JP7563451B2 - Information processing device, program, and information processing method - Google Patents

Description

The present invention relates to an information processing device, a program, and an information processing method.

Conventionally, prediction models such as deductive prediction models and learning models generated by machine learning with an inductive approach have been used to predict the functions of compounds. Among these, deductive prediction models are prediction models that predict the functions of compounds from known principles and regularities of compounds. Moreover, learning models are prediction models that express the correlation between explanatory variables and objective variables, obtained as a result of inductive learning using, for example, descriptors related to the structure of a compound as explanatory variables and the function exhibited by the compound as the objective variable (for example, Patent Document 1). A method of developing materials using information processing including machine learning is called materials informatics (hereinafter referred to as "MI").

Information on compounds whose functions are predicted using such predictive models, and information on compounds used to generate learning models, can be obtained, for example, from publicly available databases.

JP 2002-73626 A

However, the number of compounds for which necessary information can be obtained from public databases is limited, and in order to increase the number of candidate compounds for function prediction or to increase the number of compounds used to generate learning models and further improve prediction accuracy, it is necessary to obtain compound information from private databases as well. However, obtaining compound structure information from private databases poses the problem that it may lead to the leakage of confidential information related to compound structures.

The object of the present invention is to provide an information processing device, program, and information processing method that can increase the security of confidential information related to the structure of compounds.

In order to achieve the above object, the present invention provides an information processing device comprising:
an information providing unit that provides the first external device with encryption algorithm information for performing encryption according to a predetermined encryption algorithm;
a first data acquisition unit that acquires encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction unit that predicts a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Equipped with
The prediction model represents the correlation between encrypted structural data obtained by encrypting structural data relating to the structure of a compound according to the encryption algorithm, and functional data relating to the function of the compound.

The present invention as set forth in claim 2 provides the information processing device as set forth in claim 1,
an encryption unit that encrypts structure data relating to a compound structure according to the encryption algorithm to generate encrypted structure data;
a learning model generation unit that generates a learning model as the prediction model based on the encrypted structure data and function data related to the function of the compound;
Equipped with.

The present invention provides an information processing device according to claim 2,
the first data acquisition unit acquires, from the first external device, encrypted structural data of a learning object encrypted according to the encryption algorithm, and functional data of a learning object related to a function of a compound corresponding to the encrypted structural data of the learning object;
The learning model generation unit generates the learning model using at least the encrypted structure data of the learning object and the functional data of the learning object acquired by the first data acquisition unit.

In order to achieve the above object, the invention of an information processing device according to claim 4 comprises:
an encryption unit that encrypts structure data relating to a compound structure according to a predetermined encryption algorithm to generate encrypted structure data;
a learning model generating unit that generates a learning model that represents a correlation between the encrypted structure data and the functional data, based on the encrypted structure data and functional data related to a function of the compound;
an information providing unit that provides a first external device with encryption algorithm information for performing encryption according to the encryption algorithm;
a first data acquisition unit that acquires, from the first external device, encrypted structure data of a learning object encrypted according to the encryption algorithm, and function data of a learning object related to a function of a compound corresponding to the encrypted structure data;
Equipped with
The learning model generation unit generates the learning model using at least the encrypted structure data of the learning object and the functional data of the learning object acquired by the first data acquisition unit.

The invention described in claim 5 is the information processing device described in claim 4,
The first data acquisition unit acquires encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device,
The information processing device includes a prediction unit that predicts, based on the learning model, a function of a compound corresponding to the encrypted structure data of the function prediction target acquired by the first data acquisition unit.

The present invention as set forth in claim 6 provides an information processing device as set forth in any one of claims 2, 3, and 5,
a structure data generating unit that generates the structure data,
The encryption unit encrypts the structure data generated by the structure data generation unit to generate the encrypted structure data of a function prediction target;
The prediction unit predicts, based on the learning model, a function of a compound corresponding to the encrypted structure data of the function prediction target generated by the encryption unit.

The present invention as set forth in claim 7 provides an information processing device as set forth in any one of claims 2, 3, and 5,
a second data acquisition unit that acquires structural data relating to a structure of the compound from a second external device that publishes the structure of the compound;
The encryption unit encrypts the structure data acquired by the second data acquisition unit to generate the encrypted structure data of a function prediction target;
The prediction unit predicts, based on the learning model, a function of a compound corresponding to the encrypted structure data of the function prediction target generated by the encryption unit.

The invention described in claim 8 is the information processing device according to any one of claims 2 to 7,
a third data acquisition unit that acquires the structure data and the function data from an external predetermined database;
The encryption unit generates the encrypted structure data based on the structure data acquired by the third data acquisition unit,
The learning model generation unit generates the learning model using at least the encrypted structure data and the functional data acquired by the third data acquisition unit.

The invention described in claim 9 provides an information processing device according to any one of claims 1 to 8,
The encryption algorithm is incapable of reversing the conversion to the structure data before encryption.

In order to achieve the above object, the present invention provides a program as set forth in claim 10,
A computer installed in the information processing device,
an information providing means for providing the first external device with encryption algorithm information for performing encryption according to a predetermined encryption algorithm;
a data acquisition means for acquiring encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction means for predicting a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Function as a
The prediction model represents the correlation between encrypted structural data obtained by encrypting structural data relating to the structure of a compound according to the encryption algorithm, and functional data relating to the function of the compound.

In order to achieve the above object, the present invention provides a program as set forth in claim 11,
A computer installed in the information processing device,
an encryption means for encrypting structural data relating to a compound structure according to a predetermined encryption algorithm to generate encrypted structural data;
a learning model generating means for generating a learning model representing a correlation between the encrypted structure data and functional data relating to a function of the compound, based on the encrypted structure data and the functional data;
an information providing means for providing a first external device with encryption algorithm information for performing encryption according to the encryption algorithm;
a data acquisition means for acquiring, from the first external device, encrypted structure data of a learning subject encrypted according to the encryption algorithm, and function data of a learning subject relating to a function of a compound corresponding to the encrypted structure data;
Function as a
The learning model generating means generates the learning model using at least the encrypted structure data of the learning subject and the functional data of the learning subject acquired by the data acquiring means.

