JP3892166B2 - Method for predicting molecular reaction characteristics - Google Patents

Description

式（１）において、ｎａｔｏｍは分子１を構成する原子の個数を示し、ｃｈａｒｇｅ（ｊ）は各々の原子の電荷を示し、ＦＦfield は前述の空間占有度を示す。静電的因子（ＦＦelectro ）が大きいということは、例えば正電荷を有する反応相手が接近しにくいことを示す。
【００４５】
立体的因子（ＦＦsteric）は、各々の原子のフロンティア球面５上の各々のプローブ点７毎に、プローブ点７に設定した所定の立体特性を有するプローブ原子と分子を構成する全ての原子との間のファンデルワールズエネルギーを求め、このファンデルワールズエネルギーの相当する原子のフロンティア球面４上の和として与えられる。立体的因子（ＦＦsteric）は、分子力場ＭＭЗの中のファンデルワールズエネルギーを計算するための数学的手法を用いて演算され、式（２）によって示される。
【００４６】
【数２】

プローブ原子としては、例えば、ｓｐ3 炭素、ｓｐ2 炭素、またはｓｐ炭素を用いることが好適であり、式（２）においては、プローブ原子としてｓｐ3 炭素を用いている。式（２）において、εは原子の硬さを表すＭＭЗのパラメータを示し、ri（またはrj）はＭＭЗにおいて定義されたファンデルワールズ半径を示し、rij は相当する原子のフロンティア球面５上のｉ番目のプローブ点とｊ番目の原子との間の距離を示す。立体的因子（ＦＦsteric）が大きいということは、立体的反発が大きいため反応相手が近づきにくいことを示す。
【００４７】
なお、上述の説明において、空間占有度（ＦＦfield ）として、各々の原子２ａ，２ｂ，２ｃ・・のフロンティア球面５上に存在するプローブ点７の個数を採用した例であったが、空間占有度（ＦＦfield ）として、各々の原子２ａ，２ｂ，２ｃ・・のフロンティア球面５の面積を採用することも可能である。空間占有度（ＦＦfield ）としてフロンティア球面５の面積を採用することは、フロンティア球面５の面積が解析的に求められる場合に有用である。
【００４８】
また、上述の説明においては、分子の空間的大きさを反映するように分子包囲面２０をまず設定することを前提とするとし、そして、この分子包囲面２０として、複数の原子球面４によって構成される最外郭包絡面を採用する場合を説明した。しかしながら、当初から分子包囲面を設定することを前提とすることなく、分子１を構成する複数の原子２ａ，２ｂ，２ｃ・・の各々の原子毎に原子の中心を球面中心として原子球面４を求め、分子包囲面は、複数の原子球面４によって構成される最外郭包絡面として結果的に得られるものであると考えることも可能である。
【００４９】
また、上述の説明では、複数の原子球面４を求めて前記分子包囲空間２１を複数の構成要素空間２３に分割する場合を例にとったが、前記分子包囲空間２１を複数の構成要素空間２３に空間分割する手順としてはこれに限らない。
【００５０】
例えば、前記分子包囲面２０を複数の構成要素空間にボロノイ分割することも可能である。この場合、構成要素空間２２として、ボロノイ分割によって得られるボロノイ領域とする。このために、まず、例えば、分子のフロンティア分子軌道によって分子包囲面２０を設定する。次に、ボロノイ図の手法（例えば、ACMConputing Surveys,vol.23,no.3(1991), pp.345-405.（邦訳：杉原厚吉訳、Ｖｏｒｏｎｏｉ図−一つの基本的な幾何データ構造に関する概論，ｂｉｔ １９９３年９月号別冊、コンピュータ・サイエンス、共立出版、東京、１９９３、ｐｐ．１３１−１８５）を参照）に従い、分子１を構成する原子２ａ，２ｂ，２ｃ・・の中心を母点とし前記分子包囲面２０で囲まれた分子包囲空間２１をボロノイ分割し、原子毎に得られるボロノイ領域を得る。得られたボロノイ領域の輪郭面は、前記分子包囲面２０上にあるフロンテイア包囲面と、隣接するボロノイ領域の境界面とから構成されることになる。
【００５１】
前記構成要素空間２２としてボロノイ領域を採用する場合、分子１を構成する原子２ａ，２ｂ，２ｃ・・の大きさの違いを反映させるために、各原子のファンデルワールズ半径の比に基づき重み付けをし、ボロノイ分割して求めたボロノイ領域を補正して用いることが有用である。
【００５２】
また、前記構成要素空間２２としてボロノイ領域を採用した場合に、空間占有度（ＦＦfield ）として、ボロノイ領域の体積に関する量を採用することが有用である。
【００５３】
上述のように、前記構成要素空間２２としてボロノイ領域を採用することにより次のような効果を奏することができる。すなわち、本願発明の分子の反応特性予測方法を適用する目的に応じて、その目的に好適な分子包囲空間を選択することが可能になる。例えば、官能基の特性を反映させて官能基毎に構成要素空間を設定することも可能になる。
【００５４】
次に、図２乃至図４を参照して、ＦＲＡＵの適用例を以下に説明する。
【００５５】
分子の反応特性の類比の有無を表す識別子としてＦＲＡＵ特性（ＦＦ）がいかに有効であるかが以下に立証される。
【００５６】
図３及び図４に示すように、主にＭｇ原子またはＢ原子を含む４２種類の金属試薬を例にとる。各々の金属試薬はそれなりの機能を有しているので、ＦＲＡＵの有効性を検証する上で適している。
【００５７】
図３及び図４には、記号”ａ”、”ｂ”・・・”ｅ”、”ｆ”、”ｇ”で表示された単位方形によって、特性が類似する金属試薬が分類されている。
【００５８】
図３及び図４に示す金属試薬において、Ｍｇ原子、Ｂ原子、水素原子、炭素原子等について、計１５２個の原子の各々についてＦＲＡＵ特性（ＦＦ）を求める。
【００５９】
そのために、まず、上述の金属試薬の幾何学的及び電子的な構造をab initio RHF/3-21G^**の分子軌道（ＭＯ）計算によって最適化した。次に、金属試薬の各々の原子のＦＲＡＵ特性（ＦＦ）を、上述した最適化された幾何学的構造と原子の電荷とを用いて演算した。半径がファンデルワールズ半径である原子球面４を採用し、プローブ点７に＋１の単位電荷を置き、静電的因子（ＦＦelectro ）を求めた。プローブ原子としてｓｐ³炭素を用い、立体的因子（ＦＦsteric）を求めた。金属試薬の各々の原子のＦＲＡＵ特性（ＦＦ）としては、図３及び図４に示す金属試薬のマグネシウム（Ｍｇ）原子、水素原子、炭素原子について、計１５２個の原子のＦＲＡＵ特性（ＦＦ）を求めた。
【００６０】
さらに、識別子としてのＦＲＡＵ特性（ＦＦ）を用いて、上述の金属試薬の集合は、自己組織化ニューラルネットワークの一つであるコホーネンニューラルネットワーク（Kohonen neural network）（T.Kohonen,Biol.Cybern. 1982,43,59.を参照）の手法を用いて解析された。
【００６１】
コホーネンニューラルネットワーク（Kohonen neural network）は、多次元空間における入力データを、入力データの間のトポロジー的関係を保持してコホーネンマップ（Kohonen map ）に射映する。コホーネンマップにおけるデータを分類したグループの間の境界は、例えば、Ｕ- マトリックス（A.Ultsch等, Proc. Transputer Anvender Treffen/World Transputer Congress TAT/WTC 93 Aachen, Springer-Verlag, New York, 1993, pp. 194-203.を参照）によって認識される。
【００６２】
図２に、反応特性予測用マップとして得られたコホーネンマップ（Kohonen map ）１０を示す。コホーネンマップ１０において、各々の小さい単位方形はニューロンを示す。ニューロンの個数は、図３及び図４に示す金属試薬におけるＭｇ原子、Ｂ原子、水素原子、炭素原子等の対象とする原子の個数１５２を含むように、１８×１８個のニューロンが設けられている。
【００６３】
３種類のＦＲＡＵ特性（ＦＦ）である空間占有度（ＦＦfield ）と静電的因子（ＦＦelectro ）と立体的因子（ＦＦsteric）によって、それらの値に対応する一つの３次元ベクトルが構成される。３種類のＦＲＡＵ特性（ＦＦ）から構成された一つの３次元ベクトルが、コホーネンマップ１０に対する一つの入力データとなる。ここでは、１５２個の入力データが存在する。
【００６４】
次に、コホーネンマップ１０の作成過程を説明する。
【００６５】
コホーネンマップ１０の各々のニューロンに１５２個の入力データを次のようにして植え付ける。
