について、１００回づつ計算される。そして、温度に相当するパラメータは、２５から０．５回ごとに０．０１減らし、そのパラメータは以後一定とする。
この大きな２つの段階、構造情報の取得とＣα原子の構築は１０回繰り返され、最小の目的関数値をもつＣα原子の座標が最適解として算出される。
ステップＩ−８２：主鎖原子座標の構築及び最適化
ステップＩ−８１（１）のＣαの原子座標に主鎖の他の原子を付加し、シミュレーティッドアニーリング法によって目的関数を最小化するようにする。まず、Ｃα原子の立体的な重ねあわせを行い、Ｃαの原子間距離が２．５Å以下の残基が取り上げられる。Ｃαを除く主鎖の原子座標はＣα原子間距離が最小になるように参照タンパク質の座標から取得しモデル構造とする。
参照タンパク質の中に相当する残基が無い場合、主鎖の原子座標はデータベース中の相当する４残基のタンパク質断片から作成される。この過程の中で、残基ｉの主鎖原子はｉ−１番目からｉ＋２番目までのＣα原子間の最小のｒｍｓｄ値を持つ残基から選ばれる。その際Ｎ末端の残基では、Ｃα原子座標の重ね合わせ範囲がｉ番目からｉ＋３番目までとなり、Ｃ末端の残基およびそのひとつ前の残基では同様にｉ−３番目からｉ番目までおよびｉ−２番目からｉ＋１番目までとなる。
主鎖原子の目的関数を基にシミュレーティッドアニーリング法によって主鎖原子座標が最適化される。
目的関数は下記式（７）のとおりである。

となる。
Ｕｂｏｎｄは下記式（８）のとおりである。

ここでｂｉ０は、標準の結合長でありそれぞれの化学結合の種類によって異なる。Ｋｂは定数であり２２５とする。
Ｕａｎｇは結合角の関数で、下記式（９）のとおりである。

ここでθｉはｉ番目の結合角であり、化学結合の種類によって異なる。Ｋａは定数で４５と設定される。
Ｕｎｏｎ−ｂｏｎｄは非結合の相互作用の関数で、下記式（１０）のとおりである。

ここでεｉ，ｊとｒｉ，ｊ＊は定数で原子の種類によって異なる。
Ｋｎｏｎは定数で０．２５とし、カットオフは８Åとする。
ＵＳＳはＣｙｓ残基が生成するジスルフィド結合の関数で、下記式（１１）のとおりである。

ここでＫＳＳＣαおよびＫＳＳＣβは定数であり７．５である。
Ｕｐｏｓは原子の位置に関する関数で、下記式（１２）のとおりである。

ここで＜ＷｉＸｉ＞は下記式（１３）のように与えられる。

式（１２）の＜ＷｉＸｉ＞は、目的タンパク質および参照タンパク質の間の構造の重ねあわせから求める。
Ｋｐｏｓは定数であり０．３である。
Ｕｔｏｒは主鎖のねじれ角のものであり、下記式（１４）のとおりである。

ここでφｉ０とφｉ０はＲａｍａｃｈａｎｄｒａｎマップ上での最も近いねじれ角のθｉおよびψｉとする。またωｉ０は０としてｃｉｓ−Ｐｒｏ残基の場合のみπ（ｒａｄｉａｎ）とする。ＫｔおよびＫωは定数であり、それぞれ１０および５０とする。
ＵｃｈｉはＣαのキラリティーに関するものであり、下記式（１５）のとおりである。

ここでτｉはＮ−Ｃα−Ｃβ−Ｃで定められるねじれ角でありＫｃｈｉは５０とする。Ｕｈｙｄｒはホモロガスなタンパク質中で保存された主鎖の水素結合に関するもので、下記式（１６）のとおり定められる。

水素結合は、Ｎ原子とＯ原子の距離が２．９±０．５Åにあるときに設定される。複数の参照タンパク質中で水素結合があるか否かを判定するときは、７５％以上の参照タンパク質が存在すると認めた場合に水素結合ありと判定する。Ｋｈｙｄｒは定数であり０．６である。
次にＣβを含む主鎖原子の最適化がシミュレーティッドアニーリングによって行われる。このアニーリングの過程で主鎖とＣβの原子の摂動が初期の位置に対して１．０Å以内になるようにする。このアニーリングの段階は主鎖とＣβの原子に対して２００回行われる。温度に相当するパラメータは５０もしくは２５から始まり一回毎に０．５倍にしてゆき０．０１になるまで続け、その後一定値とする。
主鎖の立体配置を幅広くサンプリングするために、本発明の方法では、好ましくは上記の方法を６回行い、最小の目的関数値を持つ主鎖の原子座標を最適解とする。そして、温度に相当するパラメータは、はじめの２回は５０からスタートして３回目から２５からスタートすることとする。
ステップＩ−８３：側鎖原子座標の構築及び最適化
側鎖の構築は、大きく２段階に分かれており、「構造保存部位の側鎖構築」（スッテプＩ−８３（１））と「全体の側鎖構築」（スッテプＩ−８３（２））に分けられる。
ステップＩ−８３（１）：構造保存部位の側鎖構築
算出された主鎖原子に対して、以前の研究における方法を用いてホモロガスなタンパク質から側鎖のねじれ角を得る。この方法の詳細は、″Ｔｈｅ ｒｏｌｅ ｏｆ ｐｌａｙｅｄ ｂｙ ｅｎｖｉｒｏｎｍｅｎｔａｌ ｒｅｓｉｄｕｅｓ ｉｎ ｓｉｄｅ−ｃｈａｉｎ ｔｏｒｓｉｏｎａｌ ａｎｇｌｅｓ ｗｉｔｈｉｎ ｈｏｍｏｌｏｇｏｕｓ ｆａｍｉｌｉｅｓ ｏｆ ｐｒｏｔｅｉｎｓ：Ａ ｎｅｗ ｍｅｔｈｏｄ ｏｆ ｓｉｄｅ ｃｈａｉｎ ｍｏｄｅｌｉｎｇ．″Ｏｇａｔａ Ｋ ａｎｄ Ｕｍｅｙａｍａ Ｈ，Ｐｒｏｔ．Ｓｔｒｕｃｔ．Ｆｕｎｃｔ．Ｇｅｎｅｔ．１９９８，３１，２５５−３６９に記載されている。
この方法の中でホモロガスなタンパク質の中で保存されている側鎖の割合を算出し、この情報を基にして側鎖のモデリングを行う。側鎖の保存された部位の側鎖の原子座標は固定した主鎖原子に対して置かれる。例えば、ホモロガスなタンパク質中でアルギニン残基のχ１角が保存されていれば、Ｃγ原子の座標を置くことができ、Ｐｈｅ残基でχ１とχ２角が保存されていれば、全ての側鎖原子を置くことができる。式（７）を用いたシミュレーティッドアニーリングの最適化の過程は、主鎖とＣβの原子のみ行われて、原子の摂動は１．０Å以内となるようにした。この主鎖とＣβの原子のアニーリングの段階は２００回行われる。そして、温度に相当するパラメータは２５からスタートして一回毎に０．５倍にしてゆき０．０１になるまで小さくなるようにする。式（７）中のＵｎｏｎ−ｂｏｎｄは主鎖原子と部分的に作成された側鎖原子について行われる。そのとき側鎖原子の座標は最適化の過程を通じて保存されるようにする。
構造の情報であるＭｉと水素結合のＮ−Ｏのペアは最適化の過程で用いられる。主鎖原子の配置を得るために、上記プロセスを３回繰り返し、目的関数の最小の主鎖原子の座標を算出構造とする。
ステップＩ−８３（２）：全体の側鎖の構築
側鎖の構築は固定した主鎖およびＣβ原子のもとで行う。これは上記したＯｇａｔａ Ｋ ａｎｄ Ｕｍｅｙａｍａ Ｈ，Ｐｒｏｔ．Ｓｔｒｕｃｔ．Ｆｕｎｃｔ．Ｇｅｎｅｔ．１９９８，３１，２５５−３６９に開示されている研究成果をもって行われ、それを用いることにより短時間で正確なモデルを与えることができる。次に主鎖構造は低温におけるモンテカルロ法によって最適化され、温度は０．００１に設定され式（７）の目的関数Ｕｎｏｎ−ｂｏｎｄを用い、全ての主鎖と側鎖の原子で計算される。そして、Ｎ、Ｃα、Ｃ、Ｃβ原子の最適化の過程で側鎖のねじれ角を最適化された状態を保つように側鎖の座標を再配置する。原子の摂動は０．５Å以内とする。次に側鎖は削除され、上記の側鎖構築が繰り返される。このプロセスは２．４Åの原子同士のぶつかり合いがなくなり、且つＮ−Ｃα−Ｃβ−Ｃのねじれ角が−１２０±１５°の範囲に収まるまで繰り返される。
ステップＩ−９０：立体構造の完成
上記の通りステップＩ−７０において得られたアライメントを基にステップＩ−８０においてＦＡＭＳ等のモデリングソフトを用いて立体構造を構築し、モデルが完成する。上記ステップＩ−８０で示す方法をＦＡＭＳと称することがある。この方法の詳細は、Ｋｏｊｉ Ｏｇａｔａ ａｎｄ Ｈｉｄｅａｋｉ Ｕｍｅｙａｍａ，″Ａｎ ａｕｔｏｍａｔｉｃ ｈｏｍｏｌｏｇｙ ｍｏｄｅｌｉｎｇ ｍｅｔｈｏｄ ｃｏｎｓｉｓｔｉｎｇ ｏｆ ｄａｔａｂａｓｅ ｓｅａｒｃｈｅｓ ａｎｄ ｓｉｍｕｌａｔｅｄ ａｎｎｅａｌｉｎｇ″Ｊｏｕｒｎａｌ ｏｆ Ｍｏｌｅｃｕｌａｒ Ｇｒａｐｈｉｃｓ ａｎｄ Ｍｏｄｅｌｉｎｇ １８，２５８−２７２，２０００に記載されている。
ステップＩ−１００：立体構造の検査
ステップＩ−１００において、出来上がった立体構造モデルの疎水コアが形成されているかどうかを検査し、ステップＩ−１１０の最終構造に至る。
ステップＩ−１１０：最終構造の構築
かくして目的タンパク質の立体構造を規定する原子座標を得ることができる。
本発明の方法（アライメント修正法）によれば、ホモロジーモデリングによるタンパク質の立体構造構築において、アミノ酸配列の並置（アライメント）が与えられた場合において、参照タンパク質の座標から立体構造情報を得てアライメントが修正されるため、より正確な立体構造の構築が可能となる。アライメントの修正は、タンパク質の二次構造情報、疎水コア値情報、疎水コア距離情報等に基づいて行われ、これによりタンパク質立体構造の構成単位である二次構造およびその近傍にアミノ酸配列断片の挿入および欠損が入らないモデルを得ることができる。また本発明の方法は、アライメントを修正する際に、タンパク質の立体構造を形成する物理的な疎水性相互作用を考慮した方法であり、疎水性相互作用の働く空間およびその近傍にアミノ酸配列断片の挿入および欠損が入らないようにすることが可能である。
ＩＩ．Ａｄｖａｎｃｅｄ ＦＡＭＳによる立体構造の構築法
次に、本発明の第二および第三の態様であるＡｄｖａｎｃｅｄ ＦＡＭＳについて説明する。
第６図は、本発明のＡｄｖａｎｃｅｄ ＦＡＭＳによる立体構造構築方法の一例を示すフローチャートである。以下、第６図の各ステップを例としてこの発明を詳述する。
ステップＩＩ−１０：目的タンパク質のアミノ酸配列
先ず、立体構造の構築対象となる目的タンパク質のアミノ酸配列を入力する。用いられるアミノ酸配列としては、上記で詳述した通り、データベースに登録されているもの、配列が始めて解析されたもの等如何なる由来の配列であってもよい。また部分構造のみが解析されているタンパク質の配列も完全な立体構造の情報を得るために、本発明における立体構造の構築対象とすることができる。
ステップＩＩ−２０：高ホモロジー用アライメントソフトによるデータベース検索及び配列アライメント
ステップＩＩ−１０において入力されたアミノ酸配列を、高ホモロジー用アライメントを行うソフトウエア、例えばＦＡＳＴＡを用いてアライメントを行う。即ち、ＦＡＳＴＡにより立体構造データベース、例えばＰＤＢから参照タンパク質を選定し、選定された参照タンパク質のアミノ酸配列とのアライメント（並置）が行われる。
ここでいう参照タンパク質とは立体構造データベースから取得される配列と座標のデータベースでありプロテインデータバンク（ＰＤＢ）として登録された公共データベースから得られるものである。ＰＤＢのデータベースには２６２４３個の立体構造が登録されているが（２０００年８月時点）、９５％以上の配列の一致度があるものは同じカテゴリーと判断し、カテゴリー中からもっとも長い配列のものを、更に同じ長さの場合はＸ線の分解能の最も高い構造をそのカテゴリーの代表とした。本発明で用いた立体構造データベースはこの代表を集めたデータベースである。３９２２個の代表構造を本発明においてＰＤＢデータベースとして用いている。
ここで、高ホモロジー用アライメントソフトＦＡＳＴＡの詳細は、前記の通り、″Ｅｆｆｅｃｔｉｖｅ ｐｒｏｔｅｉｎ ｓｅｑｕｅｎｃｅ ｃｏｍｐａｒｉｓｏｎ．″Ｐｅａｒｓｏｎ ＷＲ，（１９９６）Ｍｅｔｈｏｄｓ Ｅｎｚｙｍｏｌ；２６６：２２７−５８に記載されている。このプログラムは、目的配列と一致度の高い配列を立体構造データベースから探索し、最終的な目的配列と参照タンパク質との一致度をｅ値として算出する。
ｅ値は配列間の適合の間に存在するランダムなバックグラウンドノイズを定量的に記述したものである。それは二つの配列がどれだけマッチしているかを示すものでもありスコアに対し指数関数的に減少する性質を持ち、結果の有意な閾値を設定する方法として有用である。このステップにおいては、例えばＦＡＳＴＡで算出されたｅ値が０．０１以下の値を持つとき有意なアライメントであると判断している。
ステップＩＩ−３０：プロファイルアライメントソフトによるデータベース検索及び配列アライメント
ステップＩＩ−１０において入力されたアミノ酸配列を、低ホモロジー領域のアライメントを行うソフトウエア、即ちプロファイルアライメントを行うソフトウエア、例えばＰＳＩ−ＢＬＡＳＴを用いてアライメントを行う。ＰＳＩ−ＢＬＡＳＴはプロファイルアライメントを行うようにプログラムされている。ＰＳＩ−ＢＬＡＳＴの詳細は、前記の通り、″Ｍａｔｃｈｉｎｇ ａ ｐｒｏｔｅｉｎ ｓｅｑｕｅｎｃｅ ａｇａｉｎｓｔ ａ ｃｏｌｌｅｃｔｉｏｎ ｏｆ ＰＳＩ−ＢＬＡＳＴ−ｃｏｎｓｔｒｕｃｔｅｄ ｐｏｓｉｔｉｏｎ−ｓｐｅｃｉｆｉｃ ｓｃｏｒｅ ｍａｔｒｉｃｅｓ．″Ｓｃｈａｆｆｅｒ ＡＡ，Ｗｏｌｆ ＹＩ，Ｐｏｎｔｉｎｇ ＣＰ，Ｋｏｏｎｉｎ ＥＶ，Ａｒａｖｉｎｄ Ｌ ａｎｄ Ａｌｔｓｃｈｕｌ ＳＦ，Ｂｉｏｉｎｆｏｒｍａｔｉｃｓ １９９９，１２，１０００−１１に記載されている。
プロファイルとして用いるアミノ酸配列としては、例えば、公開データベースであるＰＩＲ：ｈｔｔｐ：／／ｗｗｗ−ｎｂｒｆ．ｇｅｏｒｇｅｔｏｗｎ，ｅｄｕ／ｐｉｒ／（Ｎａｔｉｏｎａｌ Ｂｉｏｍｅｄｉｃａｌ Ｒｅｓｅａｒｃｈ Ｆｏｕｎｄａｔｉｏｎ（ＮＢＲＦ））等のアミノ酸配列を用いればよい。
このプログラム（ＰＳＩ−ＢＬＡＳＴ）は、プロファイルのデータベース中の（ここではＰＩＲ中の）有意なアライメント関係のみから情報を引き出し、アミノ酸配列の部位特異的スコア行列を作成する。次にプログラム内部では目的配列の変わりに作成された部位特異的スコア行列と一致度の高い配列をデータベースから探索し、その部位特異的スコア行列は、より有意なアライメントが検出されなくなるまで回数を重ねる毎に逐次更新されてゆく。そして最終的な部位特異的スコア行列と参照タンパク質との一致度をｅ値として算出する。
前記の通り、ｅ値は配列間の適合の間に存在するランダムなバックグラウンドノイズを定量的に記述したものである。それは二つの配列がどれだけマッチしているかを示すものでもありスコアに対し指数関数的に減少する性質を持ち、結果の有意な閾値を設定する方法として有用である。
下記第１表および第２表に、ＣＡＳＰ４（Ｆｏｕｒｔｈ Ｃｏｍｍｕｎｉｔｙ Ｗｉｄｅ Ｅｘｐｅｒｉｍｅｎｔ ｏｎ ｔｈｅ Ｃｒｉｔｉｃａｌ Ａｓｓｅｓｓｍｅｎｔ ｏｆ Ｔｅｃｈｎｉｑｕｅｓ ｆｏｒ Ｐｒｏｔｅｉｎ Ｓｔｒｕｃｔｕｒｅ Ｐｒｅｄｉｃｔｉｏｎ：ｈｔｔｐ：／／ｐｒｅｄｉｃｔｉｏｎｃｅｎｔｅｒ．１１ｎｌ．ｇｏｖ／ｃａｓｐ４／Ｃａｓｐ４．ｈｔｍｌ）において、正しいフォールドでモデリングできたもののリストおよび間違ったフォールドでモデリングしたもののリストとｅ値を示す。正しいフォールドを正しくモデリングできたものが０．１以下である確立は８４％で、間違ってモデリングしたものの内０．１以下である確立は４４％である。ここで、正しくモデリングできたものとしての定義は、タンパク質のフォールドの仕方と二次構造が正しく予測されているものとしている。従ってＲＭＳＤ値が大きくても正しいとしているものもある（Ｔ００８９）。下記第１表および第２表の比較によってｅ値が０．１以下ならかなり正しいモデリングが可能と言える。
タンパク質フォールドを正しくモデリングした第１表においてｅ値が３．２（ＦＡＳＴＡ）と７．１のような大きな数値もあるが、タンパク質フォールドを間違ってモデリングした第２表において０．０３１、０．０４２、０．４８、０．９６のような０．１の近傍のｅ値が存在するので、０．１のｅ値の閾値は妥当と考えられる。またｅ値が５×１０^−１７や７×１０^−２２のような小さい値でも、タンパク質フォールドの評価にエラーを起こすことがあることから、ｅ値がかなり小さいことが正しいフォールドを完全に保証するものではないことに注意する必要がある。なお、第１表および第２表中の「○」又は「◎」は不等式「ｅ＜０．１」又は「Ｉ．Ｄ＞２０」を満たすことを示す。