In order to achieve the above object, the invention of an information processing method according to claim 12 comprises:
An information processing method executed by an information processing device,
an information providing step of providing encryption algorithm information for performing encryption according to a predetermined encryption algorithm to a first external device;
a data acquisition step of acquiring encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction step of predicting a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Including,
The prediction model represents the correlation between encrypted structural data obtained by encrypting structural data relating to the structure of a compound according to the encryption algorithm, and functional data relating to the function of the compound.

In order to achieve the above object, the invention of an information processing method according to claim 13 comprises:
An information processing method executed by an information processing device,
an encryption step of encrypting structural data relating to the structure of the compound according to a predetermined encryption algorithm to generate encrypted structural data;
a learning model generating step of generating a learning model representing a correlation between the encrypted structure data and the functional data, based on the encrypted structure data and functional data relating to a function of the compound;
an information providing step of providing encryption algorithm information for performing encryption according to the encryption algorithm to a first external device;
a data acquisition step of acquiring, from the first external device, encrypted structure data of a learning subject encrypted according to the encryption algorithm, and function data of a learning subject relating to a function of a compound corresponding to the encrypted structure data;
Including,
In the learning model generating step, the learning model is generated using at least the encrypted structure data of the learning subject and the functional data of the learning subject acquired in the data acquiring step.

The present invention has the effect of increasing the security of confidential information relating to the structure of compounds.

FIG. 1 is a diagram showing a schematic configuration of a compound information processing system. FIG. 13 is a diagram illustrating an example of an encryption algorithm for generating encrypted molecular structure data. FIG. 2 is a block diagram showing the main functional configuration of an MI server. FIG. 13 is a diagram showing an example of the contents of labeled encrypted molecular structure data. FIG. 2 is a block diagram showing the main functional configuration of the client server. FIG. 1 is a diagram illustrating a first method for predicting the function of a compound. FIG. 1 is a diagram illustrating a second method for predicting the function of a compound. FIG. 13 is a diagram illustrating a third method for predicting the function of a compound. FIG. 13 is a diagram illustrating a fourth method for predicting the function of a compound.

Below, embodiments of the information processing device, program, and information processing method of the present invention are described with reference to the drawings.

FIG. 1 is a diagram showing a schematic configuration of a compound information processing system 100.
The compound information processing system 100 includes an MI server 1 (information processing device), a public database server 2 (hereinafter referred to as "public DB server 2") (a specified database), a reagent database server 3 (hereinafter referred to as "reagent DB server 3") (a second external device), and a client server 4 (a first external device). The MI server 1, the public DB server 2, the reagent DB server 3, and the client server 4 are connected to each other so as to be able to communicate with each other via a communication network N. The communication network N is, for example, the Internet, but is not limited to this.

The MI server 1 is a device owned by a provider of information provision services related to materials informatics (MI), and performs various information processing related to MI. That is, the MI server 1 generates a learning model (also called a "trained model") that predicts the function of a compound based on information related to the compound through machine learning, and uses the learning model to acquire or generate information useful for material development by MI and transmit it to the client server 4. More specifically, the MI server 1 acquires a target value of the function of the compound from the client server 4, searches for a compound that exhibits the function of the target value, and transmits information related to the structure of the identified compound to the client server 4.

In detail, the MI server 1 uses a large number of combinations of encrypted molecular structure data (encrypted structure data) relating to the structure of a compound and functional data relating to the function of the compound to generate a learning model that represents the correlation between the encrypted molecular structure data and the functional data through an inductive approach by machine learning. The encrypted molecular structure data corresponds to the explanatory variables of the machine learning, and the functional data corresponds to the target variables of the machine learning. Among them, the encrypted molecular structure data is data obtained by encrypting the molecular structure data (structural data) relating to the structure of the compound according to a predetermined encryption algorithm. The number of data sets consisting of the encrypted molecular structure data and the functional data used to generate the learning model is, for example, tens of thousands of sets or more. One method for improving the prediction accuracy of the learning model is to increase the number of these data sets.
Hereinafter, the encrypted molecular structure data and the functional data used to generate the learning model are also referred to as “encrypted molecular structure data to be learned” and “functional data to be learned”, respectively. The learning model generated by the MI server 1 is one of the prediction models for predicting the function of a compound.

The molecular structure data that is the source of the encrypted molecular structure data is not particularly limited as long as it is capable of identifying the molecular structure, i.e., the elements that make up the molecule and their bonding patterns.

As an encryption algorithm for generating encrypted structure data from molecular structure data, for example, one can use one that extracts characteristics of the molecular structure of a compound according to predetermined rules and quantifies them.

FIG. 2 is a diagram for explaining an example of an encryption algorithm for generating encrypted molecular structure data.
In the encryption algorithm of Fig. 2, the structural formula shown in the upper part of the figure is converted into the code shown in the lower part of the figure according to its characteristics. Each digit of the code is set to 0 or 1. The conversion rule by the encryption algorithm of Fig. 2 can be, for example, as follows.
That is, first, each atom constituting a molecule is numbered according to the Morgan method.
Next, atomic information is given according to the Daylight rule, and fragment information contained in the molecule is added.
Next, the duplicate fragments are removed.
Finally, the resulting fragment is assigned to a predetermined digit by a hash function. For example, if a particular fragment is included in the molecule, a predetermined digit of the code is set to 1.
The encrypted molecular structure data generated by such an encryption algorithm can be said to be a kind of descriptor representing the characteristics of the molecular structure. That is, from the encrypted molecular structure data, the characteristics of the molecular structure can be specified from various aspects based on the position of the digits whose value is 1. On the other hand, since the hash function is a one-way function, it is impossible to reversely convert the encrypted molecular structure data back to molecular structure data. That is, in this embodiment, an irreversible encryption algorithm is used.

The type of learning model generated by the MI server 1 is not particularly limited as long as it represents the correlation between the encrypted molecular structure data and the functional data. As the learning model, various known models such as linear regression, principal component analysis, decision tree, random forest, support vector machine, or random forest can be used.