【００６６】
まず、初期状態においては、コホーネンマップ１０には、各々のニューロンに種々の大きさと方向を有するベクトルが予めランダムに分配されている。次に、初期状態にあるコホーネンマップ１０を、入力データの３次元ベクトル分布を反映するように入力データの３次元ベクトルでトレーニングする。このトレーニングは、入力データの３次元ベクトルともっとも近いユークリッド距離をもつニューロンとその近傍ニューロンのベクトルを、その入力データの３次元ベクトルの大きさと向きに沿うように所定の手法によって修正する操作を、全入力データについて所定の回数実行することによって達成される。最後に、このようにしてトレーニングされたコホーネンマップ上のニューロンのベクトルと、最も近いユークリッド距離を持つ入力データが、そのニューロン上に割り付けられる。このようにして全ての入力データ（１５２個）が割り付けられ、図２に示すコホーネンマップ１０が得られる。
【００６７】
図２に示すコホーネンマップ１０は、実際はトーラス形状を有し、最上の線１０ａと最低の１０ｂとは同一の線を示し、左側の線１０ｃと右側の１０ｄとは同一の線を示す。
【００６８】
図２に示すコホーネンマップ１０において、記号”ａ”、”ｂ”・・・”ｅ”、”ｆ”、”ｇ”で表示された単位方形は、図３及び図４に示す同一の分類記号で示す金属試薬に対応する。
【００６９】
以下には、簡単のために、１５２個の入力データの内の一部のデータに着目し、説明する。
【００７０】
コホーネンマップ１０では原子の種類の違いでグループを形成して分布した。コホーネンマップ１０には図３及び図４に示す金属試薬のマグネシウム（Ｍｇ）原子、水素原子、炭素原子の分布の様子を示してある。
【００７１】
図２において、水素（Ｈ）原子の分布を詳しく見ると、それぞれ結合している金属原子の違いでグループを形成していることがわかる。すなわち、マグネシウム（Ｍｇ）原子に結合している水素原子（図３における“ａ”）とホウ素（Ｂ）原子に結合している水素原子（図３における“ｂ”）とでそれぞれグループを形成している。
【００７２】
炭素（Ｃ）原子の分布を詳しく見ると、それぞれ結合している金属原子の違いでグループを形成していることがわかる。すなわち、マグネシウム（Ｍｇ）原子に結合している炭素原子（図３における“ｃ”）とホウ素（Ｂ）に結合している炭素原子（図３における“ｄ”）とでそれぞれグループを形成している。
【００７３】
マグネシウム（Ｍｇ）原子の分布を詳しく見ると、反応における役割ごとにグループを形成していることがわかる（前述のとおり、コホーネンマップ１０の実際の形状はトーラス型であるので、上下、左右は連結している）。すなわち、それぞれ還元剤（図３における“ｅ”）、Grignard試薬（図３における“ｆ”）、塩基（図３における“ｇ”）とでグループを形成している。さらに、Grignard試薬グループと塩基グループ同士はGrignard試薬グループ同士と還元剤グループ同士や塩基グループと還元剤グループ同士よりも近くに位置しており、Ｕ−マトリックス法から得られる境界も、Grignard試薬グループ−塩基グループ間では低く（すなわち差が小さく）、Grignard試薬グループ−還元剤グループ間、塩基グループ−還元剤グループ間は高い（すなわち差が大きい）結果となった。
【００７４】
このことは、Grignard試薬が見方を変えれば強塩基であり、強弱の違いはあるが同じ塩基に属するものであること、また、別の見方ではGrignard試薬はアルカンのマグネシウム塩であり、ここに示した塩基はアミンのマグネシウム塩である点でも類似の化合物であることと本質的に矛盾しない結果となっている。このことは、ＦＲＡＵ特性値が、化合物が本来持っている特性をよく反映していることを示している。
【００７５】
さらに、ここには示していないが、Grignard試薬グループ、塩基グループ、還元剤グループそれぞれの内容を詳しく見ると、構造類似したものが近くに位置していることがわかる。たとえば、Grignard試薬はアルキル鎖の共通のものが近くに位置している。
【００７６】
以上のように、ＦＲＡＵ特性値は構造や反応における役割の類似性をよく反映していることがコホーネンニューラルネットワークによる解析で明らかにすることができた。
【００７７】
さらに、ここには示していないが、各ニューロン上の各ＦＲＡＵ特性値の内容と隣接するニューロン間の各ＦＲＡＵ特性値の差とを詳しく調べることで、どのＦＲＡＵ特性値がグループの境界線を主として決めているかを知ることができた。すなわち、どのＦＲＡＵ特性値が試薬の構造類似性・反応における役割の類似性と大きく関連しているかを知ることができた。
【００７８】
このように、ＦＲＡＵ特性値とコホーネンニューラルネットワークによる解析を組合わせた手法は、化合物の持つ反応特性の内容を明らかにする可能性を持つ。
【００７９】
以上のように、コホーネンマップ１０から、同一の種類の反応特性を示す原子は同一あるいは隣接するニューロンに集められていることが認められる。
【００８０】
図２乃至図４を参照すれば明らかなように、ＦＲＡＵに従って得られるＦＲＡＵ特性（ＦＦ）と金属試薬の役割や構造との間に、極めてよい相関関係があることが認められる。
【００８１】
以上、説明したようにＦＲＡＵは、分子１の周りの電子的特徴及び立体的特徴をＦＲＡＵ特性（ＦＦ）として数値的に特性化することができ、さらに、分子１を細かく分割するサイト毎に、すなわち分子１を構成する原子毎にＦＲＡＵ特性（ＦＦ）を付与することができるのである。
【００８２】
また、ＦＲＡＵは、コホーネンニューラルネットワークを応用することによって、反応を主に支配するのは、どの要素か、またどのサイトか、またそれらの要素やサイトがどの程度に反応に寄与するかを示すことができ、これによって、分子の反応特性を定量的に分析し理解することができるのである。
【００８３】
また、上述のＦＲＡＵによる手法に対応するプログラムをコンピュータ読み取り可能な記録媒体に記録することにより、この記録媒体を用いてコンピュータによって極めて迅速にＦＲＡＵ特性（ＦＦ）を演算することが可能になる。
【００８４】
なお、上述の説明においては、自己組織化ニューラルネットワークとしてコホーネンニューラルネットワークを適用した例を示したが、他の自己組織化ニューラルネットワークを適用することも可能である。
【００８５】
【発明の効果】
以上のように、本発明の構成によれば、合成化学の分野をはじめ薬物開発の分野を含む広い化学分野で適用可能であり、母核を共通にする制限を受けることなく、分子の大きさ等を異にする多種類の分子の間で反応特性の類比の有無を高い精度で予測することが可能になる。
【図面の簡単な説明】
【図１】（ａ）は反応特性予測の対象である分子を格子模型で示す図であり、（ｂ）は分子を構成する原子のファンデルワールズ半径球面と原子球面を示す図であり、（ｃ）は原子球面のインテリア球面とフロンティア球面とを示す図であり、（ｄ）は（ｃ）における一原子を抽出して示す図であり、（ｅ）は原子球面上に所定均等間隔毎に設定したプローブ点と、分子包囲面と分子包囲空間と構成要素空間とを示す図である。
【図２】本発明を適用して得られたコホーネンを示す図であり、各々の小さい単位方形はニューロンを示す。
【図３】図２に示すコホーネンに表れる表示記号で分類される分子を示す図。
【図４】図３と同様にコホーネンに表れる表示記号で分類される分子を示す図。
【符号の説明】
１ 分子
２ａ，２ｂ，２ｃ 原子
３ ファンデルワールズ半径球面
４ 原子球面
５ フロンティア球面
６ インテリア球面
７ プローブ点
２０ 分子包囲面
２１ 分子包囲空間
２２ 構成要素空間
２３ 構成要素包囲面[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a molecular reaction characteristic prediction method, and more particularly to a molecular reaction characteristic prediction method, a reaction characteristic prediction map, and a computer-readable recording medium for predicting a correlation characteristic of molecular reaction characteristics between a plurality of molecules. About.