ステップＩＩ−４０：参照タンパク質間の高ホモロジー用アライメントソフトによる再アライメント及び配列的冗長性排除
ステップＩＩ−２０およびＩＩ−３０において、高ホモロジー用アライメントソフトウエア（ＦＡＳＴＡ）およびプロファイルアライメント用ソフトウエア（ＰＳＩ−ＢＬＡＳＴ）において算出された複数のアライメント間で、高ホモロジー用アライメントソフトウエア（ＦＡＳＴＡ）のアライメントを行い、参照タンパク質同士のアミノ酸配列を総当りで比べ冗長性を排除し、目的配列との間でｅ値の高い方のアライメント及び参照タンパク質を排除している。その際、具体的には、例えばＦＡＳＴＡおよびＰＳＩ−ＢＬＡＳＴが同じＰＤＢのＩＤの参照タンパク質をアライメントの対象としている場合には、そのＰＤＢのＩＤについてＦＡＳＴＡのアライメントを採用し、ＰＳＩ−ＢＬＡＳＴのアライメントを排除する。
このステップにおいてどれだけ冗長性を排除するかが、後のステップにおいて非常に重要となる。本発明では、ＦＡＳＴＡのｅ値がある値以下のもの同士を同じクラスターとして、３９２２個のＰＤＢデータベースがいくつのクラスターに分けられるかを調査した。その結果を第７図に示す。第７図から明らかな通り、ｅ値が０．０１より大きくなるとクラスター数が急激に減少し、構造に多様性が無くなることが分かる。この結果に基づきｅ値が０．０１以下の場合を同じクラスターとして冗長であると判断している。この様に、冗長性の判断において、ｅ値を適切に設定することが重要である。このｅ値を低く設定しすぎれば、似たような配列が多くなりこのステップを行う意味が無くなる。逆に高く設定しすぎれば、クラスターの数が激減し構造に多様性がなくなる。第７図から、冗長性判断のためｅ値を０．０１以下と設定すればこのようなことが起こらず、適切に判断が可能であることがわかる。
更に詳述すれば、例えばプロファイルアライメントを行うＰＳＩ−ＢＬＡＳＴのｅ値０．１以下で立体構造データベースであるＰＤＢに対して検索を行うと、通常の場合、複数のアライメントの候補が得られることが多い。それらに対して全て次のステップＩＩ−５０にあるＣα原子座標の構築を行うことは、大変な時間を要し、且つ同じような立体構造が多数出来ることになる。このようないわば大量に出来る文字配列上「似たようなアライメント」を削除するのがこのステップの目的である。アライメントは目的配列と参照タンパク質の配列との関係を示したものであり、大量に出来たとしても目的配列の部分は同一である。つまりこの場合アライメントが似ているか否かは、参照タンパク質側の配列が似ているか否かで決まる。文字配列上似ているか否かの判定の為に参照タンパク質のアミノ酸配列同士の高ホモロジー用のアライメント例えばＦＡＳＴＡを行う。例えばＦＡＳＴＡでｅ値が０．０１以下であるならば、それは立体構造的に似たようなタンパク質同士の関係があり、それらを参照タンパク質としてモデリングすることは、同じようなタンパク質座標が構築されることを意味する。
そこで参照タンパク質同士は高ホモロジー用アライメントで行い、互いに似ていると判断された場合、どちらを捨てるかの判断は、低ホモロジー用アライメント（プロファイルアライメント）ソフト、例えばＰＳＩ−ＢＬＡＳＴで「目的配列とより似ていない（ｅ値がより高い）」と判断された方を捨てる。このことによりこのステップの目的が達成できる。
ステップＩＩ−５０：Ｃα原子の初期座標の構築・最適化及び立体構造的冗長性排除
ステップＩＩ−４０からの配列的な冗長性を排除したアライメントの結果を受けて、参照タンパク質との挿入および欠損のあるアミノ酸残基についての情報をえる。アライメントにおいて連続して三残基以上のアミノ酸が対応しているギャップの無い領域を選び出し、その領域においては、これらの残基ペアにおいて、目的タンパク質のＣα原子は参照タンパク質と同一のものを当てはめておく。Ｃα原子が求められなかった場合には、予め作成してある断片のデータベースから座標を当てはめる（第３図参照）。
ステップＩＩ−５１：Ｃα原子のシミュレーティッドアニーリング法による構築 上記ステップ５０で作成されたＣα原子はシミュレーティッドアニーリングのプロセスを用いて参照タンパク質の座標から構成される関数を用いて最適化される。この目的関数等は前記ステップＩ−８１（１）の通りであり、同様の方法で行われる。
ステップＩＩ−５２：立体構造的冗長性排除
ステップＩＩ−５１で算出されたＣαの座標を用いて参照タンパク質間の立体構造フィッティングを行う。上記ステップによって複数のアライメントが、候補として選ばれてきた場合、個々に選定された参照タンパク質およびアライメント情報に基づいて目的タンパク質のＣα原子座標を算出し、Ｃα原子座標同士で立体構造をフィッティングを行い、各座標の根平均二乗距離（ＲＭＳＤ）が一定の値以下の場合は同一の立体構造であると判断して、目的タンパク質との間でｅ値の高い方の参照タンパク質およびアライメントを排除して、さらに参照タンパク質およびアライメントが選定される。この場合、根平均二乗距離（ＲＭＳＤ）の値が３．０Å以下の場合は基本的に同一の立体構造であると判断し、目的配列との間でｅ値の高い方の参照タンパク質及びアライメントを排除するのが好ましい。ここで参考資料としてＣＡＳＰ４のターゲットについてモデリングの際にＣα原子座標のみ算出した場合と側鎖まで構築した場合のずれを示すと下記第３表の通りであり、平均は０．４０４±０．１６２Åとなっており、Ｃαのみでおおよその評価が可能である。

このステップにより、更に不必要な参照タンパク質およびアライメントが排除され、立体構造の構築上有用なもののみが選定される。このステップは上記ステップＩＩ−４０から得られる参照タンパク質およびアライメントのみならず、他の如何なる手法で得られたものに対しても適用することができる。
ステップＩＩ−６０：アライメントにアミノ酸配列の挿入部分配列がある場合のＣα原子の構築
得られたアライメントに、挿入部分配列がある場合、予め作成された部分アミノ酸配列の座標データベースから適切な配列を読み込みＣα原子座標が構築される。データベースから得られる長さ以上の挿入部分配列（ｎ）がある場合、次の通りＣα原子座標を構築することができる。
即ち、（ｉ）目的タンパク質の挿入部分配列を上記データベースから得られる長さに削除して参照タンパク質とのアライメントを行い、（ｉｉ）このアライメントを元に削られた目的タンパク質のＣα原子座標を構築し、（ｉｉｉ）削られた目的タンパク質の削除部分の末端から一定数のアミノ酸残基を削除し（但し、削除するアミノ酸残基数は、上記データベースから得られる長さより少ない数とする）、（ｉｖ）この削られた目的タンパク質と目的タンパク質とのアライメントを基に目的タンパク質のＣα原子座標を構築し、（ｖ）挿入部分配列（ｎ）から、削られた目的タンパク質の元の挿入部分配列（ｎ）に相当する部分に残存する削除後の残基数を引いた残基数が、上記データベースから得られる長さ以下となるまで上記操作を繰り返し行う。
上記方法の一例を示す第８図を用いてより具体的にこの方法を説明する。即ち、アライメントに挿入部分配列ｎ残基がある場合、モデリングソフト、例えばＦＡＭＳの場合では最大２４残基の断片部分配列のデータベースを用いて挿入する。モデリングソフトとしてＦＡＭＳを用いた場合ｎ＞２４の場合には、従来ではその挿入部分のモデリングが不可能であったが、第８図の方法によって構築できる。
まず、挿入部分配列ｎ残基の中央部分の配列を除き、挿入部分配列を２４残基に削る。この削られたアライメントを基にＦＡＭＳのＣα原子を構築する機能を用いて、削られた目的配列のＣαの座標を構築する。次に、削られた目的配列の座標を参照とするアライメントを作成する。ただし、削られた目的配列の挿入された２４残基の部分は真の構造ではなく、あくまで中央部分が削られたものであり、中央部分を次の段階でモデリングしなくてはならない。そのまま、中央部分の構築を行うと挿入部分が歪な形になるので、例えば末端部分を７残基程度両末端では１４残基分削る必要がある。
この例では１４残基削っているが、この数が２４残基以上ではいけない。２４残基挿入し１４残基削った後、１０残基分挿入部分がモデルされている。従って、モデリングを要する挿入部分、即ち、挿入部分配列（ｎ）から、削られた目的タンパク質の元の挿入部分配列（ｎ）に相当する部分に残存する削除後の残基数を引いた残基数（１０）は、ｎ−１０残基となる。ｎ−１０＜２５であれば、ＦＡＭＳの機能から立体構造の構築を行う。ｎ−１０＞２４であれば、再び２４残基に削って上記操作を繰り替えす。
上記の操作を行うことにより、何残基の挿入部分配列があっても立体構造の構築が可能である。
このステップＩＩ−６０は、上記ステップから得られる参照タンパク質およびアライメント等を対象とするのみならず、他のいかなる手法で得られたものに対しても適用することができる。
上記で選定されたアライメント、参照タンパク質、Ｃα原子座標等の情報に基づく立体構造の構築方法は特に限定されないが、例えば次のステップ７０〜９０の方法で行うのが好ましい。ここで、本明細書の各ステップにおいて記載されている計算回数、定数、カットオフ値等のパラメーターは、本発明者が最も好ましいと考えているものの一例であり、本発明の範囲を何ら限定するものでない。
ステップＩＩ−７０：主鎖原子座標の構築及び最適化
ステップＩＩ−５２又はステップＩＩ−６０のＣαの原子に主鎖の他の原子を付加し、シミュレーティッドアニーリング法によって目的関数を最小化するようにする。アライメントにモデリングソフトでの処理が不可能な挿入部分配列がない場合はステップＩＩ−５２で構築されたＣα原子に主鎖の他の原子の付加が行われる。
まず、Ｃα原子の立体的な重ねあわせを行い、Ｃαの原子間距離が２．５Å以下の残基が取り上げられる。Ｃαを除く主鎖の原子座標はＣα原子間距離が最小になるように参照タンパク質の座標から取得しモデル構造とする。
参照タンパク質の中に相当する残基が無い場合、主鎖の原子座標はデータベース中の相当する４残基のタンパク質断片から作成される。この過程の中で、残基ｉの主鎖原子はｉ−１番目からｉ＋２番目までのＣα原子間の最小の根平均二乗距離（ＲＭＳＤ）値を持つ残基から選ばれる。その際Ｎ末端の残基では、Ｃα原子座標の重ね合わせ範囲がｉ番目からｉ＋３番目までとなり、Ｃ末端の残基およびそのひとつ前の残基では同様にｉ−３番目からｉ番目までおよびｉ−２番目からｉ＋１番目までとなる。
主鎖原子の目的関数を基にシミュレーティッドアニーリング法によって主鎖原子座標が最適化される。この目的関数等は前記ステップＩ−８２の通りであり、同様の方法で行われる。
ステップＩＩ−８０：側鎖原子座標の構築及び最適化
側鎖の構築は、大きく２段階に分かれており、「構造保存部位の側鎖構築」（スッテプＩＩ−８１）と「全体の側鎖構築」（スッテプＩＩ−８２）に分けられる。
ステップＩＩ−８１：構造保存部位の側鎖構築
算出された主鎖原子に対して、以前の研究における方法を用いてホモロガスなタンパク質から側鎖のねじれ角を得る。この方法の詳細は、″Ｔｈｅ ｒｏｌｅ ｏｆ ｐｌａｙｅｄ ｂｙ ｅｎｖｉｒｏｎｍｅｎｔａｌ ｒｅｓｉｄｕｅｓ ｉｎ ｓｉｄｅ−ｃｈａｉｎ ｔｏｒｓｉｏｎａｌ ａｎｇｌｅｓ ｗｉｔｈｉｎ ｈｏｍｏｌｏｇｏｕｓ ｆａｍｉｌｉｅｓ ｏｆ ｐｒｏｔｅｉｎｓ：Ａ ｎｅｗ ｍｅｔｈｏｄ ｏｆ ｓｉｄｅ ｃｈａｉｎ ｍｏｄｅｌｉｎｇ．″Ｏｇａｔａ Ｋ ａｎｄ Ｕｍｅｙａｍａ Ｈ，Ｐｒｏｔ．Ｓｔｒｕｃｔ．Ｆｕｎｃｔ．Ｇｅｎｅｔ．１９９８，３１，２５５−３６９に記載されている。
このステップＩＩ−８１は前記ステップＩ−８３（１）と同様に行われる。
ステップＩＩ−８２：全体の側鎖の構築
側鎖の構築は固定した主鎖およびＣβ原子のもとで行う。これは上記したＯｇａｔａ Ｋ ａｎｄ Ｕｍｅｙａｍａ Ｈ，Ｐｒｏｔ．Ｓｔｒｕｃｔ．Ｆｕｎｃｔ．Ｇｅｎｅｔ．１９９８，３１，２５５−３６９に開示されている研究成果をもって行わる。
このスッテプＩＩ−８２は前記ステップＩ−８３（２）と同様に行われる。
ステップＩＩ−９０：最終構造の構築
かくしてステップＩＩ−１０〜ＩＩ−８０を行うことにより、目的タンパク質の立体構造を規定する原子座標を得ることができる。前記ステップＩＩ−５２迄の過程でアミノ酸配列的および立体構造的に冗長性を排除したアライメントが得られており、効率的な立体構造の構築が可能である。
前記したアライメント修正法は、上記Ａｄｖａｎｃｅｄ ＦＡＭＳに組み合わせて用いることができる。また、上記各ステップは、コンピュータプログラムにより、全自動で実行することができる。かくして、目的タンパク質の配列が与えられれば全自動のホモロジーモデリング（立体構造の構築）が可能となる。
ＩＩＩ．ドメイン単位に区切った立体構造データベース
次に、本発明の第四の態様である立体構造配列のドメイン区切り方法およびドメイン単位に区切った立体構造データベースについて説明する。
この発明において、「ドメイン」とは、前記の通り、タンパク質の３次構造の単位を意味する。
本発明のドメイン区切り方法は、ドメイン区切りが未知の立体構造の配列について、ドメイン区切りが既知の立体構造の配列に対して配列上の相同性検索を行い、得られたドメイン毎の相同性配列のコンセンサスに基づいて、ドメイン区切りを決定することを特徴とする方法である。
ドメイン区切りが既知の立体構造の配列としては、前記したドイン単位に区切った立体構造データベース（ドメイン構造データベース）、例えば、ＳＣＯＰ、ＣＡＴＨ等の配列が用いられる。相同性の検索は、これらデータベースに対して行い、ドメイン区切りが未知の立体構造の配列と相同性を有する配列を検出する。ここで「相同性」とは、配列の一致度や類似度を意味する。相同性の検索は、適当なアライメントソフト、例えば、前記ＦＡＳＴＡやＰＳＩ−ＢＬＡＳＴ等が用いられる。
これらアライメントソフトを用いて検出されたドメイン毎の相同性配列のコンセンサスに基づいて、ドメイン区切りが未知の立体構造の配列について、ドメイン区切りを決定する。ここで、コンセンサスとは、一致意見または合意を意味する。ドメイン区切りのコンセンサスは、多数決法を採用して行えばよい。
また、ＳＣＯＰに登録されている各ドメインには、その性質を表す識別子（ＩＤ）が付与されている。ドメイン区切りと同時に、この識別子（ＩＤ）についても、コンセンサスに基づいて新たな識別子を付与するのが好ましい。これにより、新たに決定されたドメインについて、その帰属を推定することができる。
上記方法を用いて、ドメイン区切りが未知の立体構造の配列について、ドメイン立体構造データベースを構築することができる。
ＳＣＯＰ等のドメイン構造データベースの更新頻度は、ＰＤＢ等の立体構造データベースとの更新頻度より少ない（更新のスピードにタイムラグがある）。このため、ＰＤＢ等の立体構造データベース中には、ドメイン区切りが決定されていない新規な配列の登録が数多く含まれる。本発明の方法により、これら新規な構造の登録に対しても効率的に対応することができる。
かくして得られるドメイン構造データベースを用いて参照タンパク質を選定することにより、ドメイン単位の立体構造の構築が可能となる。立体構造の構築は、例えば前記した方法により行うことができる。
また、ドメインとは、前記の通り、タンパク質の立体構造で３次構造の単位であり、本発明の方法を用いれば、効率的にドメイン単位の立体構造が得られ、ドメイン間相互作用の研究の良いデータセットの提供も可能となる。
ＩＶ．アミノ酸配列データベースを介した参照タンパク質探索方法
次に、本発明の第五の態様であるアミノ酸配列データベースを介した参照タンパク質の探索方法について説明する。
この発明の、参照タンパク質の探索方法は、
（ｉ）任意の目的タンパク質について、アミノ酸配列データベースに対して相同性検索を行い、目的タンパク質の配列と相同性を有するアミノ酸配列データベースの配列を検出し、
（ｉｉ）検出されたアミノ酸配列について、立体構造データベースの配列に対して相同性検索を行い、検出されたアミノ酸配列データベースの配列と相同性を有する立体構造データベースの配列を検出し、
（ｉｉｉ）検出された立体構造データベースの配列と目的タンパク質の配列とのアライメントを得て、該立体構造デーベースの配列を参照タンパク質とすることを特徴とする方法である。
先ず、相同性検索に用いられるアミノ酸配列データベースとしては、特に限定されないが、例えば、前記したＧＣＲＤｂ、ＧＰＣＲＤ、ＥｘＰＡＳｙ、ＯＲＤＢ、ＰＩＲ、Ｓｗｉｓｓ Ｐｌｏｔ、ＴｒＥＭＢＬ、ＴｒＥＭＢＬＮＥＷ、ＤＡＤ等が挙げられる。これらの中で、使いやすさの観点からＰＩＲを用いるのが好ましい。
相同性の検索は、適当なアライメントソフト、例えば、前記したＦＡＳＴＡやＰＳＩ−ＢＬＡＳＴを用いて行われる。
かくして検出される目的タンパク質の配列と相同性を有する酸配列データベースの配列について、立体構造データベースの配列に対する相同性検索が行われ、アミノ酸配列データベースの配列と相同性を有する立体構造データベースの配列が検出される。用いられる立体構造データベースとしては、特に限定されないが、例えば、前記したＰＤＢ、ＣＣＤＣ、ＳＣＯＰ、ＣＡＴＨ等が挙げられる。これらのデータベースは単独で用いても、組み合わせて用いても良い。相同性の検索は、上記の通り、適当なアライメントソフトを用いて行えば良い。
かくして、アミノ酸配列データベースの配列を介して、目的タンパク質の配列と立体構造データベースの配列とのアライメントを得ることができる。このアライメントに基づき参照タンパク質が選定される。
用いるアミノ酸配列データベースは、似たような配列が検出されるのを防ぐために、あらかじめ、似ているものを排除しておく（冗長性を排除しておく）ことが好ましい。この場合、排除の基準となる配列の一致度（類似度）は特に限定されないず、配列の検出数、処理の効率に基づいて決定すれば良く、例えば、配列の類似度（一致度）が９５％以上のものを省くのが好ましい。
アミノ酸配列データベースから相同性を有する複数（多数）の配列が検出されることがある。この場合、処理の効率化を計るために、これらの配列の冗長性を排除しておくことが好ましい。
冗長性の排除は、高ホモロジーアライメントソフト、例えばＦＡＳＴＡを用いて、閾値を段階的に下げてゆくクラスター解析により行うのが好ましい。具体的には、閾値は初め９０％から５％刻みで下げてゆく。クラスターの代表としては、もともとのＯＲＦ（Ｏｐｅｎ Ｒｅａｄｉｎｇ Ｆｒａｍｅ）のシーケンスに対し最も低いｅ値を持つ配列を採用する。クラスター数が１０以下になるクラスター解析の閾値のところでやめる。但し、この数は自動解析に適応する数であれば１０にこだわらない。特に、大量なアミノ酸配列データベースの配列に対して自動モデリングを実行する為にはこのようにデータベースの配列数を少なくする必要がある。
検出された立体構造データベースの配列と目的タンパク質の配列とのアライメントが複数得られた場合、各アライメント間の一致部分はそのままとし、不一致部分はそれらのコンセンサスに基づいて確定させ、単一のアライメントを得る。即ち、複数のアライメントのコンセンサス（一致意見）に基づいて、単一のアライメントを算出する。アライメントのコンセンサスは、多数決法等を利用して行えばよい。
このアライメントのコンセンサスを取る場合、前記ドメイン区切りや識別子（ＩＤ）の場合と異なり、考慮すべき事項が多く、別のプログラムを用いて行うのが好ましい。この場合、アルゴリズムは、多数の要素の集合から一つを選び出すことに相当し、統計手法として平均値、中央値、最頻値、最大値、最小値等が考えられる。多数決とは、最頻値に相当するが、中央値、平均値の意味も少し含んでいる。特に大小関係が無い場合は、最頻値を選択するのが好ましい（自然である）。
この発明は、目的タンパク質を立体構造データベースに対して相同性検索を行っても、相同性を有する立体構造データベースの配列が検出されなかった場合に行うのが特に好ましい。即ち、この発明により、ＯＲＦの立体構造探索において、通常のアライメントソフト、例えばＰＳＩ−ＢＬＡＳＴ、ＰＲＳ−ＢＬＡＳＴ、ＦＡＳＴＡ等で立体構造の精度が保証された類似のアミノ酸配列（参照タンパク質）を立体構造データベースから探し出せない場合においても、アミノ酸配列データベースを介することにより、参照タンパク質を探し出すことができる。このことは、第３５図に示す通り、モデリングにより立体構造が構築可能なＯＲＦ数が格段に増加したことを意味する。
かくして探索された参照タンパク質およびアライメント情報に基づいて、例えば、前記した方法により立体構造を構築することができる。
Ｖ．本発明の特に好ましい態様
上記した本発明の第一〜第五の態様は、ホモロジーモデリングを行うことができる適当なプログラムが動作するコンピュータを用いて、単独で或いは互いに組み合わせて実施することができる。本発明の最も好ましい態様の一例を第９図に示す。この発明の態様における各ステップの詳細は、前記の通りである。これにより、適当なコンピュータを用いて、如何なるＯＲＦの配列についても、漏らさずに、網羅的に立体構造の構築（モデリング）が可能である。第９図に示す方法は、特に自動化に適した方法である。これにより、目的タンパク質の配列が与えられれば、大量なアミノ酸配列データベースの如何なる配列に対してもコンピュータを用いた自動モデリングを実行することができる。近年ＯＲＦの配列情報は加速度的に増加しており、本発明の方法により、大量のＯＲＦの配列情報について、効率的なモデリングが可能となる。
ＶＩ．タンパク質の立体構造を規定する原子座標が記録されている記録媒体、データベース
上記方法で得られたタンパク質の立体構造を規定する原子座標をコンピュータが利用可能な所定の形式で適当な記録媒体に格納することにより目的タンパク質の立体構造データベースが構築できる。本発明のデータベースは、好ましくは、上記原子座標とともに参照タンパク質と目的タンパク質のアライメント情報を含む。また、データベースには、所望によりコード番号、参照タンパク質の参照領域の情報、目的タンパク質の情報、Ｃα原子間距離等が含まれる。
本発明においてデータベースとは、上記原子座標を適当な記録媒体に書き込み、所定のプログラムに従って検索を行うコンピュータシステムをも意味する。ここで適当な記録媒体としては、例えば、フロッピーディスク、ハードディスク、磁気テープ等の磁気媒体；ＣＤ−ＲＯＭ、ＭＯ、ＣＤ−Ｒ、ＣＤ−ＲＷ等の光ディスク、半導体メモリ等を挙げることができる。
ＶＩＩ．薬物の分子設計方法
薬物分子設計を行うことができる適当なプログラムが動作するコンピュータで、上記方法で得られた薬物分子の標的となるタンパク質（以下これを「標的タンパク質」と称することがある）の構造座標の全て又は一部、又はそれらが記録されたデータベース若しくは記録媒体の構造座標を使用して、標的タンパク質と相互作用をする薬物分子（拮抗薬または作動薬）を同定、検索、評価又は設計等を行うことができる。
薬物分子の同定、検索、評価又は設計は、本発明の方法で得られた立体構造座標と薬物分子の立体構造座標との相互作用の有無やその程度に基づいて行われる。本明細書において、薬物分子の同定、検索、評価又は設計等を、単に薬物の分子設計ということがある。
タンパク質の立体構造座標と薬物候補分子の立体構造座標との相互作に基づいて分子設計を行う際に用いられるコンピュータとしては、適当なプログラムが動作するように調整されているコンピュータであれば特に制限はない。また、コンピュータの記憶媒体にも特に制限はない。分子設計に用いるプログラムは、例えばモレキュラーシミュレーション社製のコンピュータ・プログラムＩｎｓｉｇｈｔ ＩＩ等を挙げることができる。特に、この目的のために特別に作成されたＬｕｄｉやＤＯＣＫといったプログラムを単独又は組み合わせて用いることで、より容易に同定、検索、評価又は設計することができる。また、タンパク質の立体構造座標と薬物分子とのドッキング評価は、例えば、ＮＥＣ社製のＢＩＯＣＥＳ（バージョン３．１０）を用いて行うことができる。
ここで、薬物分子は、既知のものであっても、新たに合成された新規な化学構造を有する薬物分子であっても、その立体構造が得られるものであれば、いずれの薬物分子も本発明の方法で用いることができる。薬物分子の立体構造座標は、Ｘ線結晶解析やモデリング等のいずれの方法で得られたものでも良い。３次元構造座標が決定されているものは、適当なデータベース、例えばＣＣＤＣ（Ｃａｍｂｒｉｄｇｅ Ｃｒｙｓｔａｌｌｏｇｒａｐｈｉｃ Ｄａｔａ Ｃｅｎｔｒｅ：ｈｔｔｐ：／／ｗｗｗ．ｃｃｄｃ．ｃａｍ．ａｃ．ｕｋ／）やＰＤＢ（Ｐｒｏｔｅｉｎ Ｄａｔａ Ｂａｎｋ：ｈｔｔｐ：／／ｗｗｗ．ｒｃｓｂ．ｏｒｇ／ｐｄｂ／）から収得することができる。
更には、特開２０００−１７８２０９号公報に記載されている方法によっても、薬物分子を設計することができる。この様に、本発明の方法で得られたタンパク質の立体構造座標を用いることで、薬物分子のコンピュータによる分子設計が可能となる。ただし、本発明の分子設計方法は、これらのプログラムや手法を用いるものに限定されるものではない。
薬物の分子設計には、通常、概念的に２つの段階がある。最初の段階は、リード化合物を見つけだすものであり、次の段階はリード化合物の最適化である。どちらの段階も、標的タンパク質の立体構造座標を使用して、それ自体既知の方法により行うことができる。これにより最適な医農薬候補分子を得ることができる。
ＶＩＩＩ．分子設計方法により得られる医農薬候補分子のスクリーニング方法
上記方法により同定、検索、評価又は設計された医農薬候補分子は、その分子の性質に応じて、例えばそれ自体既知の化学合成法により得ることができる。しかしながら、薬物分子は、天然化合物、合成化合物のいずれでも良く、また、高分子化合物、低分子化合物のいずれでも良い。得られた医農薬候補分子は、更に、それ自体既知の方法により、試験管内や生体内における薬理学的または生理学的試験によりその活性を調べ、所望の活性を有する医農薬候補分子を選抜することにより実際に医農薬として応用可能なものを得ることができる。
ＩＸ．医農薬組成物の製造方法
上記スクリーニング方法により選択された医農薬、例えば医薬分子は、それ自体単独で治療対象となる疾患等の患者に投与することができるが、これらの有効成分の１種又は２種以上を混合して投与することもできる。また、薬理学的に許容される製剤用添加物等を用いて該物質を医薬品組成物として製剤化し、これを投与するのが好ましい。例えば、必要に応じて糖衣を施した錠剤、カプセル剤、顆粒剤、細粒剤、散剤、丸剤、マイクロカプセル剤、リポソーム製剤、トローチ、舌下剤、液剤、エリキシル剤、乳剤、懸濁剤等として経口的に、あるいは無菌の水性液もしくは油性液として製造した注射剤や、座剤、軟膏、貼付剤等として非経口的に使用できる。これらは、例えば、該物質を生理学的に認められる担体、香味剤、賦形剤、ベヒクル、防腐剤、安定剤、結合剤などとともに一般に認められた製剤実施に要求される単位用量形態で混和し、充填又は打錠等の当業界で周知の方法を用いて製造することができる。これらの医薬組成物における有効成分量は指示された範囲の適当な容量が得られるようにするものである。
農薬分子について、実際に農薬として使用する場合には、担体若しくは希釈剤、添加剤および補助剤等と公知の方法で混合して、通常農薬として用いられている製剤形態（組成物）、例えば粉剤、粒剤、水和剤、乳剤、水溶剤、フロアブル剤等に調製して使用される。
実施例
以下、実施例を挙げて本発明をさらに具体的に説明するが、下記の実施例は、本発明の具体的な認識を得る一助と見なすべきものであり、本発明の範囲を何ら制限するものではない。なお、下記の実施例において、立体構造モデルの構築、アライメント、参照タンパク質の探索等は、ＤＥＬＬ社製パーソナルコンピュータ（Ｄｉｍｅｎｓｉｏｎ ＸＰＳ Ｔ４５０，ＣＰＵ：ＰｅｎｔｉｕｍＩＩＩ ４５０ＭＨｚ，Ｒｅｄ Ｈａｔ Ｌｉｎｕｘ ７．１Ｊ、メモリ：３８４Ｍｂｙｔｅｓ）を用いて行った。
実施例１ アライメント修正法の効果
大腸菌ゲノムの立体構造データベースＰＤＢ（２００１年３月時点）に対するＰＳＩ−ＢＬＡＳＴの検索結果、第１位のホモロジーが１００％（即ちＸ線解析などによる立体構造が存在する）だったものの第２位についてアライメント修正プログラムを実行した結果を示す。第２位のアライメントの修正前および修正後の両方に対してアライメントを基にしてＣαの原子座標のみモデル構築を行い、第１位で構築した正解のモデル構造に対し立体構造フィッティングを行った。
修正時のパラメータとしては次の２つである。
（ｉ）二次構造判定に用いる最大水素結合長（３．０５Å、３．１０Åおよび３．５０Å）。
（ｉｉ）プロファイルの認識方法（ここではＰＳＩ−ＢＬＡＳＴを用いたがＦＡＳＴＡ等でもよい）：プロファイルの情報の連続した領域が何残基続いたときに、アライメントを固定するか（０：プロファイル無視、１：１残基以上、２：２残基以上、３：３残基以上のプロファイル情報が続いている場合にアライメント上のその部分を固定する）。
下記第４表に示すように修正後は、９８個中７４個（７５．５％）ものＯＲＦについてアライメントがＲＭＳＤが良くなるように小さな数値に修正されることが示された。約４分の３が数値上改良されたモデリング精度を示すことは発明者らの提案しているアライメント修正法が有用であることを示す。この表におけるモデリングに使われた参照タンパク質はＯＲＦのＰＳＩ−ＢＬＡＳＴによる通常のＰＤＢデータベース検索においてｅ値が０．００１以下を示すもののみを用いた。