In addition, the MI server 1 predicts the function of the compound corresponding to the encrypted molecular structure data by applying the generated learning model to the encrypted molecular structure data of the compound whose function is to be predicted (hereinafter also referred to as "encrypted molecular structure data for which function is predicted"). The MI server 1 performs function prediction for a large number of encrypted molecular structure data, and identifies encrypted molecular structure data for which a prediction result matching the target value of function received from the client server 4 has been obtained. The MI server 1 then transmits information related to the identified encrypted molecular structure data to the client server 4.

The public DB server 2 stores molecular structure data relating to the molecular structures of a large number of compounds, and function data relating to the functions exhibited by the compounds. The public DB server 2 provides these molecular structure data and function data in response to a request from another device (in this embodiment, the MI server 1). The molecular structure data provided to the MI server 1 from the public DB server 2 is encrypted in the MI server 1 and converted into encrypted molecular structure data. This encrypted molecular structure data is used for machine learning to generate a learning model together with the function data, and can also be used as encrypted molecular structure data for function prediction targets.

The reagent DB server 3 stores molecular structure data relating to the molecular structures of many compounds that are available for sale. It can also be said that the reagent DB server 3 provides a catalog of purchasable reagents (compounds). The reagent DB server 3 provides molecular structure data in response to a request from another device (in this embodiment, the MI server 1). In this embodiment, it is assumed that no functional data relating to the function of a compound is provided from the reagent DB server 3. The molecular structure data provided from the reagent DB server 3 to the MI server 1 is encrypted in the MI server 1 and converted into encrypted molecular structure data. This encrypted molecular structure data is used as encrypted molecular structure data for a function prediction target.

The client server 4 is a device owned by a client who receives the MI-related information provision service provided by the MI server 1. The client server 4 transmits to the MI server 1 data specifying the target value of the function of the compound desired by the client, and receives from the MI server 1 information relating to the structure of the compound exhibiting said function. In order to receive the necessary MI-related information provision service, the client server 4 also transmits to the MI server 1 encrypted molecular structure data of the function prediction target, encrypted molecular structure data of the learning target, and function data of the learning target.

In this specification, the molecular structure data relating to the structure of a compound stored in the client server 4 is considered to be confidential information. In this embodiment, the molecular structure data is encrypted in the client server 4, and the resulting encrypted molecular structure data is transmitted to the MI server 1, thereby making it possible to receive the necessary information provision services without disclosing the molecular structure data, which is confidential information, to the MI server 1. The mechanism for protecting the confidential information in the client server 4 in this manner will be described in detail later.

Next, the detailed configuration of the MI server 1 and the client server 4 will be described.
FIG. 3 is a block diagram showing the main functional configuration of the MI server 1. As shown in FIG.
The MI server 1 includes a control unit 11 , an operation unit 12 , a display unit 13 , a communication unit 14 , etc., and these units are connected to each other via a bus 15 .

The control unit 11 is a processor (computer) that controls the overall operation of the MI server 1. The control unit 11 has a CPU 111 (Central Processing Unit), a RAM 112 (Random Access Memory), and a memory unit 113.

The CPU 111 reads out various control programs 113c and setting data stored in the storage unit 113, stores them in the RAM 112, and executes the programs 113c to perform various arithmetic processing.
The RAM 112 provides a working memory space for the CPU 111 and stores temporary data. The RAM 112 may include a non-volatile memory.

In addition to the above-mentioned program 113c, various data for performing information processing related to the MI are stored in the storage unit 113. As the storage unit 113, for example, a hard disk drive (HDD) is used, and a dynamic random access memory (DRAM) or the like may also be used in combination.
The data stored in the memory unit 113 includes general data 113a, client-derived data 113b, encryption algorithm information D1, and learning model data D2.

General data 113a is data relating to the structure and function of a compound that has been obtained without going through the client server 4, i.e., data obtained from the public DB server 2 or the reagent DB server 3, or data generated within the MI server 1.
Specifically, the general data 113a includes molecular structure data A1 of the learning target, its encrypted molecular structure data A2, and function data A3. The general data 113a also includes molecular structure data B1 of the function prediction target and its encrypted molecular structure data B2. Of these, the molecular structure data A1 and the function data A3 are obtained from the public DB server 2. Furthermore, the molecular structure data B1 is obtained from the public DB server 2 or the reagent DB server 3. Furthermore, as described below, the molecular structure data B1 may be generated within the MI server 1.

The client-derived data 113b is data relating to the structure and function of a compound, which is acquired from the client server 4. The client-derived data 113b includes labeled encrypted molecular structure data C2L of the target for function prediction, encrypted molecular structure data C2 of the target for learning, and functional data C3 of the target for learning.

FIG. 4 is a diagram showing an example of the contents of the labeled encrypted molecular structure data C2L.
The labeled encrypted molecular structure data C2L is data in which a unique label (here, a natural number) is associated with each of a plurality of encrypted molecular structure data that are targets of function prediction.

The encryption algorithm information D1 shown in Fig. 3 is information related to an encryption algorithm for generating encrypted molecular structure data from molecular structure data. When a specific encryption program for generating encrypted molecular structure data is executed, the encryption algorithm information D1 is referenced, so that encryption can be performed according to a specific encryption algorithm. Alternatively, the encryption algorithm information D1 may be the encryption program itself.
The encryption algorithm information D1 is used when the control unit 11 of the MI server 1 encrypts the molecular structure data A1 and B1 to generate molecular structure data A1 and B2. In addition, the encryption algorithm information D1 is transmitted to the client server 4 for encryption processing in the client server 4.

The learning model data D2 is data relating to a learning model generated by machine learning based on the encrypted molecular structure data A2, functional data A3, and/or the encrypted molecular structure data C2, and functional data C3 of the learning target. By applying the learning model represented by the learning model data D2 to the encrypted molecular structure data of the function prediction target, it is possible to predict the function of the compound corresponding to the encrypted molecular structure data. In this specification, generating the learning model data D2 is also referred to as "generating a learning model."