[0002]
[Prior art]
In drug development, activity prediction and drug design are performed, and the correlation between the structure of the drug and the receptor, the interaction between the two, and the pharmacological activity of the drug on the living body is examined. It is predicted that it has the desired activity, and drug design is performed.
[0003]
Also, in the field of synthetic chemistry, reactivity prediction and synthesis design are performed, and the correlation between the structure of the reactants, the interaction between the reactants, etc., and the reactivity is investigated, and the assumed reaction It is predicted that a sufficient amount of product will be obtained, and an optimal synthesis route is designed.
[0004]
In the field of drug development, a CoMFA method (Coparative molecular field analysis) is conventionally known as a method for predicting reaction characteristics of reactant molecules (for example, RDIII Cramer et al., J. Am. Chem. Soc., 1988,110,5959).
[0005]
In the CoMFA method, a CoMFA field is generated to perform a three-dimensional representation of a chemical structure. In the CoMFA field, for example, a rectangular parallelepiped region surrounding a molecule whose reaction characteristics are to be predicted is formed, and lattice points distributed in a lattice form in the surrounding region are considered as probe points, and a probe atom is placed at each probe point. , By calculating the energy of interaction between the probe atoms and constituent parts such as substituents constituting the molecule.
[0006]
In the CoMFA method, the portion occupying the main part of the molecule whose reaction characteristics are to be predicted is used as a common nucleus, and the substituted portion is substituted with various substituents for molecules having a common mother nucleus. Based on the correlation characteristics of the CoMFA field to be used, it is determined whether or not there is an analogy of the mutual reaction characteristics between molecules having a common mother nucleus.
[0007]
[Problems to be solved by the invention]
However, the CoMFA method assumes that the mother nucleus is shared, and when trying to predict the reaction characteristics of a certain molecule, the mutual reaction characteristics of that molecule and the molecule that shares the mother nucleus are only Whether there is an analogy is determined.
[0008]
For this reason, it has not been possible to determine whether or not there is an analogy of the mutual reaction characteristics between molecules that have greatly different molecular sizes and do not share a mother nucleus.
[0009]
Also, in the CoMFA method, rather than decomposing into small sites that make up the molecule for which reaction characteristics are to be predicted and obtaining characteristic values for each site, one CoMFA field is obtained for the entire molecule. It was unknown how each of the constituent sites contributed to the reaction, and the reaction characteristics of the molecules could not be predicted and considered with high accuracy.
[0010]
Further, in the CoMFA method, since the rectangular parallelepiped surrounding the superposed drug molecules varies depending on the size of the target drug molecule group, there is a limit that the obtained characteristic value depends on the target molecular group.
[0011]
The CoMFA method is applied to activity prediction and drug design in drug development, and the CoMFA method cannot be applied to reactivity prediction and synthesis design in the field of synthetic chemistry. It should be said that there is not yet a method for predicting reaction characteristics in the field of synthetic chemistry, and a method for predicting reaction characteristics that also serves as a guideline for activity prediction and drug design in drug development.
[0012]
Therefore, the object of the present invention is to solve the above-mentioned problems of the prior art, and can be applied in a wide range of chemistry fields including the field of drug development including the field of synthetic chemistry. It is intended to provide a molecular reaction characteristic prediction method, a reaction characteristic prediction map, and a computer-readable recording medium that can predict the presence / absence of an analogy of reaction characteristics between molecules.
[0013]
[Means for Solving the Problems]
In order to achieve the above object, a molecular reaction characteristic prediction method according to the present invention is a molecular reaction characteristic prediction method that predicts a reaction characteristic of a molecule to be predicted, and reflects the spatial size of the molecule. In this way, a molecular enclosure surface is set, and a space surrounded by the molecule enclosure surface is defined as a molecular enclosure space, and the molecular enclosure space is divided into a plurality of component spaces that characterize the reaction characteristics of molecules according to a predetermined space division procedure. Each of the plurality of atoms constituting the component space is divided so as to be included in the component space, and a contour surface surrounding the component space is defined as a component surrounding surface, and the component surrounding surface among the component surrounding surfaces A portion on the molecular surrounding surface is defined as a frontier surrounding surface of each component space, and probe points are provided at predetermined equal intervals on the frontier surrounding surface of each component space. For each component space, the degree of occupying the molecular enclosure space is determined as the space occupancy of each component space, and for each probe point on the frontier enclosure surface of each component space, Obtain the electrostatic energy between the unit charge set at the probe point and the charges of all the atoms constituting the molecule, and correspond to the sum of the component space corresponding to the electrostatic energy on the frontier surrounding surface. The probe atom and the molecule having a predetermined steric characteristic set in the probe point for each probe point on the frontier surrounding surface of each component space obtained as an electrostatic factor of the component space Van der Waals energies between all the atoms that constitute, and before the component space corresponding to the van der Waals energies Obtain the sum on the frontier enveloping surface as the corresponding three-dimensional factor of the component space, and set the space occupancy, the electrostatic factor, and the three-dimensional factor as the reaction characteristic values of the corresponding component space, Obtaining the reaction characteristic value for each of the component spaces, selecting a plurality of data molecules having known reaction characteristics, and for each of the component spaces corresponding to the atoms constituting the data molecule for each data molecule A reaction characteristic value is obtained, and the reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. It is characterized in that the reaction characteristic of the molecule to be predicted is predicted by determining that the closer the reaction characteristic is, the closer the reaction characteristic is.
[0014]
Here, the molecular surrounding surface is an outermost envelope surface constituted by a plurality of atomic spheres by obtaining an atomic sphere with the center of the atom as a spherical center for each of a plurality of atoms constituting the molecule. Features.
[0015]
In addition, each of the component space is a space surrounding each atom of a plurality of atoms constituting the molecule.
[0016]
Further, an atomic sphere is obtained with the center of the atom as a spherical center for each of a plurality of atoms constituting the molecule, and a portion of the atomic sphere that intersects the atomic sphere of another atom is defined as an interior spherical surface, and the atomic spherical surface Of these, a portion excluding the interior spherical surface is a frontier spherical surface, and the component space is a space surrounded by a surface from which the interior spherical surface is cut out and the frontier spherical surface.