上記第４表中のＯＲＦ名ｃｌｐＰを例に改良された立体構造を示す（第１０図〜第１２図参照）。このタンパク質では、修正前のアライメントでは９．５４２Åなのに対し、二次構造を判定する最大水素結合長を３．１Åとして、二残基以上のプロファイルの情報が連続する場合を固定したアライメント修正後は８．２７９Åに改善された。第１０図〜第１２図から明らかな通り、修正後のモデルは正解構造の二次構造を良く予測している。
上記ｃｌｐＰを例にアライメント修正の詳細について述べる。
修正前のアライメントは、第１３図に示す通りである。
具体的な修正の中身は第１４図〜第１８図に示す通りであり、修正部分は３箇所である。
１箇所目と２箇所目の修正前の修正部分を第１４図に示す。第１４図の５行目はプロファイルの情報を示し、６行目は二次構造情報、７行目は、疎水コア値、および疎水コア距離を示す。第１４図では２箇所挿入がある。そして、それぞれβシート（６行目の文字「Ｅ」に相当）とαヘリックス（同様に「Ｈ」に相当）の中に入っている。本発明のアライメント修正法を適用すると第１５図に示す通り２次構造を分断しないように修正される。
３箇所目の修正前は、第１６図に示す通りであり、ヘリックスを分断していたのに対し、第１７図に示す通りαヘリックスに挿入が入らないようになった。
その結果として、最終的なアライメントは第１８図の通りとなる。
このように修正することにより、ＯＲＦ名ｃｌｐＰについてＲＭＳＤが９．５４２Åから８．２７９Åになる、だけでなく二次構造を保持した構造を算出し、より真実に近い構造を算出することができる。ここでのプログラムのアルゴリズムはすべて自動化が可能なものとなっている。
実施例２ ドメイン単位に区切った立体構造データベースの効果
ここでは立体構造データベースであるＰＤＢをＳＣＯＰにしたがってドメイン単位に区切ったデータベースを利用している。これはドメイン単位にモデルが構築されることを意味する。
大腸菌ゲノムのタンパク質鎖伸張因子でｔｕｆＢと呼ばれるタンパク質を例にこのデータベースを使用したモデリングの利点を示す。このタンパク質は次に示すアミノ酸配列（ＳＥＱ ＩＤ Ｎｏ．１）を持つ。

このタンパク質は、上記実施例１のアライメント修正法の効果の表中にも示したが、公開されているＰＤＢに当てたとき第１９図に示すアライメントを持ちＲＭＳＤが１０．５２７Åであった。配列の相同性が７３．３％程度あるときは、通常において立体構造はもっと似ていてＲＭＳＤはもっと低くあるべきである。
立体構造を調べてみると第２０図および第２１図の様になっており３個のドメインの相対対置が異なるのために全体のＲＭＳＤを下げていることがわかる。
これを、発明者がＳＣＯＰのデータベースを拡張したドメイン区切りしたデータベース中で検索を行うと第２２図〜第２４図に示す三つのアライメントに分けることが出来る。参照された三つの参照タンパク質（ＩＥＸＭ−Ａの三つのドメイン構造タンパク質）がＯＲＦのｔｕｆＢの別々の場所に並置される。
さらにこのアライメントでモデル構築を行い、それぞれ立体構造フィッティングを行うと、正解構造に対してＲＭＳＤはそれぞれ１．５８６Å、１．０１３Å、３．５９９Åとなった。ドメイン区切りでモデルを構築することにより、ドメイン同士の相対配置によって悪く評価されてしまいがちな、ＲＭＳＤを精度良く妥当な数字で評価することが可能となる。同様の傾向はｔｕｆＢのみならず、同じ大腸菌の中のｔｕｆＡおよびｐｙｒＩにも見られる。
実施例３ Ａｄｖａｎｅｅｄ ＦＡＭＳによるデータベースサーチの効果
Ａｄｖａｎｃｅｄ ＦＡＭＳで効果的に参照タンパク質が選ばれることを、２０００年に行われたＣＡＳＰ４のターゲットでもあったタンパク質ＦｔｓＡ（Ｔｈｅｒｍｏｔｏｇａ ｍａｒｉｔｉｍａ由来）、Ｔ００８９について述べる。この立体構造は現在１Ｅ４ＦというＰＤＢコードで公開されているが、この構造含まないデータベースを用いて立体構造予測を行った。
このタンパク質のアミノ酸配列（ＳＥＱ ＩＤ Ｎｏ．２）は、次の通りである。

ＰＳＩ−ＢＬＡＳＴサーチの結果、下記第５表の３９個のＰＤＢが配列的な相同性があるものとして検出された（プロファイルとして２０００年７月時点のＰＩＲを、検索対象として２０００年７月時点のＰＤＢ用いた、さらにＰＤＢデータベースは９５％以上配列相同性のあるものを除いた）。３９個のうちｅ値が０．１以下のものは下記第５表のように７個であった。

この検索結果に対してＡｄｖａｎｃｅｄ ＦＡＭＳの機能である配列的及び立体構造的な冗長性を排除した後、下記第６表の３個のＰＤＢが候補として残った。

下記第７表の様に効果的にターゲットを絞ることが出来る。
正解構造であるＰＤＢ １Ｅ４Ｆに対して、立体構造フィッティングを行い正解構造とのＲＭＳＤを算出した。下記第７表の中でＲＭＳＤ値が小さいほど、正解構造に近いことを示す。

上記の結果をみると、７ＨＳＣのＡ鎖を参照タンパク質としてモデリングした構造がもっとも正解と近い構造であることがわかる。これはＰＳＩ−ＢＬＡＳＴでは７位にヒットしていたものである。Ａｄｖａｎｃｅｄ ＦＡＭＳで３位にヒットしている。Ａｄｖａｎｃｅｄ ＦＡＭＳで最も低いｅ値を持って１位にヒットしている参照タンパク質を使ってモデリングしたものが正解構造に必ずしも最も近いとは保証されないことに注目すべきである。即ち前記ＩＩで詳述したＡｄｖａｎｃｅｄ ＦＡＭＳの手法は正解構造に最も近いモデリング構造を算出する有益な手段ということになる。
この方法を遺伝子配列の集まりであるゲノム等に適用することにより、極めて正解構造に近い構造を網羅的に且つ効率的に算出することが可能となる。
実施例４ アミノ酸配列データベースを介した参照タンパク質探索の効果（その１）
アミノ酸配列データベース（ＰＩＲなどのデータベース）を介した参照タンパク質探索法を用いることにより効果的に参照タンパク質が選ばれることを、２０００年に行われたＣＡＳＰ４のターゲットでもあったタンパク質ＹＢＡＢ（Ｈ．ｉｎｆｌｕｅｎｚａｅ由来）、Ｔ００９１について述べる。２００１年１１月現在、立体構造はまだＰＤＢに登録されていない（ｈｔｔｐ：／／ｐｒｅｄｉｃｔｉｏｎｃｅｎｔｅｒ．１１ｎｌ．ｇｏｖ／ｃａｓｐ４／ｔａｒｇｅｔｓ／ｔｅｍｐｌａｔｅｓ／ｔ００９１．ｄｏｃ．ｈｔｍｌ）。
この配列に対しアミノ酸配列データベース２００１年５月のＰＩＲをプロファイルとして用いたＰＳＩ−ＢＬＡＳＴを用いて２００１年５月時点のＰＤＢ中から類似アミノ酸配列を探索した。但し、ＰＤＢのシーケンスそのものではなく、ＰＤＢのドメイン毎に区分けしたデータベースを使用した。配列中でドメイン区切りの判断は、ＳＣＯＰ（ｈｔｔｐ：／／ｓｃｏｐ．ｍｒｃ−１ｍｂ．ｃａｍ．ａｃ．ｕｋ／ｓｃｏｐ／）によるものである。ＳＣＯＰにまだ登録されていないＰＤＢについては発明者がＦＡＳＴＡを使ってＳＣＯＰに従って分類した。さらにドメイン区分けしたＰＤＢからアミノ酸配列が９５％以上の似ているものは省いている。ＰＳＩ−ＢＬＡＳＴ検索の１位はｅ値が０．２６であった。ｅ値が低いことはアミノ酸配列の相同性が高いことを意味し、０．１以下であるとフォールドが正しい可能性が高いと考えて良いのだが、上記に得られた参照タンパク質からモデリングしたとしてもモデルの精度について信頼に乏しい。
このような状況下でＰＩＲなどのアミノ酸配列データベースを介してのＯＲＦに対する立体構造検索に関する本発明を適用した詳細を以下に示す。
ＰＳＩ−ＢＬＡＳＴを用いて２００１年５月時点のＰＩＲ（アミノ酸配列が９５％以上の似ているものは省いている。１８３０１６シークエンス配列）に対し探索し、ｅ値が０．１以下かつアライメント領域長が３０残基以上のもの（今回は２６配列が該当）のみ取り出した。
次に、検索されてきた２６のＰＩＲの配列を１０以下の数に絞り込むために、ＦＡＳＴＡを用いて閾値を段階的に下げてゆくクラスター解析を行った。大量なＯＲＦに対して自動モデリングを実行する為にはこのようにＰＩＲの配列の数を少なくする必要がある。具体的には、閾値は初め９０％から５％刻みで下げてゆく。クラスターの代表としては、もともとのＯＲＦのシーケンスに対しもっとも低いｅ値を持つ配列を採用した。クラスター数が１０以下になるクラスター解析の閾値のところでやめた。但し、この数は自動解析に適応する数であれば１０にこだわらない。ここでは具体的には４０％のところでクラスター数が７個になったのでクラスタリングをやめ、それぞれのクラスター内から課題の配列（ｔ００９１．ｓｅｑ）に対し最も低いｅ値をもつアミノ酸配列をクラスターの代表にした。それらは以下の通りである。

このクラスターの代表の７配列について、ＰＳＩ−ＢＬＡＳＴを用いて前述のドメイン単位に加工しなおしたＰＤＢデータベースの配列中から類似タンパク質配列を再探索した。その結果、上記７配列中、次の３配列

に対してｅ値が０．１以下の配列が上記のドメイン区切り毎の立体構造データベースから検出された。ＰＤＢのＩＤは３つのいずれのアミノ酸配列に対しても１Ｇ８Ｋ＿Ｇであった。これら３つのアミノ酸配列ついて、ＯＲＦの配列とＰＩＲの配列とのアライメントを第２５図〜第２７図に、ＰＩＲの配列とＰＤＢの配列とのアライメントを第２８図〜第３０図に、それぞれ示す。
ＣＡＳＰ４の課題（ｔ００９１．ｓｅｑ）の配列とアミノ酸配列データベースＰＩＲ中の配列とのアライメント、およびＰＩＲの配列と立体構造の配列とのアライメントから、課題（ｔ００９１．ｓｅｑ）のＯＲＦの配列と立体構造の配列（１Ｇ８Ｋ＿Ｇ）とのアライメントを作成した。
ＰＩＲの配列（Ａ８６４９６）を介したアライメントを第３１図に、ＰＩＲの配列（Ｅ７０１５７）を介したアライメントを第３２図に、ＰＩＲの配列（Ｈ６４２１４）を介したアライメントを第３３図に、それぞれ示す。
上記３つのアライメントから明らかなように、どのＰＩＲを介しているかによってアライメントが若干異なる。さらに若干異なるアライメントに対する調整として、発明者が開発したＣ言語のプログラムである多数決法などの方法を利用して第３４図に示すような最終アライメントを得た。ここで採用された多数決法プログラムはすべて自動モデリングに対応して書かれている。
このように、ＯＲＦの立体構造探索において、ＰＩＲなどのアミノ酸配列を介することにより、通常のＰＳＩ−ＢＬＡＳＴ検索では、立体構造の精度が保証された類似のアミノ酸配列を立体構造データベースから探し出せない場合においても、探し出せることが示された。
このことは第３５図に示すように立体構造構築可能なＯＲＦ数が増加したことになる。この場合ＰＳＩ−ＢＬＡＳＴの替わりにＲＰＳ−ＢＬＡＳＴを利用しても同様のことが言える。
実施例５ アミノ酸配列データベースを介した参照タンパク質探索の効果（その２）
アミノ酸配列データベースを介した参照タンパク質探索法を用いることにより効果的に参照タンパク質が選ばれることを、２０００年に行われたＣＡＳＰ４のターゲットでもあったタンパク質ＰＰＸ１（Ｓｔｒｅｐｔｏｃｏｃｃｕｓ ｍｕｔａｎｓ由来）、Ｔ００８７について述べる。立体構造はＰＤＢに１Ｉ７４として登録されている。
この配列に対しアミノ酸配列データベース２００１年５月のＰＩＲをプロファイルとして用いたＰＳＩ−ＢＬＡＳＴを用いて２００１年５月時点のＰＤＢ中から類似アミノ酸配列を探索した。但し、ＰＤＢの配列そのものではなく、ＰＤＢのドメイン毎に区分けしたデータベースを使用した。さらにドメイン区分けしたＰＤＢからアミノ酸配列が９５％以上の似ているものは省いている。ＰＳＩ−ＢＬＡＳＴ検索の結果０．１以下のｅ値で検出されたＰＤＢはなかった。このような状況下で本発明を適用した詳細を以下に示す。
ＰＳＩ−ＢＬＡＳＴを用いて２００１年５月時点のＰＩＲ（アミノ酸配列が９５％以上の似ているものは省いている。１８３０１６配列）に対し探索し、ｅ値が０．１以下かつアライメント領域長が３０残基以上の８２個のＰＩＲ配列を取り出した。
次に検索されてきた８２のＰＩＲの配列を１０以下の数に絞り込むために、ＦＡＳＴＡを使って段階的クラスター解析を行った。具体的には閾値は初め９０％から５％刻みで下げてゆく。クラスターの代表としては、もともとのシーケンスに対しもっとも低いｅ値を持つ配列を採用した。クラスター数が１０以下になるところでやめるべきであるが、３５％のところでクラスター数が３２個であった。候補が１０以下に絞れていないが、これより閾値を下げることはクラスターの間でホモロジーの低いＰＩＲを候補として残すことになる。そこでこの問題をクリアーするため、この３５％の時点でクラスタリングをやめ、３２個のクラスターの中でｅ値の低いものの順に上位１０個を選択した。もちろん各々のクラスターの代表はｅ値の最も低いアミノ酸配列を選んでいる。このようにＣＡＳＰ４の課題の配列（ｔ００９１．ｓｅｑ）に対しアミノ酸配列データベースから１０個を選択した。
それらを次に示す。