The control unit 11 having these configurations functions as an encryption unit (encryption means), a learning model generation unit (learning model generation means), an information provision unit (information provision means), a first data acquisition unit (first data acquisition means), a second data acquisition unit (second data acquisition means), a third data acquisition unit (third data acquisition means), a prediction unit (prediction means), and a structured data generation unit (structured data generation means) when the CPU 111 executes the program 113c.
The encryption unit encrypts molecular structure data A1, B1 relating to the structure of a compound in accordance with an encryption algorithm indicated by encryption algorithm information D1 to generate encrypted molecular structure data A2, B2.
The learning model generation unit performs machine learning based on the encrypted molecular structure data A2 and functional data A3 of the learning target, and/or the encrypted molecular structure data C2 and functional data C3 of the learning target, to generate learning model data D2.
The information providing unit provides the client server 4 with encryption algorithm information D1 for performing encryption according to the encryption algorithm (causing the communication unit 14 to transmit the information).
The first data acquisition unit acquires, from the client server 4 via the communication unit 14, the labeled encrypted molecular structure data C2L of the function prediction target, the encrypted molecular structure data C2 of the learning target, and the functional data C3 of the learning target.
The second data acquisition unit acquires molecular structure data B1 from the reagent DB server 3 via the communication unit 14.
The third data acquisition unit acquires the molecular structure data A1 and the function data A3 from the public DB server 2 via the communication unit 14.
The prediction unit predicts the function of a compound corresponding to the encrypted molecular structure data A2, B2, and the labeled encrypted molecular structure data C2L, the functions of which are to be predicted, based on a learning model represented by learning model data D2.
The structure data generation unit mechanically generates molecular structure data B1 using a genetic algorithm or the like, and stores the data in the storage unit 113.

The operation unit 12 is realized by input devices such as a keyboard and a mouse, and a touch panel that is integral with the display unit 13. The operation unit 12 accepts operation input from these input devices and the touch panel, and outputs an operation signal corresponding to the operation input to the control unit 11.

The display unit 13 is realized by a liquid crystal display device or an organic EL display device, etc., and displays various information under the control of the control unit 11.

Under the control of the control unit 11, the communication unit 14 transmits and receives data between the public DB server 2, the reagent DB server 3 and the client server 4 via the communication network N.

FIG. 5 is a block diagram showing the main functional configuration of the client server 4. As shown in FIG.
The client server 4 includes a control unit 41 , an operation unit 42 , a display unit 43 , a communication unit, etc., and these units are connected to each other via a bus 45 .

The control unit 41 is a processor that controls the overall operation of the client server 4. The control unit 41 has a CPU 411, a RAM 412, and a memory unit 413.

The CPU 411 reads out various control programs 413a and setting data stored in the storage unit 413, stores them in the RAM 412, and executes the programs 413a to perform various arithmetic processing.
The RAM 412 provides a working memory space for the CPU 411 and stores temporary data. The RAM 412 may include a non-volatile memory.

In addition to the above program 413a, the storage unit 413 stores molecular structure data C1, labeled encrypted molecular structure data C2L, encrypted molecular structure data C2, functional data C3, and encryption algorithm information D1. As the storage unit 413, for example, a HDD is used, and a DRAM or the like may also be used in combination.

The encryption algorithm information D1 stored in the memory unit 413 was sent from the MI server 1 and is the same as the encryption algorithm information D1 stored in the memory unit 113 of the MI server 1. By using the encryption algorithm information D1, the control unit 41 of the client server 4 can encrypt the molecular structure data C1 according to the same encryption algorithm as the MI server 1 and generate encrypted molecular structure data C2.

The molecular structure data C1 is data relating to the molecular structure of a compound held by a client, and is managed by the client as confidential information.
As described above, the labeled encrypted molecular structure data C2L is data in which unique labels are associated with multiple pieces of encrypted molecular structure data (see FIG. 4). The encrypted molecular structure data included in the labeled encrypted molecular structure data C2L is generated by the control unit 41 encrypting the molecular structure data C1 according to the encryption algorithm indicated by the encryption algorithm information D1.
The encrypted molecular structure data C2 is data generated by the control unit 41 encrypting the molecular structure data C1 in accordance with the encryption algorithm indicated by the encryption algorithm information D1. The encrypted molecular structure data C2 may include the same encrypted molecular structure data as that included in the labeled encrypted molecular structure data C2L, or may be different from each other.
The function data C3 is data related to the function of the compound corresponding to the molecular structure data C1 (and the encrypted molecular structure data C2). The function data C3 is not considered to be confidential information.

The configurations of the operation unit 42, display unit 43 and communication unit 44 are similar to the configurations of the operation unit 12, display unit 13 and communication unit 14 of the MI server 1, so explanation is omitted.

Next, we will explain how to predict the function of a compound in the compound information processing system 100. There are multiple methods for predicting the function of a compound, depending on what is used as the encrypted molecular structure data for the learning target and the function prediction target.

To generate the learning model data D2 in the MI server 1, at least one of the following two encrypted structure data to be learned is used.
(a1) Encrypted molecular structure data A2 included in general data 113a.
(a2) Encrypted molecular structure data C2 contained in the client-derived data 113b.

The encrypted molecular structure data that is the subject of function prediction in the MI server 1 includes the following three types:
(b1) Labeled encrypted molecular structure data C2L included in the client-derived data 113b.
(b2) Among the encrypted molecular structure data B2 included in the general data 113a, the encrypted molecular structure data B2 is obtained by encrypting the molecular structure data B1 acquired from an external source (for example, the reagent DB server 3).
(b3) Encrypted molecular structure data B2 included in general data 113a, which is obtained by encrypting molecular structure data B1 generated within MI server 1.