[0017]
Further, an atomic sphere is obtained with the center of the atom as a spherical center for each of a plurality of atoms constituting the molecule, and a portion of the atomic sphere that intersects the atomic sphere of another atom is defined as an interior spherical surface, and the atomic spherical surface Of these, the portion excluding the interior spherical surface is a frontier spherical surface, and the frontier surrounding surface is the frontier spherical surface.
[0018]
Further, an atomic sphere is obtained with the center of the atom as the spherical center for each of a plurality of atoms constituting the molecule, and the atomic sphere has a van der Waals radius of each atom or a van der Waals radius of each atom. A spherical surface having a spherical radius that is a diameter obtained by commonly adding a predetermined thickness.
[0019]
Further, the molecular surrounding surface is a surrounding surface surrounding a space formed by the frontier molecular orbitals of the molecule.
[0020]
Further, the predetermined space division procedure is to perform Voronoi division of the molecular enclosure space with the center of the atoms constituting the molecule as a generating point, and the component space is a Voronoi region formed by the Voronoi division. Features.
[0021]
Also, the above Space occupancy Is the volume of the component space.
Also, the above Space occupancy Is the number of the probe points existing on the frontier surrounding surface.
[0022]
Also, the above Space occupancy Is the area of the frontier surrounding surface.
[0023]
The electrostatic factor is normalized by dividing the sum of the electrostatic energy on the frontier surrounding surface of the corresponding component space by the corresponding space occupancy, and the three-dimensional factor is The sum of the van der Waals energies on the frontier surrounding surface of the corresponding component space is divided by the corresponding space occupancy and normalized.
[0024]
The probe atom is sp ^Three Carbon, sp ² It is carbon or sp carbon.
[0025]
Also, For each of the data molecules of the plurality of data molecules, a three-dimensional reaction characteristic value is obtained from the values of the space occupancy, the electrostatic factor, and the steric factor obtained for each atom constituting the data molecule. A vector is formed, and these three-dimensional vectors are used as input data. Process according to the method of self-organizing neural network, and display the processing result on the reaction characteristic prediction map And predicting the reaction characteristic of the molecule to be predicted by examining the position of the molecule to be predicted in the reaction characteristic prediction map. It is characterized by that.
[0027]
The self-organizing neural network may be a Kohonen neural network, and the reaction characteristic prediction map may be a Kohonen map.
[0028]
Further, the Kohonen map is displayed in a planar shape.
Further, in the method for predicting molecular reaction characteristics according to the present invention, for each atom constituting the molecule to be predicted, an atomic sphere surrounding the atom is drawn, and a portion of the atomic sphere that intersects the atomic sphere of another atom The interior sphere, and the portion of the atomic sphere excluding the interior sphere is a frontier sphere, probe points are provided on the atomic sphere at predetermined equal intervals, and a surface for cutting the interior sphere for each atom and the surface The degree of occupying the space enclosed by the frontier sphere is determined as the space occupancy of the corresponding atom, and for each probe point on the frontier sphere of each atom, the unit charge set at the probe point and the Obtain the electrostatic energy between the charges of all atoms constituting the molecule, and the frontier sphere of the atom corresponding to the electrostatic energy For each of the probe points on the frontier sphere of each atom, and a probe atom having a predetermined steric characteristic set at the probe point and the molecule are formed. Van der Waals energies between all atoms to be obtained, and the sum of the atoms corresponding to the van der Waals energies on the frontier sphere is obtained as a three-dimensional factor of the corresponding atoms, and the space occupancy and the electrostatic The reaction characteristic value of the corresponding atom is determined as the reaction characteristic value of the corresponding atom and the steric factor, and the reaction characteristic value and reaction characteristic obtained for the molecule to be predicted are known. By comparing the reaction characteristic values obtained for the plurality of data molecules, and determining that the reaction characteristics are more similar as the reaction characteristic values are closer, Wherein the predicting response properties of the molecule is a prediction target.
[0029]
Furthermore, the reaction characteristic prediction map according to the present invention is a reaction characteristic prediction map used for predicting a reaction characteristic of a molecule to be predicted, and the plurality of data molecules are obtained by the method according to claim 1. For each of the data molecules, a three-dimensional vector of the reaction characteristic value is formed from the space occupancy obtained for each atom constituting the data molecule, the electrostatic factor, and the three-dimensional factor value, These three-dimensional vectors are processed as input data according to a method of a self-organizing neural network, and a reaction characteristic prediction map that displays the processed results, in the reaction characteristic prediction map of the molecule to be predicted It is used for predicting the reaction characteristic of the molecule to be predicted by examining the position.
Furthermore, the recording medium according to the present invention is a computer-readable recording medium recording a program for predicting the reaction characteristics of a molecule to be predicted, and the molecular envelopment reflects the spatial size of the molecule. A surface is set, and a space surrounded by the molecule-enclosing surface is defined as a molecule-enclosing space, and according to a predetermined space division procedure, the molecule-enclosing space is composed of a plurality of component spaces that characterize the reaction characteristics of molecules. Each of the atoms is divided so as to be included in the component space, and a contour surface surrounding the component space is defined as a component surrounding surface, and each of the component surrounding surfaces is on the molecule surrounding surface. And the probe points are provided at predetermined equal intervals on the frontier enveloping surfaces of the component spaces, For each component space, the degree of occupying the molecular enclosure space is determined as the space occupancy of each component space, and for each probe point on the frontier enclosure surface of each component space, Obtain the electrostatic energy between the unit charge set at the probe point and the charges of all atoms constituting the molecule, and correspond to the sum of the electrostatic energy on the frontier surrounding surface of the component space A probe atom having a predetermined three-dimensional characteristic set at the probe point for each probe point on the frontier surrounding surface of each component space, and The component space corresponding to the van der Waals energy is obtained by obtaining van der Waals energy between all atoms constituting the molecule. The sum of the frontier surrounding surface is determined as a corresponding three-dimensional factor of the component space, and the space occupancy, the electrostatic factor, and the three-dimensional factor are set as reaction characteristic values of the corresponding component space, Obtaining the reaction characteristic value for each of the component spaces, selecting a plurality of data molecules whose reaction characteristics are known, and for each of the component spaces corresponding to the atoms constituting the data molecule for each of the data molecules The reaction characteristic value is obtained, and the reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known, and the reaction characteristic value is obtained. It is characterized in that the reaction characteristic of the molecule to be predicted is predicted by determining that the reaction characteristic is more similar as is closer.
[0030]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of a method for predicting molecular reaction characteristics according to the present invention will be described below with reference to the drawings. The molecular reaction characteristic prediction method according to the present invention is a new technique for numerically characterizing electronic and three-dimensional characteristics of a three-dimensional field surrounding a molecule. F ield-characterization for r eaction a nalysis and u nderstanding).
[0031]
In FRAU, first, the molecular surrounding surface 20 (see FIG. 1E) is set so as to reflect the spatial size of the molecule. The molecular surrounding surface 20 can be set as an enclosing surface surrounding a molecular surrounding space formed by, for example, a so-called frontier molecular orbital of a molecule, or, as described in detail later, An outermost envelope surface constituted by a plurality of atomic spherical surfaces 4 can also be used.
[0032]
The space surrounded by the molecule surrounding surface 20 is divided into a plurality of component space 22 (see FIG. 1 (e)) by dividing a molecule surrounding space 21 (see FIG. 1 (e)) into a predetermined space dividing procedure. The component space 22 is a space unit that characterizes the reaction characteristics of molecules. A contour surface surrounding the component space is defined as a component surrounding surface 23, and a portion of each component surrounding surface 23 on the molecular surrounding surface is defined as a frontier surrounding surface 5 of each component space (FIG. 1). (See (d)).