このクラスターの代表の１０配列について、ＰＳＩ−ＢＬＡＳＴを用いて前述のＰＤＢ配列中から類似タンパク質配列を再探索した。その結果、上記１０配列中３配列が０．１以下のｅ値を持つ立体構造ドメインデータベースのＰＤＢ配列を釣ってきた。
それらを次に示す。

これらのアミノ酸配列について、ＯＲＦの配列（Ｔ００８７）とＰＩＲの配列とのアライメントを第３６図〜第３８図、ＰＩＲの配列とＰＤＢの配列とのアライメントを第３９図〜第４１図に、それぞれ示す。、
最終的に、ＣＡＳＰ４の課題（ｔ００８７．ｓｅｑ）の配列とアミノ酸配列データベースＰＩＲ中の配列とのアライメント、ＰＩＲの配列とＰＤＢの立体構造の配列とのアライメントから、第４２図〜第４４図に示す様に課題（ｔ００８７．ｓｅｑ）の配列と立体構造の配列とのアライメントを作成した。
さらにこのアライメントに基づきモデルを構築し、正解構造であるＰＤＢ：１Ｉ７４＿Ａと立体構造フィッティングしてみた。その結果、それぞれＲＭＳＤの値が、１５．８６２Å、１３．６０６Å、１３．１５９Åとなっていた。この結果でもってＣＡＳＰ４に提出していたとしたら、かなりの好成績として認められているだろう。ちなみに１３．１５９ÅはＣＡＳＰ４コンテスト参加グループ７３チーム中１３位に位置している。
また、通常のＰＳＩ−ＢＬＡＳＴで検出することの出来ないターゲットについて自動的に算出したアライメントによって国際コンテストのレベルで正解に迫ることができることを示した。
産業上の利用可能性
従来の方法等と比較し、本発明の優れた点は次の通りである。
（１）ホモロジーモデリングによる立体構造の構築において、アミノ酸配列のアライメント（並置）が与えられた場合において、参照タンパク質の座標から立体構造情報を得てアライメントが修正されるため、より正確な立体構造の構築が可能となる。
（２）アライメントの修正は、タンパク質の二次構造情報、疎水コア値情報、疎水コア距離情報等に基づいて行われ、これにより、タンパク質の立体構造の構成単位である二次構造およびその近傍にアミノ酸配列断片の挿入および欠損が入らない立体構造モデルを得ることができる。
（３）また、アライメントを修正する際に、タンパク質の立体構造を形成する物理的な疎水性相互作用の働く空間およびその近傍にアミノ酸配列断片の挿入および欠損が入らないようにすることができる。
（４）ホモロジーモデリングの手法を用いる際に、２種類の性質の異なるアライメントソフト、例えばＦＡＳＴＡおよびＰＳＩ−ＢＬＡＳＴの出力結果を用いることにより、高ホモロジーから低ホモロジーのいずれの目的タンパク質に対しても、精度の良い立体構造の構築が可能である。
（５）上記（４）のアライメントソフトが複数のアライメントを算出してきた場合に、例えば（ｉ）参照タンパク質の同士のＦＡＳＴＡマッチングを行い、候補となる参照タンパク質のアミノ酸配列の冗長性を排除する、（ｉｉ）アライメントからタンパク質中のＣα原子の座標を作成し、立体構造マッチッグを行い、候補となる参照タンパク質の立体構造的な冗長性を排除する等の明確な基準によりアライメントの選定が可能となった点である。これにより、不必要な参照タンパク質およびアライメントの排除ができ効率的な立体構造の構築が可能となる。
（６）アライメントが、データベースから得られる以上のアミノ酸配列部分断片を含む場合においても、精度良く立体構造の構築が可能である。
（７）ドメイン区切りが未知の立体構造の配列についても、ドメイン区切りが可能となり、立体構造情報の最新の知見に基づいて、３次構造毎の立体構造の構築が可能となった点である。
（８）ＯＲＦの立体構造探索において、通常のアライメントソフトによる検索では、立体構造の精度が保証された類似のアミノ酸配列（参照タンパク質）を立体構造データベースから探し出せない場合においても、アミノ酸配列データベースを介することにより、参照タンパク質が探し出せることになった点である。これにより、立体構造が構築可能なＯＲＦ数が格段に増加する。
（９）かくして、立体構造の構築対象となる任意の目的アミノ酸配列が与えられた場合において、コンピュータプログラムにより全自動でホモロジーモデリング（立体構造の構築）が可能となる。
上記の通り、本発明は、任意のタンパク質のより真に近い立体構造を精度よく効率的に構築することができる方法である。また、かくして得られたタンパク質の立体構造を規定する原子座標を含む本発明のデータベースは、特に医薬、農薬等の分子設計に極めて有用に用いることができる。
本発明を詳細にまた特定の実施態様を参照して説明したが、本発明の精神と範囲を逸脱することなく様々な変更や修正を加えることができることは当業者にとって明らかである。
本出願は、２０００年１２月１日の日本特許出願（特願２０００−３６７００７）、２０００年１２月４日の日本特許出願（特願２０００−３６８４１５）に基づくものであり、その内容はここに参照として取り込まれる。また、本明細書にて引用した文献の内容もここに参照として取り込まれる。
【配列表】