The following describes, as examples, the following <First Method> to <Fourth Method>, which have different combinations of encrypted molecular structure data to be learned and encrypted molecular structure data to be predicted for function. In all of the first to fourth methods, the client server 4 can receive the prediction result of the function of a compound by MI without disclosing (transmitting) the molecular structure data C1, which is confidential information, from the client server 4 to an external party.
<First Method>
Encrypted molecular structure data to be learned: (a1)
Encrypted molecular structure data of a function prediction target: (b1)
<Second Method>
Encrypted molecular structure data to be learned: (a1)+(a2)
Encrypted molecular structure data of a function prediction target: (b2)
<Third Method>
Encrypted molecular structure data to be learned: (a1)+(a2)
Encrypted molecular structure data of a function prediction target: (b3)
<Fourth Method>
Encrypted molecular structure data to be learned: (a1)+(a2)
Encrypted molecular structure data of a function prediction target: (b1)

<First Method>
FIG. 6 is a diagram illustrating a first method for predicting the function of a compound.
6 shows the flow of various data processing executed by the MI server 1, the public DB server 2, and the client server 4, and the flow of data transmission and reception between the servers. For simplicity, the operation in which the control unit 11 (41) controls the communication unit 14 (44) to transmit data via the communication unit 14 (44) will be simply referred to as "the control unit 11 (41) transmits data."
In the first method, “(a1) encrypted molecular structure data A2 included in general data 113a” is used as the encrypted molecular structure data to be learned, and “(b1) labeled encrypted molecular structure data C2L included in client-derived data 113b” is used as the encrypted molecular structure data to be predicted in function.

In the first method, the control unit 11 of the MI server 1 first acquires the molecular structure data A1 to be learned and the corresponding functional data A3 from the public DB server 2 (step S101: third data acquisition step).

The control unit 11 of the MI server 1 encrypts the acquired molecular structure data A1 and the molecular structure data A1 pre-stored in the memory unit 113 according to the encryption algorithm indicated by the encryption algorithm information D1 to generate encrypted molecular structure data A2 (step S102: encryption step).

The control unit 11 of the MI server 1 generates learning model data D2 by machine learning based on the generated encrypted molecular structure data A2, the functional data A3 acquired in step S101, and the functional data A3 previously stored in the storage unit 113 (step S103: learning model generation step). If the learning model data D2 has already been stored in the storage unit 113, the control unit 11 updates the learning model data D2 to the newly generated content.
In addition, for machine learning, only one of the encrypted molecular structure data A2 of the molecular structure data A1 obtained from the public DB server 2 and the encrypted molecular structure data A2 of the molecular structure data A1 pre-stored in the memory unit 113 may be used.

Meanwhile, the control unit 11 of the MI server 1 sends encryption algorithm information D1 to the client server 4 (step S104: information provision step).

The control unit 41 of the client server 4 that has received the encryption algorithm information D1 encrypts the molecular structure data C1 of the function prediction target according to the encryption algorithm indicated by the encryption algorithm information D1, and assigns a label to generate labeled encrypted molecular structure data C2L of the function prediction target (step S105). The control unit 41 also transmits the labeled encrypted molecular structure data C2L to the MI server 1. In response, the control unit 11 of the MI server 1 receives the labeled encrypted molecular structure data C2L (step S106: first data acquisition step).

The control unit 11 of the MI server 1 predicts the function of the compound corresponding to each encrypted molecular structure data by applying the learning model represented by the learning model data D2 to each encrypted molecular structure data contained in the acquired labeled encrypted molecular structure data C2L (step S107: prediction step).

The control unit 11 compares the predicted function result with the target value of the function received from the client server 4, and identifies encrypted molecular structure data for which the predicted function result matches the target value (step S108). Here, the predicted function result matching the target value may mean that the value of an index representing the function matches the target value, or that the index is within a predetermined range, or that the index is equal to or greater than a predetermined value, etc.

The control unit 11 transmits the label associated with the encrypted molecular structure data identified in step S108 to the client server 4 (step S109). In response, the control unit 41 of the client server 4 identifies the encrypted molecular structure data corresponding to the received label in the labeled encrypted molecular structure data C2L, and identifies the compound corresponding to the encrypted molecular structure data as a compound exhibiting a desired function.

As described above, the MI server 1 according to this embodiment is equipped with a control unit 11, which in the first method described above provides the client server 4 with learning model data D2 for performing encryption according to a predetermined encryption algorithm (information provision unit), acquires from the client server 4 labeled encrypted molecular structure data C2L of the function prediction target encrypted according to the above encryption algorithm (first data acquisition unit), and predicts the function of the compound corresponding to the labeled encrypted molecular structure data C2L of the function prediction target based on a learning model as a prediction model (prediction unit), and the prediction model represents the correlation between the encrypted structural data obtained by encrypting structural data relating to the structure of the compound according to the above encryption algorithm and the functional data relating to the function of the above compound.
In this way, by receiving the labeled encrypted molecular structure data C2L after encryption and performing function prediction, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, the confidential information of the client is not stored inside the MI server 1, and therefore the security of the confidential information can be improved.
In addition, since a prediction model (here, a learning model) that expresses the correlation between the encrypted molecular structure data and the function data is used, it is sufficient to have the encrypted molecular structure data of the target function prediction in order to predict the function, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data. Therefore, the function of a compound can be predicted by a simple process.

The control unit 11 also encrypts molecular structure data A1 relating to the structure of the compound according to a predetermined encryption algorithm to generate encrypted molecular structure data A2 (encryption unit), and generates a learning model as a predictive model based on the encrypted molecular structure data A2 and function data A3 relating to the function of the compound (learning model generation unit). This allows the learning model to be generated in the MI server 1. Furthermore, since the learning model is generated using the encrypted molecular structure data, in order to predict function using the learning model, it is sufficient to have encrypted molecular structure data of the function prediction target, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data. Therefore, compound function prediction can be performed with simple processing.

Furthermore, the control unit 11 acquires molecular structure data A1 and functional data A3 from the public DB server 2 (third data acquisition unit), generates encrypted molecular structure data A2 based on the acquired molecular structure data A1 (encryption unit), and generates learning model data D2 using at least the encrypted molecular structure data A2 and the functional data A3 acquired from the public DB server 2 (learning model generation unit). This makes it possible to generate a learning model using information on a large number of compounds disclosed by the public DB server 2. This makes it possible to improve the accuracy of prediction of the functions of compounds by the learning model.