[0033]
In the following, the case where the molecular surrounding surface 20 is the outermost envelope surface constituted by the atomic spheres 4 of a plurality of atoms constituting the molecule is taken as an example, and the case where a frontier spherical surface is adopted as the frontier surrounding surface 20 is taken as an example. In addition, the embodiment of the present invention will be described by taking as an example the case where the number of probe points existing on the frontier spherical surface is adopted as the space occupancy.
[0034]
First, with reference to FIGS. 1A, 1B, 1C, 1D, and 1E, a procedure for characterizing a field surrounding a molecule will be described.
[0035]
As shown in FIG. 1A, a molecule 1 that is a target of reaction prediction is typically configured by a plurality of atoms 2a, 2b, 2c,.
[0036]
In FIG. 1B, reference numeral 3 denotes a van der Waals radius sphere centered on the atom center of the atom 2a and having a van der Waals radius of, for example, 1 to 2 angstroms, and the other atoms constituting the molecule 1 in the same manner. Again, the van der Waals radius sphere is shown.
[0037]
In the FRAU according to the present invention, as shown in FIG. 1 (b), an atomic spherical surface 4 is drawn first with the atomic center of the atom 2a as the center and a radius equal to or larger than the van der Waals radius. The radius of the atomic spherical surface 4 is the van der Waals radius of each atom itself or a radius that is commonly added to the van der Waals radius of each atom by a predetermined thickness, for example, 0.1 angstrom. An atomic sphere 4 is drawn for all atoms 2a, 2b, 2c,.
[0038]
As shown in FIGS. 1 (c) and 1 (d), the portion of the atomic spherical surface 4 of each atom that intersects with the atomic spherical surface of another atom is the interior spherical surface 6, and the portion of the atomic spherical surface 4 excluding the interior spherical surface 6. Is the frontier spherical surface 5. FIG. 1 (d) shows the front sphere 5 and the interior sphere 6 of the atom 2 a extracted from the molecule 1. The frontier spherical surface 5 is a portion exposed to the outside, and can be assumed as a region mainly involved in a chemical reaction.
[0039]
The component space corresponds to a space surrounded by a surface from which the interior spherical surface 6 is cut and the frontier spherical surface 5.
[0040]
Next, as shown in FIG. 1E, probe points 7 are set on the atomic spherical surface 4 at predetermined equal intervals. The probe point 7 is a point for diagnosing the characteristics of the molecule 1 at that position.
[0041]
In FRAU, three characteristic factors are calculated in order to numerically characterize the electronic and steric characteristics of molecule 1. The three characteristic factors are space occupancy (FFfield), electrostatic factor (FFelectro), and steric factor (FFsteric), which are called FRAU characteristics (FF). The FRAU characteristic (FF) is calculated for each atom 2a, 2b, 2c,.
[0042]
The space occupancy (FFfield) is given as the number of probe points 7 existing on the frontier sphere 5 of each atom 2a, 2b, 2c,. A large space occupancy (FFfield) indicates that the proportion of the characteristics of the corresponding atoms in the overall reaction characteristics of molecule 1 is large.
[0043]
The electrostatic factor (FFelectro) is calculated for each probe point 7 on the frontier spherical surface 5 of each atom, and the unit charge set at the probe point 7 and all the atoms 2a, 2b, 2c,. Is obtained by adding the electrostatic energy to the probe point 7 on the frontier spherical surface 5 of the corresponding atom, and is expressed by the equation (1).
[0044]
[Expression 1]

In the formula (1), “name” indicates the number of atoms constituting the molecule 1, “charge (j)” indicates the charge of each atom, and “FFfield” indicates the above-described space occupancy. A large electrostatic factor (FFelectro) indicates that, for example, a reaction partner having a positive charge is difficult to approach.
[0045]
A steric factor (FFsteric) is calculated between each probe point 7 on the frontier sphere 5 of each atom and a probe atom having a predetermined steric characteristic set at the probe point 7 and all atoms constituting the molecule. The van der Waals energy is obtained as the sum of the atoms corresponding to the van der Waals energy on the frontier sphere 4. The steric factor (FFsteric) is calculated using a mathematical method for calculating van der Waals energy in the molecular force field MMЗ and is shown by equation (2).
[0046]
[Expression 2]

For example, sp3 carbon, sp2 carbon, or sp carbon is preferably used as the probe atom. In formula (2), sp3 carbon is used as the probe atom. In Equation (2), ε represents the parameter of MMЗ representing the hardness of the atom, ri (or rj) represents the van der Waals radius defined in MMЗ, and rij represents i on the frontier sphere 5 of the corresponding atom. The distance between the th probe point and the j th atom is shown. A large steric factor (FFsteric) indicates that the reaction partner is difficult to approach due to a large steric repulsion.
[0047]
In the above description, the number of probe points 7 existing on the frontier spherical surface 5 of each atom 2a, 2b, 2c,... Is employed as the space occupancy (FFfield). As the (FFfield), the area of the frontier spherical surface 5 of each atom 2a, 2b, 2c. Employing the area of the frontier spherical surface 5 as the space occupancy (FFfield) is useful when the area of the frontier spherical surface 5 is obtained analytically.
[0048]
In the above description, it is assumed that the molecular surrounding surface 20 is first set so as to reflect the spatial size of the molecule, and the molecular surrounding surface 20 is constituted by a plurality of atomic spherical surfaces 4. The case where the outermost envelope surface to be used is employed has been described. However, without setting the molecular enveloping surface from the beginning, for each atom of the plurality of atoms 2a, 2b, 2c,. It can be considered that the molecular enveloping surface is obtained as a result of the outermost enveloping surface constituted by a plurality of atomic spheres 4.
[0049]
Further, in the above description, a case where a plurality of atomic spherical surfaces 4 are obtained and the molecule surrounding space 21 is divided into a plurality of component spaces 23 is taken as an example. However, the molecule surrounding space 21 is divided into a plurality of component spaces 23. However, the procedure for dividing the space is not limited to this.
[0050]
For example, the molecular surrounding surface 20 can be Voronoi divided into a plurality of component spaces. In this case, the component space 22 is a Voronoi region obtained by Voronoi division. For this purpose, first, the molecular surrounding surface 20 is set by, for example, molecular frontier molecular orbitals. Next, methods of Voronoi diagrams (for example, ACM Computing Surveys, vol. 23, no. 3 (1991), pp. 345-405. , Bit September 1993 issue, Computer Science, Kyoritsu Shuppan, Tokyo, 1993, pp. 131-185)), the center of the atoms 2a, 2b, 2c,. The molecular enclosure space 21 surrounded by the molecular enclosure surface 20 is divided by Voronoi to obtain a Voronoi region obtained for each atom. The obtained contour surface of the Voronoi region is composed of a frontier surrounding surface on the molecule surrounding surface 20 and a boundary surface between adjacent Voronoi regions.
[0051]
When a Voronoi region is employed as the component space 22, weighting is performed based on the ratio of the van der Waals radii of each atom in order to reflect the difference in size of the atoms 2 a, 2 b, 2 c. However, it is useful to correct and use the Voronoi region obtained by Voronoi division.
[0052]
When a Voronoi region is employed as the component space 22, it is useful to employ an amount related to the volume of the Voronoi region as the space occupancy (FFfield).