【図面の簡単な説明】
第１図は、本発明のタンパク質の立体構造構築方法（アライメント修正法）の一例を示すフローチャートである。
第２図は、アライメント修正法において、与えられたアライメントを、二次構造、疎水コア値、疎水コア距離等の情報に基づいて修正する方法の一例を示すフローチャートである。
第３図は、ステップＩ−８１のＣα原子座標の構築方法を示す図である。アライメントの一致部分は参照タンパク質から取得し、無い部分はＮ，Ｃ両端それぞれ重なった２残基の重ね合わせの根平均二乗距離（ＲＭＳＤ）が最小のものをデータベースから取得する。
第４図は、ローカルスペースホモロジー（ＬＳＨ）を示す図である。図中のＴ残基に関する計算では、網をかけた（灰色の）残基が考慮される。図中下のアライメントにおける四角で囲った部分が考慮される残基ペアであり、＊のマークがあるところの比率がＬＳＨである。この場合ＬＳＨは５６．２％である。
第５図は、ＬＳＨと構造保存部位（ＳＣＲｓ）にある比率との関係を示す図である。ＬＳＨは目的タンパク質と参照タンパク質とのＣα原子の重ね合わせから計算され、ＳＣＲｓにある比率は目的タンパク質の全残基数に対するＳＣＲｓ中の残基数である。
第６図は、本発明の目的タンパク質の立体構造構築方法（Ａｄｖａｎｃｅｄ ＦＡＭＳ）の一例を示すフローチャートである。
第７図は、ＦＡＳＴＡのｅ値の対数とクラスター数との関係を示す図である。
第８図は、アライメントに予め作成された部分アミノ酸配列の座標データベースから得られる長さ以上の挿入部分配列（ｎ）がある場合のＣα原子座標の構築方法の一例を示す説明図である。
第９図は、本発明の特に好ましい態様の一例の基本的な流れを示す図である。
第１０図は、修正前のアライメントのモデルを示すディスプレイのプリントアウトした写真である。
第１１図は、修正後のアライメントのモデルを示すディスプレイのプリントアウトした写真である。
第１２図は、第１位の正解構造モデルを示すディスプレイのプリントアウトした写真である。
第１３図は、修正前のアライメントを示す図である。
第１４図は、１箇所目と２箇所目の修正の中身を示す図である。
第１５図は、１箇所目と２箇所目の修正後の中身を示す図である。
第１６図は、３箇所目の修正の中身を示す図である。
第１７図は、３箇所目の修正後の中身を示す図である。
第１８図は、最終的なアライメントを示す図である。
第１９図は、ＯＲＦの配列（ｔｕｆＢ）とＰＤＢの配列とのアライメントを示す図である。
第２０図は、ｔｕｆＢの第２位のモデルを示すディスプレイのプリントアウトした写真である。
第２１図は、ｔｕｆＢの第１位のモデルを示すディスプレイのプリントアウトした写真である。
第２２図は、ＯＲＦの配列（ｔｕｆＢ）とドメイン単位に区切ったデータベースの配列とのアライメント（１）を示す図である。
第２３図は、ＯＲＦの配列（ｔｕｆＢ）とドメイン単位に区切ったデータベースの配列とのアライメント（２）を示す図である。
第２４図は、ＯＲＦの配列（ｔｕｆＢ）とドメイン単位に区切ったデータベースの配列とのアライメント（３）を示す図である。
第２５図は、ＯＲＦの配列（Ｔ００９１）とＰＩＲの配列（Ａ８６４９６）とのアライメント（１）を示す図である。
第２６図は、ＯＲＦの配列（Ｔ００９１）とＰＩＲの配列（Ｅ７０１５７）とのアライメント（２）を示す図である。
第２７図は、ＯＲＦの配列（Ｔ００９１）とＰＩＲの配列（Ｈ６４２１４）とのアライメント（３）を示す図である。
第２８図は、ＰＩＲの配列（Ａ８６４９６）とＰＤＢの配列（１Ｇ８Ｋ＿Ｇ）とのアライメント（１）を示す図である。
第２９図は、ＰＩＲの配列（Ｅ７０１５７）とＰＤＢの配列（１Ｇ８Ｋ＿Ｇ）とのアライメント（２）を示す図である。
第３０図は、ＰＩＲの配列（Ｈ６４２１４）とＰＤＢの配列（１Ｇ８Ｋ＿Ｇ）とのアライメント（３）を示す図である。
第３１図は、ＰＩＲの配列（Ａ８６４９６）を介したＯＲＦの配列（Ｔ００９１）立体構造の配列（１Ｇ８Ｋ＿Ｇ）とのアライメントを示す図である。
第３２図は、ＰＩＲの配列（Ｅ７０１５７）を介したＯＲＦの配列（Ｔ００９１）と立体構造の配列（１Ｇ８Ｋ＿Ｇ）とのアライメントを示す図である。
第３３図は、ＰＩＲの配列（Ｈ６４２１４）を介したＯＲＦの配列（Ｔ００９１）と立体構造の配列（１Ｇ８Ｋ＿Ｇ）とのアライメントを示す図である。
第３４図は、ＯＲＦの配列（Ｔ００９１）と立体構造の配列（１Ｇ８Ｋ＿Ｇ）との最終的なアライメントを示す図である。
第３５図は、本発明のアミノ酸配列データベースを介した参照タンパク質探索方法により、立体構造が構築可能なＯＲＦ群が拡大することを模式的に示す図である。
第３６図は、ＯＲＦの配列（Ｔ００８７）とＰＩＲの配列（Ｅ８４０４６）とのアライメントを示す図である。
第３７図は、ＯＲＦの配列（Ｔ００８７）とＰＩＲの配列（Ｇ６４５０３）とのアライメントを示す図である。
第３８図は、ＯＲＦの配列（Ｔ００８７）とＰＩＲの配列（Ｓ７４９４０）とのアライメントを示す図である。
第３９図は、ＰＩＲの配列（Ｅ８４０４６）とＰＤＢの配列（１ＩＢＪ＿Ｃ）とのアライメントを示す図である。
第４０図は、ＰＩＲの配列（Ｇ６４５０３）とＰＤＢの配列（１ＶＰＥ）とのアライメントを示す図である。
第４１図は、ＰＩＲの配列（Ｓ７４９４０）とＰＤＢの配列（１Ｇ６Ｓ＿Ａ）とのアライメントを示す図である。
第４２図は、ＰＩＲの配列（Ｅ８４０４６）を介したＯＲＦの配列（Ｔ００８７）と立体構造の配列（１ＩＢＪ＿Ｃ）とのアライメントを示す図である。
第４３図は、ＰＩＲの配列（Ｇ６４５０３）を介したＯＲＦの配列（Ｔ００８７）と立体構造の配列（１ＶＰＥ＿）とのアライメントを示す図である。
第４４図は、ＰＩＲの配列（Ｓ７４９４０）を介したＯＲＦの配列（Ｔ００８７）と立体構造の配列（１Ｇ６ＳＡ１）とのアライメントを示す図である。Technical field
The present invention relates to a method for preparing a protein three-dimensional structure, and more particularly, to a method for selecting an alignment and reference protein by a specific method and constructing a three-dimensional structure of a target protein based on the information.
The genome has been rapidly determined all over the world, and many amino acid sequences encoded by the genome have also been determined. The method of the present invention can obtain a three-dimensional structure of an amino acid sequence with high accuracy, and is extremely useful in a field (bioinformatics) in which biological information is analyzed focusing on molecular design of medical and agricultural chemicals.
Background art
Using information on a protein with a known three-dimensional structure, it is possible to obtain an alignment with a target protein with an unknown three-dimensional structure, and create a three-dimensional structure of the target protein using a computer based on this alignment information. This method is usually called homology modeling. The accuracy of the three-dimensional structure constructed by homology modeling has improved remarkably in recent years, but there are still many problems to be solved.
In homology modeling, it is first important to obtain a highly reliable alignment in order to obtain a highly accurate three-dimensional structure. As a method for providing alignment by selecting a single or a plurality of highly similar proteins from a three-dimensional structure database such as PDB (Protein Data Bank) or SCOP (Structure Classification of Protein) when an arbitrary sequence is given. For example, there is a method using computer software such as FASTA and PSI-BLAST. FASTA is a program that matches 20 alphabetic character sequences that mean 20 natural amino acids, and PSI-BLAST matches character sequences in the same way, but whether or not the characters match. Instead, it has a property of optimizing alignment by calculating the degree of matching of characters called a profile as a substitution matrix for each site on the character sequence of the related protein, and further performing repeated calculation. By using these, it is said that a certain degree of reliable alignment is supplied.
It is said that a technique for immediately converting to 3D structure information when a certain alignment exists is possible by using modeling software FAMS (Journal of Molecular Graphics and Modeling 18, 258-272, 2000) or the like. . However, when trying to convert into a three-dimensional structure using modeling software as it is in the alignment that is the output result of the above software, the secondary structure such as α helix and β sheet, which is the unit of protein structure formation, is divided, There is a high possibility that it will deviate from the true structure. In addition, there are unstable forms in which not only the secondary structure but also the hydrophobic amino acid residues responsible for the hydrophobic interaction, which is the main cause of the three-dimensional structure of proteins, are exposed to the solvent. Or For this reason, the method considering only the character sequence is a situation in which the three-dimensional structure of the protein to be created is clearly not a true structure.
In homology modeling, as described above, in order to obtain a three-dimensional structure close to a true structure with high accuracy, it is first important to obtain a highly reliable alignment. Given the amino acid sequence of an arbitrary protein, select one or more highly similar proteins from a three-dimensional structure database such as a PDB (Protein Data Bank) or a domain structure database (such as SCOP) that is a tertiary structure constituent unit. As a method for providing alignment, for example, there is a method using computer software such as FASTA and PSI-BLAST. FASTA is a program that matches 20 alphabetic character sequences that mean 20 kinds of natural amino acids, and has high homology (corresponding to about 30% or more of amino acid matching, equivalent to about 0.01 or less in FASTA e value). It is said that a highly reliable model can be constructed by constructing a three-dimensional structure with respect to a reference protein. However, reliability with respect to a low homology reference protein (amino acid identity of about 20% or less) is not guaranteed.
On the other hand, in PSI-BLAST, matching of character sequences is performed in the same manner. However, instead of information on whether characters match, the degree of matching of characters called a profile is determined for each site on the character sequence of related proteins. It has the property of optimizing the alignment by calculating as a replacement matrix and further repeatedly calculating. As a result, even in the low homology region, it is said that if the e-value of PSI-BLAST is 0.1 or more, a certain degree of reliable alignment is supplied.
Thus, FASTA and PSI-BLAST calculate one or a plurality of alignments with different reference proteins for one arbitrary sequence. Both of these alignments consider only the character arrangement and do not include information on the three-dimensional structure. When these softwares or related software have chosen multiple alignments, there is currently no general guideline on which alignments should be constructed according to which alignment.
Further, as described above, it is said that a technique for immediately converting to information of a three-dimensional structure when a certain alignment exists is possible by using, for example, modeling software FAMS. However, the alignment calculated by the above two softwares often includes insertions (insertion subsequences) of amino acid sequence fragments that are longer than the length that can be processed by modeling software. However, modeling was considered impossible.
Disclosure of the invention
In view of the above situation, the present invention has been made for the purpose of providing a method for accurately and efficiently constructing a three-dimensional structure of an arbitrary protein using a computer by homology modeling.
As a result of intensive studies to achieve the above object, the present inventors have performed alignment for creating an appropriate three-dimensional model when given arbitrary alignment and three-dimensional structure coordinates of a reference protein in protein homology modeling. As a result of repeated examinations, if the given alignment is corrected based on the secondary structure information etc. of the reference protein, insertion and deletion of amino acid sequence fragments will not occur in the secondary structure of the protein and its vicinity. It was found that a three-dimensional structure close to the true structure can be constructed with high accuracy.
In addition, in the homology modeling of proteins, the present inventors can construct a three-dimensional structure with high accuracy by using a combination of different alignment software and processing the inserted partial sequence by a specific method. I found.
Further, the present inventors have found that domain separation can be efficiently performed using a known three-dimensional structure database in a specific method for a sequence of a three-dimensional structure whose domain separation is unknown.
Furthermore, the present inventors have found that when an arbitrary amino acid sequence is aligned with a reference protein, the reference protein can be searched for via the arbitrary amino acid sequence.
The present invention has been accomplished based on these findings.
In other words, according to the first aspect of the present invention, (1) an arbitrary target protein is aligned with a reference protein having a known three-dimensional structure, and the three-dimensional structure of the target protein is constructed based on the selected reference protein and alignment information. A method for constructing a protein three-dimensional structure comprising: correcting a given alignment based on three-dimensional structure information obtained from coordinates of a reference protein, and constructing a three-dimensional structure based on the corrected alignment information Is provided.
According to a preferred aspect of the present invention, (2) when the profile alignment method is used for the alignment, the profile information is also taken into consideration and the alignment is further corrected, and (3) the three-dimensional structure information is The method according to (1) or (2) above, which is secondary structure information, hydrophobic core value information and hydrophobic core distance information, and (4) the three-dimensional structure of the reference protein with respect to the Cα atom in the amino acid. To obtain the coordinates, optimize the Cα atom coordinates to minimize the objective function, and add the other atoms of the main chain to the optimized Cα atom coordinates to minimize the objective function. The above (1) to (3) are performed by optimizing the atomic coordinates of the main chain and adding the other atoms of the side chain to the optimized atomic coordinates of the main chain to optimize the objective function. Either The method of mounting is provided.
Hereinafter, the first aspect of the present invention may be referred to as an “alignment correction method”.
According to the second aspect of the present invention, (5) a method of aligning an arbitrary target protein with a reference protein having a known three-dimensional structure and constructing a three-dimensional structure of the target protein based on the selected reference protein and alignment information The method for constructing a protein three-dimensional structure is characterized in that the alignment is performed using two types of software for high homology alignment and profile alignment, and a reference protein and alignment are selected based on the results. The
According to a preferred aspect of the present invention, (6) the reference protein and alignment are selected by comparing the sequences by performing realignment using a high homology software between a plurality of alignments calculated by two types of software. The method according to the above (5), which is carried out by excluding the reference protein having a higher e value from the target protein and the alignment when the homology between them is high, and (7) software for high homology alignment Is the FASTA, and the software for profile alignment is PSI-BLAST. (8) The method according to (5) or (6), and (8) a plurality of alignments have been selected as candidates by the software. Based on the selected reference protein and alignment information The Cα atom coordinates of the sample are calculated, and the three-dimensional structure is fitted between the Cα atom coordinates. If the root mean square distance (RMSD) of each coordinate is a certain value or less, it is determined that the three-dimensional structure is the same. The method according to any one of (5) to (7) above, wherein the reference protein and the alignment having a higher e value with the target protein are excluded, and the reference protein and the alignment are further selected, (9) The method according to (8) above, in which the root mean square distance (RMSD) is 3.0 mm, and (10) the creation of the three-dimensional structure was selected for the Cα atom in the amino acid based on the selected alignment information. Obtain coordinates from the three-dimensional structure of the reference protein, optimize the atomic coordinates of Cα to minimize the objective function, and add other atoms of the main chain to the optimized atomic coordinates of Cα. The above is done by optimizing the atomic coordinates of the main chain to minimize and adding the other atoms of the side chain to the optimized atomic coordinates of the main chain to optimize the objective function (5) The method in any one of (9) is provided.
From the third aspect of the present invention, (11) a method of aligning an arbitrary target protein with a reference protein having a known three-dimensional structure, and constructing the three-dimensional structure of the target protein based on the selected reference protein and alignment information When there is an insertion partial sequence (n) longer than the length obtained from the coordinate database of partial amino acid sequences prepared in advance in the alignment, (i) the insertion partial sequence of the target protein is deleted to the length obtained from the database. And (ii) construct Cα atom coordinates of the target protein that has been deleted based on this alignment, and (iii) a certain number of amino acid residues from the end of the deleted portion of the target protein that has been deleted. The number of amino acid residues to be deleted is the length obtained from the above database. (Iv) constructing the Cα atom coordinates of the target protein based on the alignment between the target protein and the target protein, and (v) the target protein deleted from the insertion partial sequence (n). The above operation is repeated until the number of residues obtained by subtracting the number of residues after deletion remaining in the portion corresponding to the original insertion subsequence (n) of the above is less than the length obtained from the database, and the Cα atomic coordinates are determined. A method for constructing a three-dimensional structure of a protein is provided.
According to a preferred aspect of the present invention, (12) the creation of the three-dimensional structure obtains coordinates from the three-dimensional structure of the reference protein selected for the constructed Cα atom coordinates, and minimizes the objective function so as to minimize the objective function. Optimize the coordinates, optimize the main chain atomic coordinates to minimize the objective function by adding other atoms of the main chain to the optimized Cα atomic coordinates, and optimize the main chain atomic coordinates The method according to the above (11), which is performed by adding another atom of the side chain to the structure to optimize the objective function, is provided.
Hereinafter, the second and third aspects of the present invention may be collectively referred to as “Advanced FAMS”.
According to another aspect of the invention, there is provided (13) the method according to any one of (1) to (12) above, wherein a reference protein having a known three-dimensional structure is selected from a three-dimensional structure database divided into domain units. .
According to the fourth aspect of the present invention, (14) for a sequence of a three-dimensional structure whose domain delimiter is unknown, a sequence homology search is performed on the sequence of a three-dimensional structure whose domain delimiter is known, and for each domain obtained Provided is a method for domain delimitation of a protein characterized by determining domain delimitation based on consensus of homologous sequences.
According to a preferred aspect of the present invention, (15) when the obtained homologous sequence for each domain has an identifier, the identifier of the determined domain is further determined based on the consensus of the identifier (14) Is provided.
According to another aspect of the present invention, (16) a three-dimensional structure database divided into domain units constructed by the method described in (14) or (15) above is provided.
According to another aspect of the present invention, there is provided the method according to (13), wherein (17) the three-dimensional structure database divided into domain units is the three-dimensional structure database according to (16).
According to the fifth aspect of the present invention, (18) (i) a homology search is performed on the amino acid sequence database for any target protein, and the sequence of the amino acid sequence database having homology with the sequence of the target protein is detected, (Ii) A homology search is performed on the detected amino acid sequence with respect to the sequence of the three-dimensional structure database, and a three-dimensional structure database sequence having homology with the detected amino acid sequence database sequence is detected, and (iii) detection Provided is a method for searching for a reference protein through an amino acid sequence database, wherein an alignment between the sequence of the constructed three-dimensional structure database and the sequence of the target protein is obtained, and the three-dimensional structure database sequence is used as a reference protein. .
According to a preferred aspect of the present invention, (19) the homology search is performed using alignment software, and (20) a plurality of amino acid sequence database sequences having homology with the target protein sequence. If detected, the method according to (18) or (19) above, wherein the sequence redundancy is eliminated using high homology alignment software, and (21) the criterion for making the sequence redundancy elimination variable in stages. (22) The method according to (20) or (21), wherein (22) the sequence redundancy is eliminated until the detected amino acid sequence database has 10 sequences; (23) When multiple alignments of the sequence of the detected three-dimensional structure database and the sequence of the target protein are obtained, the matching part between the alignments remains as it is. The method according to any one of (18) to (22) above, wherein the inconsistent portion is determined based on the consensus, and (24) the target protein is homologous to the three-dimensional structure database. The method according to any one of the above (18) to (23), which is carried out when a sex search is performed and a sequence of a three-dimensional structure database having homology with the target protein is not detected, is provided.
According to another aspect of the present invention, (25) any one of (1) to (13) and (17), wherein the selection of the reference protein is performed by the method according to any one of (18) to (24) above. Is provided.
According to a sixth aspect of the present invention, there are provided (26) atomic coordinates that define a three-dimensional structure of a protein obtained by the method according to any one of (1) to (13), (17) and (25). The
According to a seventh aspect of the present invention, there is provided (27) a computer-readable recording medium in which the atomic coordinates described in the above (26) are recorded.
According to an eighth aspect of the present invention, there is provided (28) a database comprising the atomic coordinates described in (26) above.
According to a ninth aspect of the present invention, (29) a three-dimensional protein produced using the atomic coordinates described in (26) above or the atomic coordinates of the recording medium or database described in (27) or (28) A drug molecule design method is provided that identifies, searches, evaluates, or designs a target drug molecule based on the interaction between the structure and the three-dimensional structure of the drug candidate molecule.
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, the present invention will be described in more detail. Several terms are used herein and have the following meanings unless otherwise indicated.
The “target protein” means a protein for which a three-dimensional structure is to be constructed in the present invention, since a complete three-dimensional structure has not been determined by X-ray crystallography or NMR analysis. The amino acid sequence of this protein is sometimes referred to as “target sequence” or “target amino acid sequence”. In this target protein, the partial structure has been analyzed but a complete three-dimensional structure has not been obtained, the function has already been specified, the function has been estimated, the amino acid sequence has been determined However, the functions are completely unknown. “Reference protein” means a protein that has already been determined in detail by X-ray crystallography or NMR analysis, and is referred to for alignment and optimization of atomic coordinates. “Alignment” means establishing a correspondence between amino acid sequences of two or more proteins, and the method will be described in detail in the description of each step below.
“Atomic coordinates” describe a three-dimensional structure in a three-dimensional space. It is a relative distance in three directions perpendicular to each other with a point in space as the origin, and is a vector quantity consisting of three numbers per atom excluding hydrogen atoms present in the protein.
“Secondary structure” means α helix, β sheet, or other secondary structure, which is a structural unit of protein tertiary structure. A region where the secondary structure is not discriminated is called a loop. “Hydrophobic core” is a hydrophobic interaction composed of side chains of hydrophobic amino acid residues such as Ala, Cys, Gln, Ile, Leu, Met, Phe, Pro, Thr, Trp, Tyr, and Val. The “hydrophobic core value” indicates the number of hydrophobic residues around the amino acid. The “hydrophobic core distance” indicates the distance between the hydrophobic residue and the hydrophobic core, and the larger the distance, the farther the residue is from the hydrophobic core. For detailed definitions of hydrophobic core, hydrophobic core value, hydrophobic core distance, etc., see Umezawa Y and Umeyama H, Chem. Pharm. Bull. (1988) Vol 36, 4652-4658, Kaneko H et al. , Research Communication in Biochemistry and Cell & Molecular Biology (1998) Vol 2, 37-54.
A “profile” is a character in an amino acid sequence to which information characterizing the position of the character is added, which means a matrix. This matrix is a brute force combination of amino acids and has 20 * 20 elements. This matrix also has the meaning of information on the degree of conservation of amino acids, ease of substitution, and similarity between amino acids.
I. Construction of protein structure by alignment correction method
First, the alignment correction method according to the first aspect of the present invention will be described. FIG. 1 is a flowchart showing an example of construction of a three-dimensional structure by the alignment correction method of the present invention.
As shown in FIG. 1, in this method, first, in step I-10, an alignment showing the relationship between the amino acid sequence of the target protein and the amino acid sequence of the reference protein is prepared. In step I-20, the coordinates of the reference protein are prepared from an appropriate three-dimensional structure database. In step I-30, the secondary structure is determined from the reference protein coordinates prepared in step I-20. In Step I-40, the hydrophobic core value and the hydrophobic distance are calculated from the reference protein coordinates prepared in Step I-20. In step I-50, it is checked whether the region of alignment includes the region of the hydrophobic core in the reference protein. In Step I-60, the secondary structure information, the hydrophobic core value information, and the hydrophobic core distance information are juxtaposed along the alignment of the reference protein on the alignment. In step I-70, alignment is corrected. A three-dimensional structure is constructed using modeling software such as FAMS based on the alignment obtained in Step I-70, and a model is completed in Step I-90. In step I-100, it is checked whether the hydrophobic core of the finished model is formed, and the final structure of step I-110 is reached.
Hereinafter, each step will be described in more detail.
Step I-10: Amino acid sequence alignment
First, in Step I-10, an alignment showing the relationship between the amino acid sequence of the target protein and the amino acid sequence of the reference protein is prepared.
The amino acid sequence of the target protein to be used may be any sequence derived from a database registered, a sequence analyzed for the first time, or the like. In addition, a protein sequence for which only a partial structure has been analyzed can also be used as a three-dimensional structure construction target in the present invention in order to obtain complete three-dimensional structure information.
As a database to be used, for example, “An Internet review: the complete neuroscientists the world Wide Web.” Bloom FE, Science 1996; 274 (5290): 1104-9 -Coupled Receptor Database): http: // www. gcrdb. usthscsa. edu /, GPCRDB: http: // www. gpcr. org / 7tm /, ExPASy: http: // www. expasy. ch / cgi-bin / sm-gpcr. p1, ORDB: http: // ycmi. med. yale. edu / senselab / ordb /, GeneBank: ftp: // ncbi. nlm. nih. gov / genbank / genomes /, PIR: http: // www-nbrf. georgetown. edu / pir / (National Biomedical Research Foundation (NBRF)), Swiss Plot: http: // www. expasy. ch / sprot / sprot-top. html (Swiss Institute of Bioinformatics (SIB), European Bioinformatics Institute (EBI)), TrEMBL (URL and administrator are both the same as Swiss Plot), TrEMBLNEW (SURL and administrator are both the same as SwissPlot) ftp. ddbj. nig. ac. Humans registered in databases such as jp (Japanese DNA Data Punk)H.sapiens), Drosophila (D.melanogaster),Nematode(C.elagans),yeast(S.cerevisiae), Arabidopsis (A.thalianaAnd the like. These databases are merely examples, and any database can be used as long as the amino acid sequence of the protein is registered.
As a three-dimensional structure database used for selection of a reference protein, for example, PDB (Protein Data Bank): http: // www. rcsb. org / pdb /, CCDC (Cambridge Crystallographic Data Center: http://www.ccdc.cam.au.uk/, SCOP (Structure Classification of Protein.r.cco.mtp: /. / Scop, CATH: http: //www.biochem.ucl.ac.uk/bsm/cath, etc. These three-dimensional structure databases can be used alone or in combination. SCOP and CATH are three-dimensional structure databases (hereinafter, this may be abbreviated as “domain structure database”) divided into domain units. The term “main” refers to a tertiary structure unit of a protein's three-dimensional structure. SCOP is constructed by manually discriminating a tertiary structure unit and determining a tertiary structure break. CATH is constructed by determining the unit of tertiary structure almost fully automatically, and it is preferable to use a combination of the three-dimensional structure database constructed by the method of the present invention described later as the domain structure database. .
For example, FASTA or PSI-BLAST (Position-Specific Iterated BLAST) is preferably used as the alignment software. FASTA is a program that searches sequences having a high degree of coincidence with a target sequence from a three-dimensional structure database and calculates the degree of coincidence between the final target sequence and a reference protein as an e value. Details of FASTA are described in “Effective protein sequence comparison.” Pearson WR, (1996) Methods Enzymol; 266: 227-58.
PSI-BLAST is programmed to perform profile alignment. The details of PSI-BLAST are “Matching a protein sequence against a collection of, PSI-BLAST-constructed position and specific score-specific I-in-V 1000-11.
PSI-BLAST, which performs profile alignment, is the tool with the best performance at present for detecting sequence similarity. This program extracts information from only significant alignment relationships in the profile database and creates a site-specific score matrix of amino acid sequences. Next, the program searches the database for sequences with a high degree of coincidence with the site-specific score matrix created instead of the target protein sequence, and the site-specific score matrix counts the number of times until no more significant alignment is detected. Each time it is overlaid, it is updated sequentially. Then, the degree of coincidence between the final site-specific score matrix and the reference protein is calculated as an e value (E value). In addition, as shown below, the same effect can be obtained by using RPS-BLAST (Reverse PSI-BLAST) instead of PSI-BLAST.
In RPS-BLAST, the profile of each reference protein is created with respect to a certain large sequence database in which only amino acid sequences are collected in the above process, and the target protein has a small e value (E Check if you have (value). RPS-BLAST is one of the execution forms of PSI-BLAST. Details of RPS-BLAST can be found in Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ. Nucleic Acids Res 1997 Sep 1; 25 (17): 3389-402 "Gapped BLAST and PSI-BLAST: a new generation of protein database programs."
The e value (E value) is a quantitative description of random background noise that exists during matching between sequences. It indicates how well the two sequences match, has a property that decreases exponentially with respect to the score, and is useful as a method for setting a significant threshold for the result. In the present invention, when the alignment has a reliability of usually 98% or more, preferably 99% or more, a three-dimensional structure is created assuming that the target protein is a protein family having a certain function such as GPCRs. It is appropriate to judge. In PSI-BLAST, the case where the e value (E value) has a statistical value of usually 0.1 or less, preferably 0.01 or less, corresponds to that.
Step I-20: Preparation of reference protein
In step I-20, the coordinates of the reference protein are prepared from an appropriate three-dimensional structure database such as PDB. At this stage, a PDB coordinate file of the reference protein is prepared. More specifically, both the coordinate file obtained by extracting only the alignment region portion judged to be high homology by the alignment software and not the entire protein region and the entire protein region are prepared.
Step I-30: Discrimination of secondary structure information
In step I-30, the secondary structure is determined from the reference protein coordinates prepared in step I-20. The secondary structure determination is applied to both the coordinates obtained by extracting only the alignment region portion prepared in Step I-20 and the entire protein. As specific software, DSSP, STRIDE, and the like are preferable, but basically, a method of discriminating from a twist angle of a protein main chain and a hydrogen bond pattern is used. Here, for example, it is preferable to use a seq23d program written in C language developed by the inventors. However, the distance between oxygen and nitrogen across the hydrogen of hydrogen bond ranges from 2.9 to 3.5 mm, so here the distance is 3.05 mm, 3.10 mm, and 3.50 mm. Adopted.
Here, “DSSP (Dictionary of secondary structures of proteins)” is a PDB format file that is used to analyze the hydrogen bond pattern of the main chain, internal rotation angle, etc., and determine the α helix and β sheet. Software. Details of this can be found in Kabsch, W. et al. & Sander, C.I. (1983) Dictionary of protein structure: pattern recognition of hydrogen-bonded and geometric features. Biopolymers, 22: 2577-2638. It is described in. “STRIDE (Protein secondary structure assignment from atomic coordinations)” is a software that uses a PDB format file as an input file to analyze the hydrogen bond pattern of the main chain, internal rotation angle, etc., and determine the secondary structure. is there. Details of this are described in Frisman, D & Argos, P .; (1995) Knowledge-based secondary structure assignment. Proteins: structure, function and genetics, 23, 566-579. It is described in.
Step I-40: Calculation of hydrophobic core value and hydrophobic core distance
In Step I-40, the hydrophobic core value and the hydrophobic core distance are calculated from the reference protein coordinates prepared in Step I-20. This step is also applied to both the coordinates obtained by extracting only the alignment region portion prepared in Step I-20 and the entire protein region, as in Step I-30. As specific software, for example, hpocore and hpodist of BIOCESS (manufactured by NEC) are preferable.
Here, “hpocore” is software that takes as input a file in which atomic coordinates are in a unique format of BIOCESS, and writes hydrophobic core value information into the file. Determine the space of hydrophobic interaction formed by the side chains of hydrophobic residues, and the number of side chains of each hydrophobic residue that are in spatial contact with the number of hydrophobic residues (hydrophobic (Core value: 0 or positive integer). "Hpodist" is software that writes the information of the hydrophobic core distance to a file in which the atomic coordinates are in a BIOSES original format and the file in which the hydrophobic core value calculated by hpocore is already written is input. , Software that calculates the quantitative distance (hydrophobic core distance) of how far each residue is from the space of hydrophobic interaction from the hydrophobic core value calculated by “hpocore” and atomic coordinates . However, here, the inventor rewrites and uses the source in C language so that the program can be automated.
Step I-50: Confirmation of alignment region in reference protein
In step I-50, it is checked whether the region of alignment includes the region of the hydrophobic core in the reference protein. Here, when the hydrophobic core region is not included, it is determined that the protein cannot be folded using the hydrophobic interaction in this alignment, and the following operation is not performed.
Step I-60: Alignment of secondary structure, hydrophobic core value, and hydrophobic core distance information
In Step I-60, the secondary structure information, the hydrophobic core value information, and the hydrophobic core distance information are juxtaposed along the alignment of the reference protein on the alignment.
Step I-70: Correction of alignment
In step I-70, alignment is corrected. The alignment is corrected in a state where information such as secondary structure, hydrophobic core value, and hydrophobic core distance is added to the alignment information. In addition, when the location of the active residue of the reference protein is known, it is preferable to include that information. In addition, profile information is also obtained at the time of alignment execution by PSI-BLAST or FASTA, and those places where the alignment is not moved will not lose useful information of the original program. Therefore, it is preferable to include profile information as shown in the effect of the alignment correction method of Example 1 described later.
FIG. 2 is a flowchart showing an example of an alignment correction method. Hereinafter, the alignment correction method will be described in detail with reference to FIG.
1) When the insertion / deletion position is in the α helix or β sheet
Even if the insertion / deletion position is in the α helix or β sheet, the boundary between the two α helix regions continues, the boundary between the two β sheet regions, or the α helix region and the β sheet region. If it exists at the following boundary, it ends without being corrected.
1-1) In the case of α helix
If the deletion is 3 residues from the end of the α helix, the process ends without change. Only when this is not the case, the hydrophobic core values in the regions other than the α-helix and β-sheets on the N-terminal side and C-terminal side are compared with the insertion / deletion. When a large difference is recognized in the hydrophobic core value, the insertion / deletion is moved to a region other than the α helix / β sheet having the smaller hydrophobic core value and having the maximum hydrophobic core distance. If a large difference is not recognized, the insertion / deletion is moved to a region other than the nearest α helix / β sheet and having the maximum hydrophobic core distance.
(I) Further, in step I-10, when preparing the alignment, it is determined by alignment software, for example, PSI-BLAST, whether or not the position of insertion / deletion corresponds to the region designated as having high homology.
(Ii) If information on the active site of the reference protein is obtained, it is determined whether or not the position of insertion / deletion is a corresponding part.
In the case of (i) and (ii) above, the region moves to the maximum position of the hydrophobic core distance while avoiding the corresponding region. After the movement, each item of (i) and (ii) is determined, and the movement is made to avoid the corresponding area and to the position where the hydrophobic core distance is the maximum. This operation is repeated until the items (i) and (ii) become negative.
1-2) In the case of β sheet
The hydrophobic core values in the regions other than the α-helix and β-sheets on the N-terminal side and C-terminal side are compared with the insertion / deletion. When a large difference is recognized in the hydrophobic core value, the insertion / deletion is moved to a region other than the α helix / β sheet having the smaller hydrophobic core value and having the maximum hydrophobic core distance. If a large difference is not recognized, the insertion / deletion is moved to a region other than the nearest α helix / β sheet and having the maximum hydrophobic core distance.
(I) Further, in step I-10, when preparing the alignment, it is determined by alignment software, for example, PSI-BLAST, whether or not the position of insertion / deletion corresponds to the region designated as having high homology.
(Ii) If information on the active site of the reference protein is obtained, it is determined whether or not the position of insertion / deletion is a corresponding part.
In the case of (i) and (ii) above, the region moves to the maximum position of the hydrophobic core distance while avoiding the corresponding region. After the movement, each item of (i) and (ii) is determined, and the movement is made to avoid the corresponding area and to the position where the hydrophobic core distance is the maximum. This operation is repeated until the items (i) and (ii) become negative.
2) When there are insertions / deletions in other secondary structures
If there is a nearby loop, the insertion / deletion is moved to the center of the loop. Further, it moves from the moved position to the maximum position of the hydrophobic core distance.
3) When there is an insertion / deletion in the loop
In step I-10, whether alignment is included in a region designated as having high homology by alignment software such as PSI-BLAST when preparing an alignment, and whether it corresponds to the location of an active residue in a reference protein It moves to the maximum position of the hydrophobic core distance so that both are rejected.
Step I-80: Modeling
Based on the corrected alignment obtained in Step I-70, a three-dimensional structure is constructed using an appropriate modeling software, for example, FAMS. The model is completed in Step I-90, and the completed model is confirmed in Step I-100. It is examined whether a hydrophobic core is formed, and the final structure of Step I-110 is reached.
Next, each step of the FAMS will be described as a suitable example of modeling (construction of a three-dimensional structure). The number of calculations, constants, cut-off values, etc. in the following Steps I-81 to I-83 show examples of preferable parameters considered by the inventor, and limit the scope of the present invention. Not.
Step I-81: Construction and optimization of initial coordinates of Cα atoms
In response to the alignment results from step I-80, information about amino acid residues inserted and deleted from the reference protein is obtained. In the alignment, a region having no gap corresponding to three or more amino acids in succession is selected, and in this region, the Cα atom of the target protein is applied to the same as the reference protein in these residue pairs. deep. If the Cα atom is not obtained, coordinates are applied from a database of fragments prepared in advance (see FIG. 3). Here, Cα atom in the present specification means a carbon atom that is the center of the skeleton of each amino acid. The Cβ atom means a carbon atom bonded to the side chain side of the Cα atom. Moreover, C atom means the carbon atom of a carbonyl group.
Step I-81 (1): Construction of the Cα atom by the simulated annealing method
The Cα atom created in step I-81 is optimized using a function constructed from the coordinates of the reference protein using a simulated annealing process. This objective function is as shown in the following formula (1).