Furthermore, the encryption algorithm does not allow reverse conversion back to the structure data before encryption. This means that the MI server 1 cannot decrypt the labeled encrypted molecular structure data C2L received from the client server 4 to identify the molecular structure data C1. Therefore, the client can receive information provision services from MI without disclosing the molecular structure data C1, which is confidential information, to any outsider, including the administrator of the MI server 1.

In addition, in the first method, the program 113c causes the control unit 11, which is a computer provided in the MI server 1, to function as an information providing means for providing the client server 4 with encryption algorithm information D1 for performing encryption according to a predetermined encryption algorithm, a first data acquiring means (data acquiring means) for acquiring, from the client server 4, labeled encrypted molecular structure data C2L of a function prediction target encrypted according to the above encryption algorithm, and a prediction means for predicting the function of a compound corresponding to the labeled encrypted molecular structure data C2L of a function prediction target based on a learning model as a prediction model, and the prediction model represents the correlation between the encrypted structural data obtained by encrypting structural data relating to the structure of the compound according to the above encryption algorithm and functional data relating to the function of the above compound.
By operating the MI server 1 with such a program, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, confidential information of the client is not stored inside the MI server 1, so the security of the confidential information can be improved. Furthermore, for function prediction using a learning model, it is sufficient to have encrypted molecular structure data of the function prediction target, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data, so that function prediction of a compound can be performed with simple processing.

Further, the first method as an information processing method includes an information providing step of providing a client server 4 with encryption algorithm information D1 for performing encryption according to a predetermined encryption algorithm, a first data acquisition step (data acquisition step) of acquiring from the client server 4 labeled encrypted molecular structure data C2L of a function prediction target encrypted according to the above encryption algorithm, and a prediction step of predicting the function of a compound corresponding to the labeled encrypted molecular structure data C2L of the function prediction target based on a learning model as a prediction model, wherein the prediction model represents a correlation between the encrypted structural data obtained by encrypting structural data relating to the structure of the compound according to the above encryption algorithm and functional data relating to the function of the above compound.
According to this method, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, confidential information of the client is not stored inside the MI server 1, so the security of the confidential information can be improved. Furthermore, for function prediction using a learning model, it is sufficient to have encrypted molecular structure data of the function prediction target, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data, so that function prediction of a compound can be performed with simple processing.
Furthermore, some of these methods can be performed manually, which makes it possible to efficiently predict the functions of a variety of compounds without the need for elaborate equipment.

<Second Method>
FIG. 7 is a diagram illustrating a second method for predicting the function of a compound.
FIG. 7 shows the flow of various data processes performed by the MI server 1, the public DB server 2, the reagent DB server 3 and the client server 4, and the flow of data transmission and reception between the servers.
In the second method, "(a1) encrypted molecular structure data A2 included in the general data 113a" and "(a2) encrypted molecular structure data C2 included in the client-derived data 113b" are used as the encrypted molecular structure data to be learned, and "(b2) encrypted molecular structure data B2 included in the general data 113a, which is obtained by encrypting molecular structure data B1 obtained from an outside source (e.g., reagent DB server 3)" is used as the encrypted molecular structure data to be predicted in function.

In the second method, first, the control unit 11 of the MI server 1 acquires the molecular structure data A1 to be learned and the corresponding functional data A3 from the public DB server 2 (step S201: third data acquisition step).

The control unit 11 of the MI server 1 encrypts the acquired molecular structure data A1 and the molecular structure data A1 pre-stored in the memory unit 113 according to the encryption algorithm indicated by the encryption algorithm information D1 to generate encrypted molecular structure data A2 (step S202: encryption step).

The control unit 11 of the MI server 1 sends encryption algorithm information D1 to the client server 4 (step S203: information provision step).

The control unit 41 of the client server 4 that has received the encryption algorithm information D1 encrypts the molecular structure data C1 of the learning target according to the encryption algorithm indicated by the encryption algorithm information D1 to generate encrypted molecular structure data C2 (step S204). The control unit 41 also transmits the encrypted molecular structure data C2 of the learning target and the corresponding functional data C3 of the learning target to the MI server 1. In response, the control unit 11 of the MI server 1 receives the encrypted molecular structure data C2 and functional data C3 of the learning target (step S205: first data acquisition step).

The control unit 11 of the MI server 1 generates learning model data D2 by machine learning based on the encrypted molecular structure data A2 generated in step S202, the functional data A3 acquired from the public DB server 2 in step S201, the functional data A3 pre-stored in the memory unit 113, and the encrypted molecular structure data C2 and functional data C3 acquired from the client server 4 in step S205 (step S206: learning model generation step). If the learning model data D2 has already been stored in the memory unit 113, the control unit 11 updates the learning model data D2 to the newly generated content.
In addition, for machine learning, only a portion of the encrypted molecular structure data A2 of the molecular structure data A1 obtained from the public DB server 2, the encrypted molecular structure data A2 of the molecular structure data A1 pre-stored in the memory unit 113, and the encrypted molecular structure data C2 obtained from the client server 4 may be used.

Meanwhile, the control unit 11 of the MI server 1 acquires molecular structure data B1 of the function prediction target from the reagent DB server 3 (step S207: second data acquisition step), and encrypts it according to the encryption algorithm indicated by the encryption algorithm information D1 to generate encrypted molecular structure data B2 (step S208).

The control unit 11 of the MI server 1 predicts the function of the compound corresponding to each encrypted molecular structure data by applying the learning model represented by the learning model data D2 to the encrypted molecular structure data B2 generated in step S208 (step S209: prediction step).

The control unit 11 compares the function prediction result with the function target value received from the client server 4, and identifies the encrypted molecular structure data B2 that matches the function prediction result image target value (step S210).

The control unit 11 transmits the molecular structure data B1 corresponding to the encrypted molecular structure data B2 identified in step S210 to the client server 4 (step S211). In response, the control unit 41 of the client server 4 identifies the compound related to the received molecular structure data B1 as a compound exhibiting the desired function.