[0053]
As described above, by adopting a Voronoi region as the component space 22, the following effects can be achieved. That is, according to the purpose of applying the method for predicting molecular reaction characteristics of the present invention, it is possible to select a molecular enclosure suitable for the purpose. For example, it is possible to set a component space for each functional group by reflecting the characteristics of the functional group.
[0054]
Next, an application example of FRAU will be described below with reference to FIGS.
[0055]
How effective the FRAU characteristic (FF) as an identifier indicating the presence / absence of an analogy of the reaction characteristics of molecules is demonstrated below.
[0056]
As shown in FIGS. 3 and 4, 42 types of metal reagents mainly containing Mg atoms or B atoms are taken as an example. Since each metal reagent has an appropriate function, it is suitable for verifying the effectiveness of FRAU.
[0057]
In FIG. 3 and FIG. 4, metal reagents having similar characteristics are classified according to unit squares indicated by symbols “a”, “b”... “E”, “f”, “g”.
[0058]
In the metal reagent shown in FIGS. 3 and 4, FRAU characteristics (FF) are obtained for each of a total of 152 atoms with respect to Mg atoms, B atoms, hydrogen atoms, carbon atoms, and the like.
[0059]
For this purpose, first, the geometric and electronic structure of the above-mentioned metal reagent is ab initio RHF / 3-21G ^** Optimized by molecular orbital (MO) calculations. Next, the FRAU characteristics (FF) of each atom of the metal reagent were calculated using the optimized geometry and the atomic charge described above. An atomic spherical surface 4 having a van der Waals radius was adopted, a unit charge of +1 was placed at the probe point 7, and an electrostatic factor (FFelectro) was obtained. Sp as a probe atom ^Three Carbon was used to determine the steric factor (FFsteric). As the FRAU characteristics (FF) of each atom of the metal reagent, the FRAU characteristics (FF) of a total of 152 atoms for the magnesium (Mg) atom, hydrogen atom, and carbon atom of the metal reagent shown in FIGS. Asked.
[0060]
Furthermore, using the FRAU characteristic (FF) as an identifier, the above-described set of metal reagents is a Kohonen neural network (T. Kohonen, Biol. Cybern. 1982, 43, 59.).
[0061]
The Kohonen neural network projects input data in a multidimensional space onto a Kohonen map while maintaining a topological relationship between the input data. The boundaries between the groups that classified the data in the Kohonen map are, for example, U-matrix (A.Ultsch et al., Proc. Transputer Anvender Treffen / World Transputer Congress TAT / WTC 93 Aachen, Springer-Verlag, New York, 1993, pp. 194-203.).
[0062]
FIG. 2 shows a Kohonen map 10 obtained as a reaction characteristic prediction map. In the Kohonen map 10, each small unit square represents a neuron. The number of neurons is 18 × 18 neurons so as to include the number 152 of target atoms such as Mg atom, B atom, hydrogen atom, and carbon atom in the metal reagent shown in FIG. 3 and FIG. Yes.
[0063]
The three types of FRAU characteristics (FF), that is, the space occupancy (FFfield), the electrostatic factor (FFelectro), and the three-dimensional factor (FFsteric) form one three-dimensional vector corresponding to these values. One three-dimensional vector composed of three types of FRAU characteristics (FF) is one input data for the Kohonen map 10. Here, there are 152 input data.
[0064]
Next, a process for creating the Kohonen map 10 will be described.
[0065]
152 pieces of input data are implanted in each neuron of the Kohonen map 10 as follows.
[0066]
First, in the initial state, in the Kohonen map 10, vectors having various sizes and directions are randomly distributed to each neuron in advance. Next, the Kohonen map 10 in the initial state is trained with the three-dimensional vector of the input data so as to reflect the three-dimensional vector distribution of the input data. In this training, an operation of correcting a neuron having a Euclidean distance closest to the three-dimensional vector of input data and a vector of its neighboring neurons by a predetermined method so as to follow the size and direction of the three-dimensional vector of the input data, This is accomplished by executing a predetermined number of times for all input data. Finally, the neuron vector on the Kohonen map trained in this way and the input data having the closest Euclidean distance are allocated on the neuron. In this way, all input data (152 pieces) are allocated, and the Kohonen map 10 shown in FIG. 2 is obtained.
[0067]
The Kohonen map 10 shown in FIG. 2 actually has a torus shape, the uppermost line 10a and the lowest line 10b indicate the same line, and the left line 10c and the right line 10d indicate the same line.
[0068]
In the Kohonen map 10 shown in FIG. 2, the unit squares indicated by the symbols “a”, “b”... “E”, “f”, “g” are the same classifications as shown in FIGS. Corresponds to the metal reagent indicated by the symbol.
[0069]
In the following, for the sake of simplicity, description will be given focusing on a part of the 152 pieces of input data.
[0070]
In the Kohonen map 10, groups were formed and distributed according to the type of atom. The Kohonen map 10 shows the distribution of magnesium (Mg) atoms, hydrogen atoms, and carbon atoms of the metal reagent shown in FIGS.
[0071]
In FIG. 2, when the distribution of hydrogen (H) atoms is examined in detail, it can be seen that a group is formed by the difference in the metal atoms bonded to each other. That is, hydrogen atoms bonded to magnesium (Mg) atoms (“a” in FIG. 3) and hydrogen atoms bonded to boron (B) atoms (“b” in FIG. 3) form groups. ing.
[0072]
When the distribution of carbon (C) atoms is examined in detail, it can be seen that groups are formed by the difference in the metal atoms bonded to each other. That is, a carbon atom bonded to a magnesium (Mg) atom (“c” in FIG. 3) and a carbon atom bonded to a boron (B) (“d” in FIG. 3) form groups. Yes.
[0073]
A closer look at the distribution of magnesium (Mg) atoms shows that groups are formed for each role in the reaction (as described above, the actual shape of the Kohonen map 10 is a torus type, so Connected). That is, a group is formed with the reducing agent (“e” in FIG. 3), the Grignard reagent (“f” in FIG. 3), and the base (“g” in FIG. 3). Furthermore, the Grignard reagent group and the base group are located closer to each other than the Grignard reagent group and the reducing agent group, and the base group and the reducing agent group. The results were low (ie, the difference was small) between the base groups, and high (ie, the difference was large) between the Grignard reagent group and the reducing agent group and between the base group and the reducing agent group.
[0074]
This means that the Grignard reagent is a strong base if it is viewed differently and belongs to the same base, although there is a difference in strength, and in another way, the Grignard reagent is a magnesium salt of an alkane, as shown here. The results are essentially consistent with the fact that the base is a similar compound in that it is a magnesium salt of an amine. This indicates that the FRAU characteristic value well reflects the characteristic inherent to the compound.
[0075]
Furthermore, although not shown here, when the details of the Grignard reagent group, the base group, and the reducing agent group are examined in detail, it can be seen that structurally similar ones are located nearby. For example, Grignard reagents have a common alkyl chain located nearby.
[0076]
As described above, it has been clarified by the analysis by the Kohonen neural network that the FRAU characteristic value well reflects the similarity of the role in the structure and reaction.
[0077]
Furthermore, although not shown here, by examining in detail the content of each FRAU characteristic value on each neuron and the difference in each FRAU characteristic value between adjacent neurons, which FRAU characteristic value mainly has a group boundary. I was able to know what I decided. That is, it was possible to know which FRAU characteristic value is greatly related to the structural similarity of the reagent and the similarity of the role in the reaction.
[0078]
As described above, the technique combining the FRAU characteristic value and the analysis by the Kohonen neural network has a possibility of clarifying the content of the reaction characteristic of the compound.
[0079]
As described above, it is recognized from the Kohonen map 10 that atoms having the same kind of reaction characteristics are collected in the same or adjacent neurons.