Here, Ulen relates to the distance between the Cα atoms of the pair of the residue adjacent to the sequence and the Cys residue, and is set as the following formula (2).

Here, Di, i + 1 is the distance between Cα of residue i and residue i + 1. DiSS is the distance between pairs of Cys residues that form disulfide bonds. K1 and KSS are constants and are set to 2 and 5, respectively.
Uang is a function of the bond angle of the Cα atom and is represented by the following formula (3).

Here, θi (rad) is the angle of the i, i + 1, i + 2nd residue Cα atom. θ0 is set to (100/180) · π (rad) from the X-ray structure of the PDB. Ka is a constant and is 1.
Upos is a function related to the position of the Cα atom, and is represented by the following formula (4).

Here, “‖” means “norm”, and “Mi” means an average distance between Cα atoms at structurally equivalent positions on the alignment based on the structure. When the value of Mi is not found for residue i, the value of Mi is set to 10. Here, it is an average coordinate of the Cα atom, as shown in the following formula (5).

Here, Xji is the atomic coordinates of Cα of the i-th residue of the j-th reference protein. wji is the weight of the i-th residue of the j-th reference protein. This weight is an important parameter for determining the approximate shape of the target protein, and this is determined by a local value within 12 km of the region of interest called local space homology (LSH) ( (See FIG. 4). The correlation between the LSH and the ratio of residue pairs present in structurally conserved regions (SCRs) is very high as shown in FIG. This means that when having a high LSH value, the position of the Cα atom is statistically within 1.0 cm compared to the reference protein structure. Uvdw is as shown in the following formula (6).

Here, Kvdw was set to 0.01 (Di, j <3.2 Å) and 0.001 (Di, j> 3.2 Å), and 6 と し た was set as a cutoff value.
Cα atoms are optimized using a simulated annealing method according to equation (1). At this stage of optimization, the Cα atom perturbation is set to be within 1.0 cm. The annealing stage is calculated 100 times for all Cα atoms. The parameter corresponding to the temperature is decreased by 0.01 every 25 to 0.5 times, and the parameter is assumed to be constant thereafter.
The two large steps, acquisition of structure information and construction of Cα atoms are repeated 10 times, and the coordinates of Cα atoms having the minimum objective function value are calculated as the optimum solution.
Step I-82: Construction and optimization of main chain atomic coordinates
Another atom of the main chain is added to the atomic coordinates of Cα in step I-81 (1), and the objective function is minimized by the simulated annealing method. First, three-dimensional superposition of Cα atoms is performed, and residues having a Cα interatomic distance of 2.5 cm or less are picked up. The atomic coordinates of the main chain excluding Cα are obtained from the coordinates of the reference protein so as to minimize the distance between the Cα atoms, and are used as a model structure.
If there is no corresponding residue in the reference protein, the atomic coordinates of the main chain are created from the corresponding 4-residue protein fragment in the database. In this process, the main chain atom of the residue i is selected from the residues having the smallest rmsd value between the Cα atoms from the (i−1) th to the (i + 2) th. At that time, in the N-terminal residue, the overlapping range of the Cα atom coordinates is from i-th to i + 3-th, and in the C-terminal residue and the residue immediately before it, the i-3-th to i-th and i-th are also the same. -2 to i + 1.
The main chain atom coordinates are optimized by the simulated annealing method based on the objective function of the main chain atom.
The objective function is as shown in the following formula (7).

It becomes.
Ubond is as shown in the following formula (8).

Here, bi0 is a standard bond length and differs depending on the type of chemical bond. Kb is a constant and is 225.
Uang is a function of the bond angle, as shown in the following formula (9).

Here, θi is the i-th bond angle and differs depending on the type of chemical bond. Ka is a constant and is set to 45.
Unon-bond is a function of non-bonded interaction and is as shown in the following formula (10).

Here, εi, j and ri, j * are constants and differ depending on the type of atom.
Knon is a constant of 0.25, and the cutoff is 8 mm.
USS is a function of a disulfide bond generated by a Cys residue and is represented by the following formula (11).

Here, KSSCα and KSSCβ are constants and 7.5.
Upos is a function related to the position of the atom, and is represented by the following formula (12).

Here, <WiXi> is given by the following equation (13).

  <WiXi> in formula (12) is obtained from superposition of the structure between the target protein and the reference protein.
Kpos is a constant and is 0.3.
Utor is a torsion angle of the main chain and is represented by the following formula (14).

Here, φi0 and φi0 are the closest twist angles θi and ψi on the Ramachandran map. In addition, ωi0 is set to 0 and π (radian) only in the case of a cis-Pro residue. Kt and Kω are constants and are 10 and 50, respectively.
Uchi relates to the chirality of Cα and is represented by the following formula (15).

Here, τi is a twist angle determined by N-Cα-Cβ-C, and Kchi is 50. Uhydr is related to hydrogen bonding of the main chain conserved in homologous protein, and is defined as the following formula (16).

The hydrogen bond is set when the distance between the N atom and the O atom is 2.9 ± 0.5cm. When it is determined whether or not there are hydrogen bonds in a plurality of reference proteins, it is determined that hydrogen bonds are present when it is recognized that 75% or more of the reference proteins are present. Khydrr is a constant and is 0.6.
Next, optimization of the main chain atoms including Cβ is performed by simulated annealing. In this annealing process, the perturbation of the main chain and Cβ atoms is made to be within 1.0 mm with respect to the initial position. This annealing step is performed 200 times for the main chain and Cβ atoms. The parameter corresponding to the temperature starts from 50 or 25 and is increased by 0.5 times each time until it reaches 0.01, and then is set to a constant value.
In order to sample the configuration of the main chain widely, in the method of the present invention, the above method is preferably performed six times, and the atomic coordinates of the main chain having the minimum objective function value are set as the optimum solution. The parameter corresponding to the temperature starts from 50 for the first two times and starts from 25 for the third time.
Step I-83: Construction and optimization of side chain atomic coordinates
The construction of side chains is roughly divided into two stages: “Side chain construction of structure-conserving sites” (Step I-83 (1)) and “Whole side chain construction” (Step I-83 (2)). Divided.
Step I-83 (1): Construction of side chain of structure-preserving site
For the calculated main chain atoms, the twist angle of the side chain is obtained from the homologous protein using the method in the previous study. The details of this method can be found in “The role of played by the affairs of the mine. The new members of the world.” Struct. Funct. Genet. 1998, 31, 255-369.
In this method, the percentage of side chains conserved among homologous proteins is calculated, and side chains are modeled based on this information. The atomic coordinates of the side chain of the conserved part of the side chain are placed with respect to the fixed main chain atom. For example, if the χ1 angle of the arginine residue is conserved in a homologous protein, the coordinates of the Cγ atom can be placed, and if the χ1 and χ2 angles are conserved in the Phe residue, all side chain atoms Can be put. The process of optimizing simulated annealing using Equation (7) was performed only for the main chain and Cβ atoms, and the perturbation of atoms was within 1.0 cm. This stage of annealing the main chain and Cβ atoms is performed 200 times. The parameter corresponding to the temperature starts from 25 and is increased by 0.5 times each time until it becomes 0.01. Unon-bond in the formula (7) is performed on the side chain atom partially created with the main chain atom. At this time, the coordinates of the side chain atoms are preserved throughout the optimization process.
The pair of Mi and hydrogen bond N—O as structural information is used in the optimization process. In order to obtain the arrangement of main chain atoms, the above process is repeated three times, and the coordinates of the main chain atom having the minimum objective function are used as the calculated structure.
Step I-83 (2): Construction of the entire side chain
The side chain is constructed under a fixed main chain and Cβ atoms. This is described in Ogata K and Umeyama H, Prot. Struct. Funct. Genet. The research results disclosed in 1998, 31 and 255-369 are carried out, and by using them, an accurate model can be given in a short time. Next, the main chain structure is optimized by the Monte Carlo method at a low temperature, the temperature is set to 0.001, and the objective function Unon-bond of Equation (7) is used to calculate for all main chain and side chain atoms. Then, the coordinates of the side chains are rearranged so that the twist angle of the side chains is optimized in the process of optimizing the N, Cα, C, and Cβ atoms. The perturbation of atoms should be within 0.5mm. The side chain is then deleted and the above side chain construction is repeated. This process is repeated until there is no collision between the 2.4Å atoms and the twist angle of N-Cα-Cβ-C falls within the range of −120 ± 15 °.
Step I-90: Completion of the three-dimensional structure
Based on the alignment obtained in step I-70 as described above, a three-dimensional structure is constructed using modeling software such as FAMS in step I-80 to complete the model. The method shown in Step I-80 above may be referred to as FAMS. The details of this method are described in Koji Ogata and Hideaki Umeyama, “An automatic homology modeling and confederation of database 2 of 18 and the like.
Step I-100: Three-dimensional structure inspection
In Step I-100, it is examined whether or not a hydrophobic core of the completed three-dimensional structure model is formed, and the final structure of Step I-110 is reached.
Step I-110: Construction of the final structure
Thus, atomic coordinates that define the three-dimensional structure of the target protein can be obtained.
According to the method of the present invention (alignment correction method), in the three-dimensional structure construction of a protein by homology modeling, when alignment of amino acid sequences is given, three-dimensional structure information is obtained from the coordinates of the reference protein and alignment is performed. Since the correction is made, a more accurate three-dimensional structure can be constructed. The alignment is corrected based on the secondary structure information of the protein, the hydrophobic core value information, the hydrophobic core distance information, etc., thereby inserting the amino acid sequence fragment in and around the secondary structure that is a structural unit of the protein three-dimensional structure. And a model that does not contain defects can be obtained. In addition, the method of the present invention is a method that takes into account the physical hydrophobic interaction that forms the three-dimensional structure of a protein when correcting the alignment. It is possible to prevent insertion and deletion.
II. Construction method of 3D structure by Advanced FAMS
Next, Advanced FAMS as the second and third aspects of the present invention will be described.
FIG. 6 is a flowchart showing an example of a method for constructing a three-dimensional structure by Advanced FAMS of the present invention. Hereinafter, the present invention will be described in detail by taking each step of FIG. 6 as an example.
Step II-10: Amino acid sequence of the target protein
First, the amino acid sequence of the target protein for which a three-dimensional structure is to be constructed is input. As described in detail above, the amino acid sequence to be used may be any sequence derived from a database, a sequence registered for the first time, or the like. In addition, a protein sequence for which only a partial structure has been analyzed can also be used as a three-dimensional structure construction target in the present invention in order to obtain complete three-dimensional structure information.
Step II-20: Database search and sequence alignment using high homology alignment software
The amino acid sequence input in Step II-10 is aligned using software that performs high homology alignment, for example, FASTA. That is, a reference protein is selected from a three-dimensional structure database such as PDB by FASTA, and alignment (alignment) with the amino acid sequence of the selected reference protein is performed.
The reference protein here is a database of sequences and coordinates acquired from a three-dimensional structure database, and is obtained from a public database registered as a protein data bank (PDB). 26243 three-dimensional structures are registered in the PDB database (as of August 2000), but those with 95% or higher sequence match are judged to be the same category and have the longest sequence among the categories. In the case of the same length, the structure having the highest X-ray resolution was selected as the representative of the category. The three-dimensional structure database used in the present invention is a database of these representatives. In the present invention, 3922 representative structures are used as the PDB database.
Here, the details of the alignment software FASTA for high homology are described in “Effective protein sequence comparison.” Pearson WR, (1996) Methods Enzymol; 266: 227-58, as described above. This program searches a three-dimensional structure database for a sequence having a high degree of coincidence with the target sequence, and calculates the degree of coincidence between the final target sequence and the reference protein as an e value.
The e value is a quantitative description of the random background noise that exists during matching between sequences. It indicates how well the two sequences match, has a property that decreases exponentially with respect to the score, and is useful as a method for setting a significant threshold for the result. In this step, for example, when the e value calculated by FASTA has a value of 0.01 or less, it is determined that the alignment is significant.
Step II-30: Database search and sequence alignment by profile alignment software
The amino acid sequence input in Step II-10 is aligned using software for aligning the low homology region, that is, software for performing profile alignment, for example, PSI-BLAST. PSI-BLAST is programmed to perform profile alignment. Details of PSI-BLAST are as described above. Bioinformatics 1999, 12, 1000-11.
As an amino acid sequence used as a profile, for example, PIR: http: // www-nbrf. An amino acid sequence such as geology, edu / pir / (National Biomedical Research Foundation (NBRF)) may be used.
This program (PSI-BLAST) extracts information from only significant alignment relationships in the profile database (here in the PIR) and creates a site-specific score matrix of amino acid sequences. Next, the program searches the database for sequences with a high degree of coincidence with the site-specific score matrix created instead of the target sequence, and the site-specific score matrix is repeated until no more significant alignment is detected. It is updated sequentially every time. Then, the degree of coincidence between the final site-specific score matrix and the reference protein is calculated as an e value.
As described above, the e value is a quantitative description of random background noise that exists during matching between sequences. It indicates how well the two sequences match, has a property that decreases exponentially with respect to the score, and is useful as a method for setting a significant threshold for the result.
In Tables 1 and 2 below, in CASP4 (Fourth Community Wide Exponential on the Critical Assessment of Techniques for Protein Structure Prediction: / c. A list of things and a list of things modeled with wrong folds and e-values. The probability that the correct fold has been correctly modeled is less than 0.1% is 84%, and the probability of being incorrectly modeled is 0.1% or less is 44%. Here, the definition of correctly modeled is that the protein folding and secondary structure are correctly predicted. Therefore, there are some that are correct even if the RMSD value is large (T0089). By comparing Table 1 and Table 2 below, it can be said that fairly correct modeling is possible if the e value is 0.1 or less.
There are large numbers such as 3.2 (FASTA) and 7.1 in the first table that correctly modeled the protein fold, but 0.031, 0.042 in the second table that incorrectly modeled the protein fold. , 0.48, 0.96, and similar e-values in the vicinity of 0.1 exist, so a threshold of e-values of 0.1 is considered reasonable. E value is 5 × 10^-17Or 7 × 10^-22It should be noted that even a small value such as may cause an error in the evaluation of the protein fold, so that a fairly small e value does not completely guarantee the correct fold. In Tables 1 and 2, “◯” or “◎” indicates that the inequality “e <0.1” or “ID> 20” is satisfied.

Step II-40: Realignment with reference software for high homology between reference proteins and elimination of sequence redundancy
In Steps II-20 and II-30, the high homology alignment software (FASTA) is used among a plurality of alignments calculated in the high homology alignment software (FASTA) and the profile alignment software (PSI-BLAST). Alignment is performed, the amino acid sequences of the reference proteins are compared altogether to eliminate redundancy, and the alignment with the higher e-value from the target sequence and the reference protein are excluded. In this case, specifically, for example, when the FASTA and the PSI-BLAST target the reference protein having the same PDB ID, the FASTA alignment is adopted for the PDB ID, and the alignment of the PSI-BLAST is performed. Exclude.
How much redundancy is eliminated in this step is very important in later steps. In the present invention, it was investigated how many clusters of 3922 PDB databases can be divided into clusters having the same FASTA e value or less as the same cluster. The result is shown in FIG. As is apparent from FIG. 7, it can be seen that when the e-value is larger than 0.01, the number of clusters rapidly decreases and the structure is not diversified. Based on this result, the case where the e value is 0.01 or less is determined to be redundant as the same cluster. Thus, it is important to appropriately set the e value in determining redundancy. If this e value is set too low, there will be many similar arrays and there is no point in performing this step. On the other hand, if the value is set too high, the number of clusters is drastically reduced and the structure is not diversified. From FIG. 7, it can be seen that if the e value is set to 0.01 or less for the redundancy judgment, such a situation does not occur and the judgment can be made appropriately.
More specifically, for example, when a search is performed on a PDB which is a three-dimensional structure database with a PSI-BLAST e value of 0.1 or less for profile alignment, a plurality of alignment candidates may be obtained in a normal case. Many. It takes a lot of time to construct the Cα atomic coordinates for all of them in the next Step II-50, and many similar three-dimensional structures can be formed. The purpose of this step is to delete “similar alignment” in such a large character array. Alignment shows the relationship between the target sequence and the sequence of the reference protein, and even if it is made in large quantities, the target sequence portion is the same. That is, in this case, whether the alignment is similar depends on whether the sequence on the reference protein side is similar. In order to determine whether or not the character sequences are similar, alignment for high homology between amino acid sequences of reference proteins, for example, FASTA is performed. For example, if the e-value is 0.01 or less in FASTA, it is related to the three-dimensionally similar proteins, and modeling them as reference proteins will build similar protein coordinates Means that.
Therefore, the reference proteins are aligned with each other by high homology alignment. When it is determined that they are similar to each other, which one is to be discarded can be determined with low homology alignment (profile alignment) software such as PSI-BLAST. Those who are judged not to be similar (e value is higher) are discarded. This achieves the purpose of this step.
Step II-50: Construction and optimization of initial coordinates of Cα atom and elimination of steric redundancy
In response to the alignment results from step II-40, which eliminates the sequence redundancy, information about amino acid residues inserted and deleted from the reference protein is obtained. In the alignment, a region having no gap corresponding to three or more amino acids in succession is selected, and in this region, the Cα atom of the target protein is applied to the same as the reference protein in these residue pairs. deep. If the Cα atom is not obtained, coordinates are applied from a database of fragments prepared in advance (see FIG. 3).
Step II-51: Construction of the Cα atom by the simulated annealing method  The Cα atom created in step 50 is optimized using a function composed of the coordinates of the reference protein using a simulated annealing process. The objective function and the like are as described in Step I-81 (1) and are performed in the same manner.
Step II-52: Elimination of steric redundancy
Three-dimensional structure fitting between reference proteins is performed using the coordinates of Cα calculated in Step II-51. When multiple alignments are selected as candidates by the above steps, the Cα atom coordinates of the target protein are calculated based on the individually selected reference protein and alignment information, and the three-dimensional structure is fitted between the Cα atom coordinates. If the root mean square distance (RMSD) of each coordinate is less than a certain value, it is determined that the three-dimensional structure is the same, and the reference protein and the alignment with the higher e value are excluded from the target protein. In addition, a reference protein and alignment are selected. In this case, if the root mean square distance (RMSD) value is less than or equal to 3.0 mm, it is basically determined that the three-dimensional structure is the same, and the reference protein and alignment having a higher e value with the target sequence are determined. It is preferable to exclude. Here, as a reference material, the difference between the case where only the Cα atom coordinates are calculated and the case where the side chain is constructed is shown in Table 3 for the CASP4 target, and the average is 0.404 ± 0.162Å. Thus, approximate evaluation is possible only with Cα.