As described above, in the second method, the control unit 11 of the MI server 1 encrypts structural data related to the structure of the compound according to a predetermined encryption algorithm to generate encrypted structural data (encryption unit), generates learning model data D2 representing the correlation between the encrypted structural data and functional data based on the encrypted structural data and functional data related to the function of the compound (learning model generation unit), provides the client server 4 with encryption algorithm information D1 for performing encryption according to the encryption algorithm (information provision unit), acquires from the client server 4 encrypted molecular structure data C2 of the learning subject encrypted according to the above encryption algorithm and functional data C3 of the learning subject related to the function of the compound corresponding to the encrypted molecular structure data C2 (first data acquisition unit), and generates learning model data D2 using at least the acquired encrypted molecular structure data C2 of the learning subject and functional data C3 of the learning subject (learning model generation unit).
In this way, by receiving the encrypted molecular structure data C2 after encryption and performing function prediction, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, the confidential information of the client is not stored inside the MI server 1, and therefore the security of the confidential information can be improved.
In addition, by increasing the security of confidential information related to compound structures in this manner, it becomes easier to collect information related to non-public compound structures (encrypted molecular structure data and functional data of the learning target), making it possible to improve the predictive accuracy of the learning model using information on a larger number of compounds.
Furthermore, a compound exhibiting a function desired by a client is often identified from compounds having a structure similar to that of an existing compound managed and owned by the client. Therefore, as in the second method, a learning model is generated using the encrypted molecular structure data C2 and the function data C3 received from the client server 4, thereby obtaining a learning model that can predict with high accuracy whether or not a compound exhibits the function desired by the client.
In addition, since the learning model is generated using the encrypted molecular structure data, the encrypted molecular structure data of the function prediction target is sufficient for function prediction using the learning model, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data. Therefore, the function of a compound can be predicted by a simple process.
This means:

The control unit 11 also acquires molecular structure data B1 relating to the structure of the compound from the reagent DB server 3 that publishes the structure of the compound (second data acquisition unit), encrypts the acquired molecular structure data B1 to generate encrypted molecular structure data B2 of the function prediction target (encryption unit), and predicts the function of the compound corresponding to the generated encrypted molecular structure data B2 of the function prediction target based on the learning model data D2 (prediction unit). This makes it possible to identify a compound exhibiting the function desired by the client from among the many compounds published by the reagent DB server 3.

In addition, in the second method, the program 113c causes the control unit 11, which is a computer provided in the MI server 1, to function as an encryption means for encrypting structural data related to the structure of a compound according to a predetermined encryption algorithm to generate encrypted structural data, a learning model generation means for generating learning model data D2 representing the correlation between the encrypted structural data and functional data based on the encrypted structural data and functional data related to the function of the compound, an information provision means for providing the client server 4 with encryption algorithm information D1 for performing encryption according to the encryption algorithm, and a first data acquisition means (data acquisition means) for acquiring from the client server 4 the encrypted molecular structure data C2 of the learning subject encrypted according to the above encryption algorithm, and the functional data C3 of the learning subject related to the function of the compound corresponding to the encrypted molecular structure data C2, and the learning model generation means generates the learning model data D2 using at least the encrypted molecular structure data C2 of the learning subject and the functional data C3 of the learning subject acquired by the first data acquisition means.
By operating the MI server 1 with such a program, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, confidential information of the client is not stored inside the MI server 1, so the security of the confidential information can be improved. In addition, by improving the security of the confidential information, it becomes easier to collect information related to the structure of undisclosed compounds, so the prediction accuracy of the learning model can be improved using information on a larger number of compounds. In addition, by generating a learning model using the encrypted molecular structure data C2 and function data C3 received from the client server 4, a learning model that can predict with higher accuracy whether or not a function desired by the client is exhibited can be obtained. In addition, for function prediction using the learning model, it is sufficient to have the encrypted molecular structure data of the function prediction target, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data, so that the function prediction of the compound can be performed with simple processing.

In addition, the second method as an information processing method includes an encryption step of encrypting structural data relating to the structure of the compound according to a predetermined encryption algorithm to generate encrypted structural data, a learning model generation step of generating learning model data D2 representing the correlation between the encrypted structural data and functional data based on the encrypted structural data and functional data relating to the function of the compound, an information provision step of providing a client server 4 with encryption algorithm information D1 for performing encryption according to the encryption algorithm, and a first data acquisition step (data acquisition step) of acquiring from the client server 4 encrypted molecular structure data C2 of the learning subject encrypted according to the above encryption algorithm, and functional data C3 of the learning subject relating to the function of the compound corresponding to the encrypted molecular structure data C2, and in the learning model generation step, the learning model data D2 is generated using at least the encrypted molecular structure data C2 of the learning subject and the functional data C3 of the learning subject acquired in the first data acquisition step.
According to this method, necessary processing can be performed without receiving the molecular structure data C1, which is confidential information, from the client server 4. Therefore, confidential information of the client is not stored inside the MI server 1, so the security of the confidential information can be improved. In addition, by increasing the security of the confidential information, it becomes easier to collect information related to the structure of undisclosed compounds, so that the prediction accuracy of the learning model can be improved using information on a larger number of compounds. In addition, by generating a learning model using the encrypted molecular structure data C2 and function data C3 received from the client server 4, a learning model that can predict with higher accuracy whether or not a function desired by the client is exhibited can be obtained. In addition, for function prediction using the learning model, it is sufficient to have the encrypted molecular structure data of the function prediction target, and there is no need to decrypt the encrypted molecular structure data to generate molecular structure data, so that the function prediction of the compound can be performed by simple processing.
Furthermore, some of these methods can be performed manually, which makes it possible to efficiently predict the functions of a variety of compounds without the need for elaborate equipment.