[0080]
As is clear from FIG. 2 to FIG. 4, it is recognized that there is a very good correlation between the FRAU characteristics (FF) obtained according to FRAU and the role and structure of the metal reagent.
[0081]
As described above, the FRAU can numerically characterize the electronic characteristics and the three-dimensional characteristics around the molecule 1 as FRAU characteristics (FF). Further, for each site where the molecule 1 is finely divided, That is, the FRAU characteristic (FF) can be given to each atom constituting the molecule 1.
[0082]
FRAU also applies the Kohonen Neural Network to indicate which elements dominate the reaction, which sites, and how much these elements and sites contribute to the reaction. This allows quantitative analysis and understanding of the reaction characteristics of the molecule.
[0083]
Further, by recording a program corresponding to the above-described FRAU method on a computer-readable recording medium, it becomes possible to calculate the FRAU characteristic (FF) very quickly by a computer using this recording medium.
[0084]
In the above description, an example is shown in which a Kohonen neural network is applied as a self-organizing neural network. However, other self-organizing neural networks can also be applied.
[0085]
【The invention's effect】
As described above, according to the configuration of the present invention, it can be applied in a wide range of chemical fields including the field of drug development including the field of synthetic chemistry, and the size of the molecule can be obtained without being restricted by the common mother nucleus. It is possible to predict with high accuracy whether or not there is an analogy of reaction characteristics among many kinds of molecules having different and the like.
[Brief description of the drawings]
FIG. 1A is a diagram showing a molecule for which reaction characteristics are predicted in a lattice model, and FIG. 1B is a diagram showing a van der Waals radius sphere and an atom sphere of atoms constituting the molecule. (c) is a diagram showing an interior spherical surface and a frontier spherical surface of an atomic sphere, (d) is a diagram showing one atom extracted from (c), and (e) is an illustration on the atomic sphere at predetermined equal intervals. It is a figure which shows the set probe point, the molecule | numerator surrounding surface, the molecule | numerator enclosure space, and component space.
FIG. 2 shows a Kohonen obtained by applying the present invention, each small unit square representing a neuron.
3 is a diagram showing molecules classified by display symbols appearing in Kohonen shown in FIG. 2;
4 is a diagram showing molecules classified by display symbols appearing in Kohonen as in FIG. 3;
[Explanation of symbols]
1 molecule
2a, 2b, 2c atoms
3 Van der Waals radius spherical surface
4 Atomic sphere
5 Frontier spherical surface
6 Interior spherical surface
7 Probe points
20 Molecular enveloping surface
21 Molecular enclosure
22 Component space
23 Component surrounding surface

Claims (13)

For each atom constituting the molecule to be predicted, an atomic sphere surrounding the atom is drawn, and the portion of the atomic sphere that intersects the atomic sphere of another atom is an interior spherical surface, and the interior spherical surface of the atomic spherical surface is the interior spherical surface. Except for the frontier spherical surface,
Probing points are provided on the atomic sphere at predetermined equal intervals,
For each atom, obtain the degree of space occupancy of the corresponding atom as the degree of occupying the space surrounded by the surface that cuts out the interior sphere and the frontier sphere,
For each of the probe points on the frontier sphere of each atom, an electrostatic energy between a unit charge set at the probe point and charges of all atoms constituting the molecule is obtained, and the electrostatic energy The sum of the corresponding atoms on the frontier sphere is obtained as an electrostatic factor of the corresponding atoms,
For each of the probe points on the frontier sphere of each atom, van der Waals energy between the probe atom having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule is obtained. , The sum of the atoms corresponding to the van der Waals energy on the frontier sphere is determined as a three-dimensional factor of the corresponding atoms,
The space occupancy, the electrostatic factor, and the steric factor are the reaction characteristic values of the corresponding atoms, and the reaction characteristic value for each atom is determined.
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The method for predicting molecular reaction characteristics , wherein the space occupancy is a volume of the space surrounded by a surface cut out from the interior spherical surface and the frontier spherical surface . For each atom constituting the molecule to be predicted, an atomic sphere surrounding the atom is drawn, and the portion of the atomic sphere that intersects the atomic sphere of another atom is an interior spherical surface, and the interior spherical surface of the atomic spherical surface is the interior spherical surface. Except for the frontier spherical surface,
Probing points are provided on the atomic sphere at predetermined equal intervals,
For each atom, obtain the degree of space occupancy of the corresponding atom as the degree of occupying the space surrounded by the surface that cuts out the interior sphere and the frontier sphere,
For each of the probe points on the frontier sphere of each atom, an electrostatic energy between a unit charge set at the probe point and charges of all atoms constituting the molecule is obtained, and the electrostatic energy The sum of the corresponding atoms on the frontier sphere is obtained as an electrostatic factor of the corresponding atoms,
For each of the probe points on the frontier sphere of each atom, van der Waals energy between the probe atom having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule is obtained. , The sum of the atoms corresponding to the van der Waals energy on the frontier sphere is determined as a three-dimensional factor of the corresponding atoms,
The space occupancy, the electrostatic factor, and the steric factor are the reaction characteristic values of the corresponding atoms, and the reaction characteristic value for each atom is determined.
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The method for predicting molecular reaction characteristics , wherein the space occupancy is the number of the probe points existing on the frontier sphere . For each atom constituting the molecule to be predicted, an atomic sphere surrounding the atom is drawn, and the portion of the atomic sphere that intersects the atomic sphere of another atom is an interior spherical surface, and the interior spherical surface of the atomic spherical surface is the interior spherical surface. Except for the frontier spherical surface,
Probing points are provided on the atomic sphere at predetermined equal intervals,
For each atom, obtain the degree of space occupancy of the corresponding atom as the degree of occupying the space surrounded by the surface that cuts out the interior sphere and the frontier sphere,
For each of the probe points on the frontier sphere of each atom, an electrostatic energy between a unit charge set at the probe point and charges of all atoms constituting the molecule is obtained, and the electrostatic energy The sum of the corresponding atoms on the frontier sphere is obtained as an electrostatic factor of the corresponding atoms,
For each of the probe points on the frontier sphere of each atom, van der Waals energy between the probe atom having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule is obtained. , The sum of the atoms corresponding to the van der Waals energy on the frontier sphere is determined as a three-dimensional factor of the corresponding atoms,
The space occupancy, the electrostatic factor, and the steric factor are the reaction characteristic values of the corresponding atoms, and the reaction characteristic value for each atom is determined.