This step further eliminates unnecessary reference proteins and alignments and selects only those that are useful in constructing conformations. This step can be applied not only to the reference protein and alignment obtained from Step II-40 above, but also to those obtained by any other technique.
Step II-60: Construction of a Cα atom when an amino acid sequence is inserted in the alignment
When the obtained alignment has an insertion partial sequence, an appropriate sequence is read from a coordinate database of partial amino acid sequences created in advance, and Cα atom coordinates are constructed. If there is an insertion partial sequence (n) longer than the length obtained from the database, the Cα atom coordinates can be constructed as follows.
That is, (i) the inserted partial sequence of the target protein is deleted to the length obtained from the database, and alignment with the reference protein is performed. (Ii) Cα atomic coordinates of the target protein cut based on this alignment are constructed. (Iii) deleting a certain number of amino acid residues from the end of the deleted portion of the target protein that has been deleted (provided that the number of amino acid residues to be deleted is less than the length obtained from the database); iv) constructing the Cα atom coordinates of the target protein on the basis of the alignment between the target protein thus cut and the target protein, and (v) the original insertion partial sequence of the deleted target protein (n) from the insertion partial sequence (n) Repeat the above operation until the number of residues after deletion remaining in the part corresponding to n) is less than the length obtained from the database. Do it back.
This method will be described more specifically with reference to FIG. 8 showing an example of the above method. That is, if there are n residues in the alignment, the insertion is performed using modeling software, for example, a database of fragment partial sequences of up to 24 residues in the case of FAMS. When FAMS is used as modeling software, when n> 24, modeling of the insertion portion has been impossible conventionally, but it can be constructed by the method shown in FIG.
First, the insertion partial sequence is trimmed to 24 residues except for the sequence at the center of the insertion partial sequence n residues. Based on the trimmed alignment, the function of constructing the FAMS Cα atom is used to construct the Cα coordinates of the trimmed target sequence. Next, an alignment is created with reference to the coordinates of the cut target array. However, the 24-residue portion inserted into the deleted target sequence is not a true structure, but the central portion is deleted, and the central portion must be modeled in the next stage. If the central part is constructed as it is, the insertion part becomes distorted. For example, it is necessary to trim the terminal part by about 14 residues at both ends.
In this example, 14 residues are trimmed, but this number should not be more than 24 residues. After inserting 24 residues and cutting 14 residues, the insertion part for 10 residues is modeled. Therefore, an insertion part that requires modeling, that is, a residue obtained by subtracting the number of residues after deletion remaining in the part corresponding to the original insertion part sequence (n) of the target protein that has been deleted from the insertion part sequence (n) The number (10) is n-10 residues. If n-10 <25, a three-dimensional structure is constructed from the FAMS function. If n-10> 24, repeat the above operation again with 24 residues.
By performing the above operation, a three-dimensional structure can be constructed regardless of the number of inserted partial sequences.
This Step II-60 can be applied not only to the reference protein and alignment obtained from the above step, but also to those obtained by any other technique.
The method for constructing a three-dimensional structure based on information such as the alignment, reference protein, and Cα atom coordinates selected above is not particularly limited. For example, the method is preferably performed by the following steps 70 to 90. Here, the parameters such as the number of calculations, constants, and cut-off values described in each step of the present specification are examples of what the inventors consider most preferable, and limit the scope of the present invention. Not a thing.
Step II-70: Construction and optimization of main chain atomic coordinates
Another atom of the main chain is added to the Cα atom in Step II-52 or Step II-60, and the objective function is minimized by the simulated annealing method. When there is no insertion partial sequence that cannot be processed by the modeling software in the alignment, another atom of the main chain is added to the Cα atom constructed in Step II-52.
First, three-dimensional superposition of Cα atoms is performed, and residues having a Cα interatomic distance of 2.5 cm or less are picked up. The atomic coordinates of the main chain excluding Cα are obtained from the coordinates of the reference protein so as to minimize the distance between the Cα atoms, and are used as a model structure.
If there is no corresponding residue in the reference protein, the atomic coordinates of the main chain are created from the corresponding 4-residue protein fragment in the database. In this process, the main chain atom of residue i is selected from the residues having the smallest root mean square distance (RMSD) value between the (i−1) th to i + 2nd Cα atoms. At that time, in the N-terminal residue, the overlapping range of the Cα atom coordinates is from i-th to i + 3-th, and in the C-terminal residue and the residue immediately before it, the i-3-th to i-th and i-th are also the same. -2 to i + 1.
The main chain atom coordinates are optimized by the simulated annealing method based on the objective function of the main chain atom. The objective function and the like are as in Step I-82 and are performed in the same manner.
Step II-80: Construction and optimization of side chain atomic coordinates
The construction of the side chain is roughly divided into two stages, and is divided into “structure-conserved site side chain construction” (Step II-81) and “whole side chain construction” (Step II-82).
Step II-81: Construction of the side chain of the structure conservation site
For the calculated main chain atoms, the twist angle of the side chain is obtained from the homologous protein using the method in the previous study. The details of this method can be found in “The role of played by the affairs of the mine. The new members of the world.” Struct. Funct. Genet. 1998, 31, 255-369.
Step II-81 is performed in the same manner as Step I-83 (1).
Step II-82: Construction of the entire side chain
The side chain is constructed under a fixed main chain and Cβ atoms. This is described in Ogata K and Umeyama H, Prot. Struct. Funct. Genet. Performed with research results disclosed in 1998, 31, 255-369.
This Step II-82 is performed in the same manner as Step I-83 (2).
Step II-90: Construction of the final structure
Thus, by performing Steps II-10 to II-80, atomic coordinates that define the three-dimensional structure of the target protein can be obtained. In the process up to the step II-52, alignment in which redundancy is eliminated in terms of amino acid sequence and three-dimensional structure is obtained, and an efficient three-dimensional structure can be constructed.
The alignment correction method described above can be used in combination with the Advanced FAMS. Each of the above steps can be executed fully automatically by a computer program. Thus, if the target protein sequence is given, fully automatic homology modeling (construction of a three-dimensional structure) becomes possible.
III. 3D structure database divided into domain units
Next, a domain structure dividing method and a structure database divided into domain units according to the fourth aspect of the present invention will be described.
In the present invention, the “domain” means a unit of a tertiary structure of a protein as described above.
The domain delimiter method of the present invention performs a homology search on a sequence of a three-dimensional structure whose domain delimiter is unknown, with respect to a three-dimensional structure whose domain delimiter is unknown, It is a method characterized by determining a domain delimiter based on consensus.
As the sequence of the three-dimensional structure whose domain delimiter is known, the above-described three-dimensional structure database (domain structure database) delimited by the doin unit, for example, a sequence such as SCOP or CATH is used. The homology search is performed on these databases, and a sequence having homology with a three-dimensional structure whose domain delimiter is unknown is detected. Here, “homology” means the degree of matching or similarity of sequences. For the homology search, appropriate alignment software such as FASTA or PSI-BLAST is used.
Based on the consensus of homologous sequences for each domain detected by using these alignment software, the domain delimiter is determined for a three-dimensional structure whose domain delimiter is unknown. Here, consensus means consensus or agreement. The domain-delimited consensus may be achieved by adopting the majority method.
Each domain registered in the SCOP is given an identifier (ID) indicating its property. At the same time as the domain delimiter, it is preferable to assign a new identifier to this identifier (ID) based on consensus. Thereby, the attribution of the newly determined domain can be estimated.
Using the above method, a domain three-dimensional structure database can be constructed for a three-dimensional structure whose domain delimiter is unknown.
The update frequency of the domain structure database such as SCOP is less than the update frequency with the three-dimensional structure database such as PDB (there is a time lag in the update speed). For this reason, a three-dimensional structure database such as PDB includes many registrations of new sequences whose domain delimiters are not determined. By the method of the present invention, it is possible to efficiently cope with registration of these new structures.
By selecting a reference protein using the domain structure database thus obtained, a three-dimensional structure of a domain unit can be constructed. The three-dimensional structure can be constructed, for example, by the method described above.
In addition, as described above, a domain is a tertiary structure unit of a protein's three-dimensional structure. By using the method of the present invention, a three-dimensional structure of a domain unit can be efficiently obtained, and research on interaction between domains can be performed. A good data set can also be provided.
IV. Reference protein search method via amino acid sequence database
Next, a method for searching for a reference protein via an amino acid sequence database according to the fifth aspect of the present invention will be described.
The method for searching for a reference protein of the present invention is as follows.
(I) For any target protein, a homology search is performed on the amino acid sequence database, and the sequence of the amino acid sequence database having homology with the sequence of the target protein is detected,
(Ii) For the detected amino acid sequence, a homology search is performed with respect to the sequence of the three-dimensional structure database, and the sequence of the three-dimensional structure database having homology with the sequence of the detected amino acid sequence database is detected,
(Iii) A method characterized by obtaining an alignment between the sequence of the detected three-dimensional structure database and the sequence of the target protein, and using the three-dimensional structure database sequence as a reference protein.
First, the amino acid sequence database used for the homology search is not particularly limited, and examples thereof include the above-mentioned GCRDb, GPCRD, ExPASy, ORDB, PIR, Swiss Plot, TrEMBL, TrEMBLNEW, DAD and the like. Among these, it is preferable to use PIR from the viewpoint of ease of use.
The homology search is performed using appropriate alignment software, for example, FASTA or PSI-BLAST described above.
The acid sequence database sequence homologous to the target protein sequence thus detected is subjected to a homology search against the sequence database sequence, and the sequence database sequence homology with the amino acid sequence database sequence is detected. Is done. The three-dimensional structure database used is not particularly limited, and examples thereof include the above-described PDB, CCDC, SCOP, and CATH. These databases may be used alone or in combination. The homology search may be performed using appropriate alignment software as described above.
Thus, an alignment between the sequence of the target protein and the sequence of the tertiary structure database can be obtained via the sequence of the amino acid sequence database. A reference protein is selected based on this alignment.
In order to prevent similar sequences from being detected in the amino acid sequence database to be used, it is preferable to exclude similar sequences in advance (exclude redundancy). In this case, the degree of coincidence (similarity) of the sequences to be excluded is not particularly limited, and may be determined based on the number of detected sequences and the efficiency of processing. % Or more are preferably omitted.
A plurality of (multiple) sequences having homology may be detected from the amino acid sequence database. In this case, in order to increase the efficiency of processing, it is preferable to eliminate the redundancy of these arrays.
Exclusion of redundancy is preferably performed by cluster analysis using a high homology alignment software, for example, FASTA, which lowers the threshold stepwise. Specifically, the threshold value is initially decreased from 90% in increments of 5%. As a representative of the cluster, an array having the lowest e value with respect to the original ORF (Open Reading Frame) sequence is adopted. Stop at the threshold of cluster analysis where the number of clusters is 10 or less. However, this number is not limited to 10 as long as it is suitable for automatic analysis. In particular, in order to execute automatic modeling for a large number of amino acid sequence databases, it is necessary to reduce the number of sequences in the database.
When multiple alignments of the detected 3D structure database sequence with the target protein sequence are obtained, the matching part between the alignments is left as it is, the mismatching part is determined based on their consensus, and a single alignment is established. obtain. That is, a single alignment is calculated based on a plurality of alignment consensus. The alignment consensus may be performed using a majority method or the like.
When taking this alignment consensus, unlike the case of the domain delimiter or identifier (ID), there are many matters to be considered and it is preferable to use another program. In this case, the algorithm corresponds to selecting one from a set of a large number of elements, and an average value, median value, mode value, maximum value, minimum value, etc. can be considered as statistical methods. The majority vote corresponds to the mode value, but also includes the meaning of the median and average values. In particular, when there is no magnitude relationship, it is preferable (natural) to select the mode value.
This invention is particularly preferably performed when the homologous search of the target protein is performed on the three-dimensional structure database, but the sequence of the three-dimensional structure database having homology is not detected. That is, according to the present invention, in the search for the three-dimensional structure of the ORF, a similar amino acid sequence (reference protein) whose three-dimensional structure accuracy is guaranteed by ordinary alignment software such as PSI-BLAST, PRS-BLAST, FASTA, etc. Even in the case where it is not possible to find a reference protein, it is possible to find a reference protein via the amino acid sequence database. This means that, as shown in FIG. 35, the number of ORFs for which a three-dimensional structure can be constructed by modeling has increased remarkably.
Based on the thus-searched reference protein and alignment information, for example, a three-dimensional structure can be constructed by the method described above.
V. Particularly preferred embodiments of the present invention
The above-described first to fifth aspects of the present invention can be implemented singly or in combination with each other using a computer on which an appropriate program capable of performing homology modeling operates. An example of the most preferred embodiment of the present invention is shown in FIG. Details of each step in the embodiment of the present invention are as described above. Thereby, it is possible to comprehensively construct a three-dimensional structure (modeling) without leaking any ORF sequence by using an appropriate computer. The method shown in FIG. 9 is a method particularly suitable for automation. Thus, if the sequence of the target protein is given, automatic modeling using a computer can be executed for any sequence in a large amount of amino acid sequence database. In recent years, ORF array information has increased at an accelerated rate, and the method of the present invention enables efficient modeling of a large amount of ORF array information.
VI. Recording media and database on which atomic coordinates that define the three-dimensional structure of proteins are recorded
By storing the atomic coordinates defining the three-dimensional structure of the protein obtained by the above method in an appropriate recording medium in a predetermined format that can be used by a computer, a three-dimensional structure database of the target protein can be constructed. The database of the present invention preferably includes alignment information of the reference protein and the target protein together with the atomic coordinates. In addition, the database includes a code number, information on a reference region of a reference protein, information on a target protein, a distance between Cα atoms, and the like as desired.
In the present invention, the database also means a computer system that writes the atomic coordinates on an appropriate recording medium and performs a search according to a predetermined program. Examples of suitable recording media include magnetic media such as floppy disks, hard disks, and magnetic tapes; optical disks such as CD-ROM, MO, CD-R, and CD-RW, and semiconductor memories.
VII. Drug molecular design method
A computer on which an appropriate program capable of designing a drug molecule operates, all of the structural coordinates of a protein (hereinafter, also referred to as “target protein”) obtained by the above method, which is a target of the drug molecule, or Identification, search, evaluation or design of a drug molecule (antagonist or agonist) that interacts with a target protein using a part or the structure coordinates of a database or recording medium in which they are recorded it can.
The identification, search, evaluation or design of the drug molecule is performed based on the presence or absence of the interaction between the three-dimensional structure coordinate obtained by the method of the present invention and the three-dimensional structure coordinate of the drug molecule. In the present specification, identification, search, evaluation, or design of drug molecules may be simply referred to as drug molecular design.
The computer used for molecular design based on the interaction between the three-dimensional structure coordinates of the protein and the three-dimensional structure coordinates of the drug candidate molecule is not particularly limited as long as the computer is adjusted so that an appropriate program can be operated. There is no. There is no particular limitation on the storage medium of the computer. Examples of the program used for the molecular design include a computer program Insight II manufactured by Molecular Simulation. In particular, by using a program such as Ludi or DOCK specially created for this purpose alone or in combination, identification, search, evaluation or design can be performed more easily. In addition, docking evaluation between the three-dimensional structure coordinates of a protein and a drug molecule can be performed using, for example, BIOCES (version 3.10) manufactured by NEC.
Here, even if the drug molecule is a known one or a newly synthesized drug molecule having a new chemical structure, any drug molecule can be used as long as the three-dimensional structure can be obtained. It can be used in the inventive method. The three-dimensional coordinate of the drug molecule may be obtained by any method such as X-ray crystallography or modeling. Those whose three-dimensional structural coordinates are determined include an appropriate database such as CCDC (Cambridge Crystallographic Data Center: http://www.ccdc.cam.ac.uk/) and PDB (Protein Data Bank: http: // www.rcsb.org/pdb/).
Furthermore, drug molecules can also be designed by the method described in JP-A No. 2000-178209. In this way, by using the three-dimensional structure coordinates of the protein obtained by the method of the present invention, molecular design by a computer of drug molecules becomes possible. However, the molecular design method of the present invention is not limited to those using these programs and methods.
Drug molecular design usually has two conceptual stages. The first step is to find the lead compound, and the next step is to optimize the lead compound. Both steps can be performed by methods known per se using the three-dimensional structure coordinates of the target protein. This makes it possible to obtain an optimal candidate molecule for medicine and agrochemical.
VIII. Screening method for candidate molecules for medicine and pesticides obtained by molecular design method
The pharmaceutical and agrochemical candidate molecule identified, searched, evaluated or designed by the above method can be obtained, for example, by a chemical synthesis method known per se, depending on the nature of the molecule. However, the drug molecule may be a natural compound or a synthetic compound, and may be either a high molecular compound or a low molecular compound. The obtained medicinal and agricultural chemical candidate molecules are further examined by pharmacological or physiological tests in vitro or in vivo by a method known per se, and medicinal and agricultural chemical candidate molecules having a desired activity are selected. By this, what can actually be applied as a medical pesticide can be obtained.
IX. Method for producing pharmaceutical and agrochemical composition
The medical pesticide selected by the above screening method, for example, a pharmaceutical molecule, can be administered to a patient having a disease to be treated by itself, or a mixture of one or more of these active ingredients. It can also be administered. Further, it is preferable to formulate the substance as a pharmaceutical composition using a pharmacologically acceptable additive for pharmaceutical preparation and the like and administer it. For example, tablets, capsules, granules, fine granules, powders, pills, microcapsules, liposome preparations, troches, sublinguals, liquids, elixirs, emulsions, suspensions, etc. As an injection, a suppository, an ointment, a patch or the like manufactured orally as a sterile aqueous liquid or oily liquid. These are, for example, admixed with the physiologically recognized carriers, flavoring agents, excipients, vehicles, preservatives, stabilizers, binders, etc., in unit dosage forms required for accepted pharmaceutical practice. , And can be produced using methods well known in the art such as filling or tableting. The amount of active ingredient in these pharmaceutical compositions is such that an appropriate volume within the indicated range is obtained.
When the agrochemical molecule is actually used as an agrochemical, it is mixed with a carrier or diluent, additive, adjuvant, etc. by a known method to form a formulation (composition) usually used as an agrochemical, such as a powder. , Granules, wettable powders, emulsions, aqueous solvents, flowables and the like.
Example
EXAMPLES Hereinafter, the present invention will be described in more detail with reference to examples. However, the following examples should be considered as helping to obtain specific recognition of the present invention and limit the scope of the present invention in any way. is not. In the following examples, the construction of a three-dimensional structure model, alignment, search for a reference protein, etc. were performed using a personal computer manufactured by DELL (Dimension XPS T450, CPU: Pentium III 450 MHz, Red Hat Linux 7.1 J, memory: 384 Mbytes). Used.
Example 1  Effect of alignment correction method
Results of PSI-BLAST search against the three-dimensional structure database PDB of the E. coli genome (as of March 2001) showed that the first position is 100% (ie, there is a three-dimensional structure by X-ray analysis, etc.). The result of executing the alignment correction program is shown. Only the Cα atomic coordinates were modeled based on the alignment before and after the correction of the second position alignment, and the three-dimensional structure fitting was performed on the correct model structure built at the first position.
The following two parameters are used for correction.
(I) Maximum hydrogen bond length (3.05Å, 3.10Å and 3.50Å) used for secondary structure determination.
(Ii) Profile recognition method (PSI-BLAST is used here, but FASTA may also be used): how many residues a continuous region of profile information lasts (0: ignore profile, If the profile information of 1: 1 residue or more, 2: 2 residue or more, 3: 3 residue or more continues, that part on the alignment is fixed).
As shown in Table 4 below, after correction, it was shown that 74 out of 98 (75.5%) ORFs were corrected to small values so that the RMSD was improved. About three-quarters of the numerically improved modeling accuracy indicates that the alignment correction method proposed by the inventors is useful. As reference proteins used for modeling in this table, only those showing an e value of 0.001 or less in ordinary PDB database search by PSI-BLAST of ORF were used.

The three-dimensional structure improved by taking the ORF name clpP in Table 4 as an example is shown (see FIGS. 10 to 12). In this protein, the alignment before correction is 9.542 mm, but the maximum hydrogen bond length for determining the secondary structure is 3.1 mm, and the case where profile information of two or more residues is fixed is fixed after alignment correction. Improved to 8.279 mm. As is clear from FIGS. 10 to 12, the corrected model predicts the secondary structure of the correct structure well.
The details of the alignment correction will be described using the clpP as an example.
The alignment before correction is as shown in FIG.
The specific contents of the correction are as shown in FIGS. 14 to 18, and there are three correction parts.
FIG. 14 shows the first and second corrected parts before correction. The fifth line of FIG. 14 shows profile information, the sixth line shows secondary structure information, the seventh line shows the hydrophobic core value, and the hydrophobic core distance. There are two insertions in FIG. Each of them is in a β sheet (corresponding to the letter “E” on the sixth line) and an α helix (also corresponding to “H”). When the alignment correction method of the present invention is applied, the secondary structure is corrected so as not to be divided as shown in FIG.
Before the third correction, as shown in FIG. 16, the helix was divided, but the α helix was not inserted as shown in FIG. 17.
As a result, the final alignment is as shown in FIG.
By correcting in this way, it is possible to calculate not only the RMSD from 9.542Å to 8.279Å for the ORF name clpP but also the structure retaining the secondary structure, and the structure closer to the truth. All of the program algorithms here can be automated.
Example 2  Effect of 3D structure database divided into domain units
Here, a database obtained by dividing a PDB that is a three-dimensional structure database into domain units according to SCOP is used. This means that models are built on a domain basis.
The advantages of modeling using this database will be shown by taking as an example a protein called tufB, which is a protein chain elongation factor of the E. coli genome. This protein has the following amino acid sequence (SEQ ID No. 1).

This protein was also shown in the table of the effect of the alignment correction method of Example 1 above, but had the alignment shown in FIG. 19 when applied to the published PDB and had an RMSD of 10.527 mm. When sequence homology is on the order of 73.3%, the conformation is usually more similar and the RMSD should be lower.
Examination of the three-dimensional structure shows that it is as shown in FIGS. 20 and 21 and the overall RMSD is lowered because the relative position of the three domains is different.
This can be divided into the three alignments shown in FIGS. 22 to 24 when the inventor performs a search in a domain-delimited database obtained by extending the SCOP database. The three referenced proteins (IEXM-A's three domain structural proteins) are juxtaposed at different locations in the tufB of the ORF.
Furthermore, when model construction was performed with this alignment and three-dimensional structure fitting was performed, the RMSD for the correct structure was 1.586 mm, 1.013 mm, and 3.599 mm, respectively. By constructing a model with domain delimiters, it is possible to evaluate RMSD accurately and with a reasonable number, which tends to be badly evaluated due to the relative arrangement of domains. Similar trends are seen not only in tufB but also in tufA and pyrI in the same E. coli.
Example 3  Effect of database search by Advanced FAMS
The fact that the reference protein is effectively selected by Advanced FAMS is that the protein FtsA (which was also the target of CASP4 in 2000)Thermotoga  maritimaOrigin), T0089 will be described. This 3D structure is now1E4FThe 3D structure was predicted using a database that does not include this structure.
The amino acid sequence (SEQ ID No. 2) of this protein is as follows.

As a result of PSI-BLAST search, 39 PDBs shown in Table 5 below were detected as having sequence homology (PIR as of July 2000 as a profile, and PIR as of July 2000 as a search target). The PDB database was used, and the PDB database was excluded from those having a sequence homology of 95% or more). Of the 39, those having an e value of 0.1 or less were 7 as shown in Table 5 below.

After eliminating the sequence and steric redundancy, which is a function of Advanced FAMS, for this search result, three PDBs in Table 6 below remained as candidates.

Targets can be narrowed down effectively as shown in Table 7 below.
PDB with correct structure1E4FThen, the three-dimensional structure fitting was performed to calculate the RMSD with the correct structure. In Table 7 below, the smaller the RMSD value, the closer to the correct structure.

From the above results, it can be seen that the structure modeled with the 7HSC A chain as the reference protein is the closest to the correct answer. This was the 7th place hit in PSI-BLAST. It hits 3rd place in Advanced FAMS. It should be noted that what is modeled using a reference protein that has the lowest e value in Advanced FAMS and hits 1st place is not necessarily the closest to the correct structure. That is, the Advanced FAMS method described in detail in II is a useful means for calculating the modeling structure closest to the correct structure.
By applying this method to a genome or the like that is a collection of gene sequences, it is possible to comprehensively and efficiently calculate a structure that is very close to the correct structure.
Example 4  Effect of reference protein search through amino acid sequence database (Part 1)
The fact that a reference protein is effectively selected by using a reference protein search method via an amino acid sequence database (database such as PIR) is a protein YBAB (target of CASP4 in 2000)H.influenzaeOrigin), T0091 will be described. As of November 2001, the three-dimensional structure is not yet registered in the PDB (http://predictioncenter.11nl.gov/casp4/targets/templates/t0091.doc.html).
A similar amino acid sequence was searched from PDB as of May 2001 using PSI-BLAST using the PIR of May 2001 as a profile for this sequence. However, instead of the PDB sequence itself, a database divided by PDB domain was used. The determination of domain breaks in the sequence is based on SCOP (http://scop.mrc-1mb.cam.ac.uk/scop/). The PDB that has not yet been registered with the SCOP was classified according to the SCOP by the inventor using FASTA. Furthermore, similar domains with amino acid sequences of 95% or more are omitted from the domain-divided PDBs. The e-value was 0.26 for the first place in the PSI-BLAST search. A low e value means that the homology of the amino acid sequence is high, and if it is 0.1 or less, it may be considered that the fold is likely to be correct. Even the accuracy of the model is not reliable.
Details of applying the present invention relating to a three-dimensional structure search for an ORF through an amino acid sequence database such as PIR under such circumstances are shown below.
PSI-BLAST was used to search for PIRs as of May 2001 (similar amino acid sequences of 95% or more are omitted. 183016 sequence sequences). E value is 0.1 or less and alignment region length Only those having 30 residues or more (26 sequences are applicable this time) were extracted.
Next, in order to narrow down the searched 26 PIR sequences to a number of 10 or less, cluster analysis was performed using FASTA to lower the threshold stepwise. In order to perform automatic modeling for a large number of ORFs, it is necessary to reduce the number of PIR sequences. Specifically, the threshold value is initially decreased from 90% in increments of 5%. As a representative of the cluster, an array having the lowest e value relative to the original ORF sequence was adopted. Stop at the threshold of cluster analysis where the number of clusters is 10 or less. However, this number is not limited to 10 as long as it is suitable for automatic analysis. Specifically, since the number of clusters is 7 at 40%, clustering is stopped, and the amino acid sequence having the lowest e value for the subject sequence (t0091.seq) is represented from each cluster. I made it. They are as follows.

With respect to seven representative sequences of this cluster, similar protein sequences were re-searched from PDB database sequences that were reprocessed into the domain units described above using PSI-BLAST. As a result, of the above 7 sequences, the following 3 sequences

On the other hand, a sequence having an e value of 0.1 or less was detected from the above three-dimensional structure database for each domain break. The PDB ID was 1G8K_G for all three amino acid sequences. With respect to these three amino acid sequences, the alignment between the ORF sequence and the PIR sequence is shown in FIGS. 25 to 27, and the alignment between the PIR sequence and the PDB sequence is shown in FIGS.
From the alignment of the CASP4 problem (t0091.seq) sequence with the sequence in the amino acid sequence database PIR and the alignment of the PIR sequence and the three-dimensional structure, the ORF sequence and three-dimensional structure of the problem (t0091.seq) An alignment with the sequence (1G8K_G) was created.
FIG. 31 shows the alignment via the PIR sequence (A86496), FIG. 32 shows the alignment via the PIR sequence (E70157), and FIG. 33 shows the alignment via the PIR sequence (H64214). .
As is clear from the above three alignments, the alignment is slightly different depending on which PIR is used. Furthermore, as an adjustment for a slightly different alignment, a final alignment as shown in FIG. 34 was obtained by using a method such as majority vote which is a C language program developed by the inventors. All the majority voting programs adopted here are written for automatic modeling.
As described above, when searching for a three-dimensional structure of an ORF, by using an amino acid sequence such as PIR, in a normal PSI-BLAST search, a similar amino acid sequence with a guaranteed three-dimensional accuracy cannot be searched from a three-dimensional structure database. Was also shown to be found.
This means that the number of ORFs capable of constructing a three-dimensional structure has increased as shown in FIG. In this case, the same can be said even if RPS-BLAST is used instead of PSI-BLAST.
Example 5  Effect of reference protein search through amino acid sequence database (Part 2)
The protein PPX1 (which was also the target of CASP4 in 2000) that the reference protein was effectively selected by using a reference protein search method via an amino acid sequence database.Streptococcus  mutansOrigin), T0087 will be described. The three-dimensional structure is registered as 1I74 in the PDB.
A similar amino acid sequence was searched from PDB as of May 2001 using PSI-BLAST using the PIR of May 2001 as a profile for this sequence. However, instead of the PDB sequence itself, a database partitioned by PDB domain was used. Furthermore, similar domains with amino acid sequences of 95% or more are omitted from the domain-divided PDBs. As a result of the PSI-BLAST search, no PDB was detected with an e value of 0.1 or less. Details of applying the present invention under such circumstances are shown below.
PSI-BLAST was used to search for PIR as of May 2001 (amino acid sequences with 95% or more of similar sequences are omitted. 183016 sequences). The e-value is 0.1 or less and the alignment region length is 82 PIR sequences of 30 residues or more were extracted.
Next, in order to narrow down the sequence of 82 PIRs that were searched to 10 or less, stepwise cluster analysis was performed using FASTA. Specifically, the threshold value is initially decreased from 90% in increments of 5%. As a representative of the cluster, a sequence having the lowest e value relative to the original sequence was adopted. Should be stopped when the number of clusters is 10 or less, the number of clusters was 32 at 35%. Although candidates are not narrowed down to 10 or less, lowering the threshold value will leave PIRs with low homology among clusters as candidates. Therefore, in order to clear this problem, the clustering was stopped at this 35% point, and the top 10 of the 32 clusters having the lowest e value were selected. Of course, the representative of each cluster has chosen the amino acid sequence with the lowest e value. In this way, 10 were selected from the amino acid sequence database for the subject sequence (t0091.seq) of CASP4.
They are shown below.

Regarding the representative 10 sequences of this cluster, similar protein sequences were re-searched from the above-mentioned PDB sequences using PSI-BLAST. As a result, the PDB sequence of the three-dimensional structure domain database in which 3 sequences out of the 10 sequences have an e value of 0.1 or less has been found.
They are shown below.

For these amino acid sequences, the alignment between the ORF sequence (T0087) and the PIR sequence is shown in FIGS. 36 to 38, and the alignment between the PIR sequence and the PDB sequence is shown in FIGS. 39 to 41. . ,
Finally, the alignment of the CASP4 assignment (t0087.seq) with the sequence in the amino acid sequence database PIR, and the alignment of the PIR sequence with the PDB conformation are shown in FIGS. 42 to 44. Similarly, an alignment between the sequence of the problem (t0087.seq) and the sequence of the three-dimensional structure was created.
Further, a model was constructed based on this alignment, and a three-dimensional structure fitting with the correct structure PDB: 1I74_A was tried. As a result, the RMSD values were 15.862 mm, 13.606 mm, and 13.159 mm, respectively. If it was submitted to CASP4 with this result, it would be recognized as a good result. By the way, 13.159１５ is ranked 13th out of 73 teams participating in the CASP4 contest.
Moreover, it was shown that the correct answer at the level of an international contest can be approached by the alignment automatically calculated about the target which cannot be detected by normal PSI-BLAST.
Industrial applicability
Compared with the conventional method etc., the superior point of the present invention is as follows.
(1) In the construction of a three-dimensional structure by homology modeling, when alignment (alignment) of amino acid sequences is given, three-dimensional structure information is obtained from the coordinates of the reference protein, and the alignment is corrected. Construction is possible.
(2) The alignment is corrected based on the secondary structure information, hydrophobic core value information, hydrophobic core distance information, etc. of the protein. A three-dimensional structure model in which insertion and deletion of amino acid sequence fragments are not introduced can be obtained.
(3) In addition, when the alignment is corrected, insertion and deletion of amino acid sequence fragments can be prevented from entering the space where the physical hydrophobic interaction that forms the three-dimensional structure of the protein works and in the vicinity thereof.
(4) When using the homology modeling method, by using the output results of two types of alignment software having different properties, such as FASTA and PSI-BLAST, for any target protein of high homology to low homology, A highly accurate three-dimensional structure can be constructed.
(5) When the alignment software of (4) has calculated a plurality of alignments, for example, (i) FASTA matching between reference proteins is performed, and redundancy of amino acid sequences of candidate reference proteins is eliminated. (Ii) The alignment can be selected based on clear criteria such as creating coordinates of Cα atoms in the protein from the alignment, performing conformation matching, and eliminating the conformational redundancy of the candidate reference protein. It is a point. As a result, unnecessary reference proteins and alignment can be eliminated, and an efficient three-dimensional structure can be constructed.
(6) Even when the alignment includes more amino acid sequence partial fragments obtained from the database, it is possible to construct a three-dimensional structure with high accuracy.
(7) A domain structure can be separated even for a sequence of a three-dimensional structure whose domain delimiter is unknown, and a three-dimensional structure for each tertiary structure can be constructed based on the latest knowledge of the three-dimensional structure information.
(8) In the search for the three-dimensional structure of the ORF, even if a similar amino acid sequence (reference protein) in which the accuracy of the three-dimensional structure is guaranteed cannot be searched from the three-dimensional structure database by a normal alignment software search, the amino acid sequence database is used. Thus, the reference protein can be found. As a result, the number of ORFs for which a three-dimensional structure can be constructed increases significantly.
(9) Thus, when an arbitrary target amino acid sequence for which a three-dimensional structure is to be constructed is given, homology modeling (construction of a three-dimensional structure) can be performed fully automatically by a computer program.
As described above, the present invention is a method capable of accurately and efficiently constructing a more true three-dimensional structure of an arbitrary protein. In addition, the database of the present invention including the atomic coordinates defining the three-dimensional structure of the protein thus obtained can be used extremely usefully for molecular design of medicines, agricultural chemicals and the like.
Although the present invention has been described in detail and with reference to specific embodiments, it will be apparent to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the invention.
This application is based on the Japanese Patent Application (Japanese Patent Application No. 2000-367007) dated December 1, 2000 and the Japanese Patent Application (Japanese Patent Application No. 2000-368415) dated December 4, 2000, the contents of which are here. Incorporated as a reference. The contents of documents cited in this specification are also incorporated herein by reference.
[Sequence Listing]

[Brief description of the drawings]
FIG. 1 is a flowchart showing an example of a protein three-dimensional structure construction method (alignment correction method) of the present invention.
FIG. 2 is a flowchart showing an example of a method for correcting a given alignment based on information such as secondary structure, hydrophobic core value, and hydrophobic core distance in the alignment correction method.
FIG. 3 is a diagram showing a construction method of Cα atomic coordinates in step I-81. The matching part of the alignment is obtained from the reference protein, and the non-existing part is obtained from the database with the smallest root mean square distance (RMSD) of the superposition of the two residues overlapping each of the N and C ends.
FIG. 4 is a diagram showing local space homology (LSH). In the calculations for T residues in the figure, the shaded (gray) residues are taken into account. In the lower alignment in the figure, the portion surrounded by a square is a residue pair to be considered, and the ratio where the mark of * is LSH. In this case, LSH is 56.2%.
FIG. 5 is a diagram showing the relationship between LSH and the ratio in structure-conserving sites (SCRs). The LSH is calculated from the superposition of the Cα atom of the target protein and the reference protein, and the ratio in the SCRs is the number of residues in the SCRs relative to the total number of residues of the target protein.
FIG. 6 is a flowchart showing an example of a method for constructing a three-dimensional structure of a target protein of the present invention (Advanced FAMS).
FIG. 7 is a diagram showing the relationship between the logarithm of FASTA e value and the number of clusters.
FIG. 8 is an explanatory diagram showing an example of a method for constructing Cα atom coordinates in the case where there is an insertion partial sequence (n) longer than the length obtained from the coordinate database of partial amino acid sequences created in advance for alignment.
FIG. 9 is a diagram showing a basic flow of an example of a particularly preferable aspect of the present invention.
FIG. 10 is a printed photograph of the display showing the alignment model before correction.
FIG. 11 is a photograph of a display printed out showing a model of alignment after correction.
FIG. 12 is a photograph of a display printed out showing the first correct model.
FIG. 13 is a diagram showing alignment before correction.
FIG. 14 is a diagram showing the contents of the first and second corrections.
FIG. 15 is a diagram showing the contents after correction at the first and second locations.
FIG. 16 is a diagram showing the contents of the third correction.
FIG. 17 is a diagram showing the contents after the third correction.
FIG. 18 is a diagram showing the final alignment.
FIG. 19 is a diagram showing alignment between the ORF sequence (tufB) and the PDB sequence.
FIG. 20 is a printed photograph of the display showing the second model of tufB.
FIG. 21 is a printed photograph of the display showing the first model of tufB.
FIG. 22 is a diagram showing an alignment (1) between an ORF sequence (tufB) and a database sequence divided into domain units.
FIG. 23 is a diagram showing an alignment (2) between the ORF sequence (tufB) and the database sequence divided into domain units.
FIG. 24 is a diagram showing an alignment (3) between the ORF sequence (tufB) and the database sequence divided into domain units.
FIG. 25 shows the alignment (1) between the ORF sequence (T0091) and the PIR sequence (A86496).
FIG. 26 shows the alignment (2) between the ORF sequence (T0091) and the PIR sequence (E70157).
FIG. 27 shows the alignment (3) between the ORF sequence (T0091) and the PIR sequence (H64214).
FIG. 28 shows the alignment (1) between the PIR sequence (A86496) and the PDB sequence (1G8K_G).
FIG. 29 shows the alignment (2) between the PIR sequence (E70157) and the PDB sequence (1G8K_G).
FIG. 30 is a diagram showing an alignment (3) between the PIR sequence (H64214) and the PDB sequence (1G8K_G).
FIG. 31 is a diagram showing alignment with the sequence (1G8K_G) of the ORF sequence (T0091) via the PIR sequence (A86496).
FIG. 32 is a diagram showing alignment between the ORF sequence (T0091) and the 3D structure sequence (1G8K_G) via the PIR sequence (E70157).
FIG. 33 is a diagram showing an alignment between the ORF sequence (T0091) and the three-dimensional structure sequence (1G8K_G) via the PIR sequence (H64214).
FIG. 34 is a diagram showing the final alignment between the ORF sequence (T0091) and the three-dimensional structure (1G8K_G).
FIG. 35 is a diagram schematically showing that the ORF group capable of constructing a three-dimensional structure is expanded by the reference protein search method via the amino acid sequence database of the present invention.
FIG. 36 is a diagram showing alignment between the ORF sequence (T0087) and the PIR sequence (E84046).
FIG. 37 shows the alignment between the ORF sequence (T0087) and the PIR sequence (G64503).
FIG. 38 shows the alignment between the ORF sequence (T0087) and the PIR sequence (S74940).
FIG. 39 is a diagram showing alignment between the PIR sequence (E84046) and the PDB sequence (1IBJ_C).
FIG. 40 is a diagram showing alignment between the PIR sequence (G64503) and the PDB sequence (1 VPE).
FIG. 41 is a diagram showing alignment between the PIR sequence (S74940) and the PDB sequence (1G6S_A).
FIG. 42 is a diagram showing alignment between the ORF sequence (T0087) and the three-dimensional structure sequence (1IBJ_C) via the PIR sequence (E84046).
FIG. 43 is a diagram showing alignment between the ORF sequence (T0087) and the three-dimensional structure (1VPE_) via the PIR sequence (G64503).
FIG. 44 is a diagram showing alignment between the ORF sequence (T0087) and the three-dimensional structure (1G6SA1) via the PIR sequence (S74940).

Claims (5)

In a method of aligning an arbitrary target protein with a reference protein having a known three-dimensional structure, and constructing the three-dimensional structure of the target protein based on the selected reference protein and alignment information, the alignment is performed for high homology alignment and profile alignment. The selection of the reference protein and the alignment is carried out using the two types of software for the alignment of all the reference proteins detected by the two types of software. and re-alignment with the wear, if homology between the reference protein is high, the three-dimensional structure of the protein, which comprises eliminating the reference protein and alignment e value is high the said purpose protein structure Method. The method according to claim 1 , wherein the high homology alignment software is FASTA, and the profile alignment software is PSI-BLAST. When multiple alignments are selected as candidates by the software, the Cα atom coordinates of the target protein are calculated based on the individually selected reference protein and alignment information, and the three-dimensional structure is fitted between the Cα atom coordinates. If the root mean square distance (RMSD) of each coordinate is less than a certain value, it is determined that the three-dimensional structure is the same, and the reference protein and the alignment with the higher e value are excluded from the target protein. The method according to claim 1 or 2 , wherein a reference protein and an alignment are further selected. The method of claim 3 , wherein the root mean square distance (RMSD) is 3.0 Å. The creation of the three-dimensional structure is based on the selected alignment information, the coordinates are obtained from the three-dimensional structure of the reference protein selected for the Cα atom in the amino acid, and the Cα atomic coordinate is optimized so as to minimize the objective function And optimize the atomic coordinates of the main chain so as to minimize the objective function by adding other atoms of the main chain to the optimized atomic coordinates of Cα, and the side chain to the optimized atomic coordinates of the main chain the method according to any one of claims 1 to 4 which is carried out by optimizing to minimize the objective function by adding other atoms.

Images