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The method for predicting molecular reaction characteristics , wherein the space occupancy is an area of the frontier spherical surface .   The atomic sphere is a spherical surface having a spherical radius that is a van der Waals radius of each atom or a diameter obtained by adding a predetermined thickness in common to a van der Waals radius of each atom. 4. The method for predicting reaction characteristics of a molecule according to any one of 3 or 3. The electrostatic factor is normalized by dividing the sum of the electrostatic energy on the frontier sphere of the corresponding component space by the corresponding space occupancy,
The steric factor is normalized by dividing the sum of the van der Waals energies on the frontier sphere of the corresponding component space by the corresponding space occupancy. 4. The method for predicting molecular reaction characteristics according to any one of 2 and 3. A molecular reaction characteristic prediction method for predicting a reaction characteristic of a molecule to be predicted,
The molecular encircling surface is set to reflect the spatial size of the molecule, and the space surrounded by the molecular encircling surface is defined as the molecular encircling space,
In accordance with a predetermined space division procedure, the molecular enclosure space is divided into a plurality of component spaces characterizing the reaction characteristics of the molecule so that each atom of the plurality of atoms constituting the molecule is included in the component space. , The contour surface surrounding the component space as a component surrounding surface,
A portion of each component surrounding surface on the molecule surrounding surface is defined as a frontier surrounding surface of each component space,
Providing probe points at predetermined equal intervals on the frontier surrounding surface of each component space,
For each of the component spaces, the degree of occupying the molecular enclosure space is determined as the space occupancy of each component space,
For each of the probe points on the frontier surrounding surface of each of the component spaces, obtain the electrostatic energy between the unit charge set at the probe point and the charges of all the atoms constituting the molecule, Obtaining the sum of the component space corresponding to the electrostatic energy on the frontier surrounding surface as an electrostatic factor of the component space corresponding to
For each probe point on the frontier surrounding surface of each component space, a van der between probe atoms having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule Finding the World's energy, obtaining the sum of the component space corresponding to the van der Waals energy on the frontier surrounding surface as a corresponding three-dimensional factor of the component space,
The reaction characteristic value of the component space corresponding to the degree of space occupancy, the electrostatic factor, and the three-dimensional factor, and obtaining the reaction characteristic value for each of the component spaces,
Selecting a plurality of data molecules whose reaction characteristics are known, and determining the reaction characteristic value for each component space corresponding to the atoms constituting the data molecule for each data molecule;
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The molecular surrounding surface is an surrounding surface that surrounds a space formed by the frontier molecular orbitals of the molecule,
The predetermined space division procedure is to perform Voronoi division of the molecular enclosure space with a center of atoms constituting the molecule as a generating point, and the component space is a Voronoi region formed by the Voronoi division,
The method for predicting molecular reaction characteristics , wherein the space occupancy is a volume of the component space . A molecular reaction characteristic prediction method for predicting a reaction characteristic of a molecule to be predicted,
The molecular encircling surface is set to reflect the spatial size of the molecule, and the space surrounded by the molecular encircling surface is defined as the molecular encircling space,
In accordance with a predetermined space division procedure, the molecular enclosure space is divided into a plurality of component spaces characterizing the reaction characteristics of the molecule so that each atom of the plurality of atoms constituting the molecule is included in the component space. , The contour surface surrounding the component space as a component surrounding surface,
A portion of each component surrounding surface on the molecule surrounding surface is defined as a frontier surrounding surface of each component space,
Providing probe points at predetermined equal intervals on the frontier surrounding surface of each component space,
For each of the component spaces, the degree of occupying the molecular enclosure space is determined as the space occupancy of each component space,
For each of the probe points on the frontier surrounding surface of each of the component spaces, obtain the electrostatic energy between the unit charge set at the probe point and the charges of all the atoms constituting the molecule, Obtaining the sum of the component space corresponding to the electrostatic energy on the frontier surrounding surface as an electrostatic factor of the component space corresponding to
For each probe point on the frontier surrounding surface of each component space, a van der between probe atoms having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule Finding the World's energy, obtaining the sum of the component space corresponding to the van der Waals energy on the frontier surrounding surface as a corresponding three-dimensional factor of the component space,
The reaction characteristic value of the component space corresponding to the degree of space occupancy, the electrostatic factor, and the three-dimensional factor, and obtaining the reaction characteristic value for each of the component spaces,
Selecting a plurality of data molecules whose reaction characteristics are known, and determining the reaction characteristic value for each component space corresponding to the atoms constituting the data molecule for each data molecule;
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The molecule surrounding surface is wrapped 囲面 surrounding the space formed by the frontier molecular orbital of the molecule,
The predetermined space division procedure is to perform Voronoi division of the molecular enclosure space with a center of atoms constituting the molecule as a generating point, and the component space is a Voronoi region formed by the Voronoi division,
The method for predicting molecular reaction characteristics , wherein the space occupancy is the number of probe points existing on the frontier surrounding surface . A molecular reaction characteristic prediction method for predicting a reaction characteristic of a molecule to be predicted,
The molecular encircling surface is set to reflect the spatial size of the molecule, and the space surrounded by the molecular encircling surface is defined as the molecular encircling space,
In accordance with a predetermined space division procedure, the molecular enclosure space is divided into a plurality of component spaces characterizing the reaction characteristics of the molecule so that each atom of the plurality of atoms constituting the molecule is included in the component space. , The contour surface surrounding the component space as a component surrounding surface,
A portion of each component surrounding surface on the molecule surrounding surface is defined as a frontier surrounding surface of each component space,
Providing probe points at predetermined equal intervals on the frontier surrounding surface of each component space,
For each of the component spaces, the degree of occupying the molecular enclosure space is determined as the space occupancy of each component space,
For each of the probe points on the frontier surrounding surface of each of the component spaces, obtain electrostatic energy between unit charges set at the probe points and charges of all atoms constituting the molecule, Obtaining the sum of the component space corresponding to the electrostatic energy on the frontier surrounding surface as an electrostatic factor of the component space corresponding to
For each probe point on the frontier surrounding surface of each component space, a van der between probe atoms having a predetermined steric characteristic set at the probe point and all atoms constituting the molecule Finding the World's energy, obtaining the sum of the component space corresponding to the van der Waals energy on the frontier surrounding surface as a corresponding three-dimensional factor of the component space,
The reaction characteristic value of the component space corresponding to the degree of space occupancy, the electrostatic factor, and the three-dimensional factor, and obtaining the reaction characteristic value for each of the component spaces,
Selecting a plurality of data molecules whose reaction characteristics are known, and determining the reaction characteristic value for each component space corresponding to the atoms constituting the data molecule for each data molecule;
The reaction characteristic value obtained for the molecule to be predicted is compared with the reaction characteristic value obtained for the plurality of data molecules whose reaction characteristics are known. The closer the reaction characteristic value, the more the reaction characteristic. By judging that they are similar, predict the reaction characteristics of the molecule to be predicted ,
The molecular surrounding surface is an surrounding surface that surrounds a space formed by the frontier molecular orbitals of the molecule,
The predetermined space division procedure is to perform Voronoi division of the molecular enclosure space with a center of atoms constituting the molecule as a generating point, and the component space is a Voronoi region formed by the Voronoi division,
The molecular reaction characteristic prediction method , wherein the space occupancy is an area of the frontier surrounding surface . The electrostatic factor is normalized by dividing the sum of the electrostatic energy on the frontier surrounding surface of the corresponding component space by the corresponding space occupancy,
7. The three-dimensional factor is normalized by dividing the sum of the van der Waals energies on the frontier surrounding surface of the corresponding component space by the corresponding space occupancy. 7. The method for predicting molecular reaction characteristics according to any one of 7 and 8. The probe atom having the predetermined steric property is sp ³ carbon, sp ² carbon, or sp carbon, according to any one of claims 1, 2, 3, 6, 7 and 8. A method for predicting the reaction characteristics of molecules. For each of the data molecules of the plurality of data molecules, a three-dimensional reaction characteristic value is obtained from the values of the space occupancy, the electrostatic factor, and the steric factor obtained for each atom constituting the data molecule. Forming a vector,
These three-dimensional vectors are processed as input data according to the self-organizing neural network method, and the processing results are displayed on the reaction characteristic prediction map.
The reaction characteristic of the molecule to be predicted is predicted by examining the position of the molecule to be predicted in the reaction characteristic prediction map. The method for predicting reaction characteristics of a molecule according to any one of the above items.   12. The molecular reaction characteristic prediction method according to claim 11, wherein the self-organizing neural network is a Kohonen neural network, and the reaction characteristic prediction map is a Kohonen map.   13. The molecular reaction characteristic prediction method according to claim 12, wherein the Kohonen map is displayed in a planar shape.

Images