<Third Method>
FIG. 8 is a diagram illustrating a third method for predicting the function of a compound.
In the third method, "(a1) encrypted molecular structure data A2 included in general data 113a" and "(a2) encrypted molecular structure data C2 included in client-derived data 113b" are used as the encrypted molecular structure data to be learned, and "(b3) encrypted molecular structure data B2 included in general data 113a, which is obtained by encrypting molecular structure data B1 generated inside MI server 1" is used as the encrypted molecular structure data to be predicted in function.

Steps S301 to S306 in the third method are similar to steps S201 to S206 in the second method, so explanation is omitted.

The control unit 11 of the MI server 1 mechanically and randomly generates a plurality of molecular structure data B1 using a genetic algorithm or the like (step S307), and encrypts the data according to the encryption algorithm indicated by the encryption algorithm information D1 to generate a plurality of encrypted molecular structure data B2 (step S308). In step S309 (prediction step), function prediction is performed using the encrypted molecular structure data B2 generated in step S308 as encrypted molecular structure data to be predicted.
Steps S310 and S311 are similar to steps S210 and S211 in the second method, and therefore a description thereof will be omitted.

As described above, in the third method, the control unit 11 of the MI server 1 generates molecular structure data B1 (structure data generation unit), encrypts the generated molecular structure data B1 to generate encrypted molecular structure data B2 of the target for function prediction (encryption unit), and predicts the function of the compound corresponding to the generated encrypted molecular structure data B2 of the target for function prediction based on the learning model data D2 (prediction unit). This increases the possibility that the client will be able to identify a compound exhibiting the desired function even in cases where sufficient data on the compound of the target for function prediction cannot be obtained from outside the MI server 1.

<Fourth Method>
FIG. 9 is a diagram illustrating a fourth method for predicting the function of a compound.
In the fourth method, “(a1) encrypted molecular structure data A2 included in general data 113a” and “(a2) encrypted molecular structure data C2 included in client-derived data 113b” are used as the encrypted molecular structure data to be learned, and “(b1) labeled encrypted molecular structure data C2L included in client-derived data 113b” is used as the encrypted molecular structure data to be predicted in function.

Steps S401 to S403 and S406 of the fourth method are similar to steps S301 to S303 and S306 of the third method, so explanation is omitted.

The control unit 41 of the client server 4 that has received the encryption algorithm information D1 in step S403 encrypts the molecular structure data C1 according to the encryption algorithm indicated by the encryption algorithm information D1 to generate the encrypted molecular structure data C2 of the learning target and the labeled encrypted molecular structure data C2L of the function prediction target (step S404). The control unit 41 also transmits the encrypted molecular structure data C2 of the learning target and the corresponding functional data C3 of the learning target to the MI server 1, and the control unit 11 of the MI server 1 receives the encrypted molecular structure data C2 and the functional data C3 of the learning target (step S405: first data acquisition step). The control unit 41 also transmits the labeled encrypted molecular structure data C2L of the function prediction target to the MI server 1, and the control unit 11 of the MI server 1 receives the labeled encrypted molecular structure data C2L (step S407: first data acquisition step).

The control unit 11 of the MI server 1 predicts the function of the compound corresponding to each encrypted molecular structure data by applying the learning model represented by the learning model data D2 to each encrypted molecular structure data included in the labeled encrypted molecular structure data C2L acquired in step S407 (step S408: prediction step).
The subsequent steps S409 and S410 are similar to steps S108 and S109 in the first method, and therefore a description thereof will be omitted.

As described above, in the fourth method, the control unit 11 of the MI server 1 acquires the encrypted molecular structure data C2 and the function data C3 of the learning target from the client server 4 (first data acquisition unit), generates learning model data D2 using at least the acquired encrypted molecular structure data C2 and function data C3 (learning model generation unit), acquires labeled encrypted molecular structure data C2L of the function prediction target encrypted according to the encryption algorithm from the client server 4 (first data acquisition unit), and predicts the function of the compound corresponding to the acquired labeled encrypted molecular structure data C2L of the function prediction target based on the learning model data D2 (prediction unit). This makes it possible to generate a learning model that can predict with high accuracy whether or not a compound exhibits the function desired by the client, and to identify a compound exhibiting the function desired by the client from the encrypted molecular structure data C2 provided by the client.

The present invention is not limited to the above-described embodiment and each of the modified examples, and various modifications are possible.
For example, in the above embodiment, examples of encrypted molecular structure data for which function prediction is to be performed include encrypted molecular structure data C2 (first method) obtained from the client server 4, encrypted molecular structure data B2 (second method) of molecular structure data B1 obtained from the reagent DB server 3, and encrypted molecular structure data B2 (third method) of molecular structure data B1 generated within the MI server 1, but this is not intended to be limiting. Any encrypted molecular structure data obtained by encrypting molecular structure data of a compound can be used as the encrypted molecular structure data for which function prediction is to be performed, and the route of acquisition is not limited to that exemplified in this embodiment.
As one example, in the second or third method, the encrypted molecular structure data C2 of the function prediction target may be acquired from the client server 4.

In the above embodiment, an example of using an irreversible encryption algorithm has been described, but the present invention is not limited to this, and a reversible encryption algorithm may be used. Even in this case, even if the encrypted molecular structure data C2 in the MI server 1 is illegally accessed from outside, the third party who illegally accessed the data cannot identify the encryption algorithm and therefore cannot decrypt the encrypted molecular structure data C2 to obtain the molecular structure data C1. Therefore, even when a reversible encryption algorithm is used, the effect of increasing the security of the client's confidential information (molecular structure data C1) can be obtained.
Furthermore, the encryption algorithm is not limited to one that uses a hash function.

In addition, in the above embodiment, the learning model was generated in MI server 1, but this is not limited to the above, and an existing learning model (e.g., a learning model generated in an external device) may be used as is. In this case, MI server 1 may not have a learning model generation function (learning model generation unit). In addition, in this aspect, when encrypted molecular structure data of the target for function prediction is obtained from outside MI server 1, MI server 1 may not have a function of encrypting the molecular structure data (encryption unit).

In addition, in the above embodiment, an example of using a learning model generated by machine learning using an inductive approach has been described, but the prediction model used to predict the function of a compound is not limited to this learning model. As a prediction model, for example, a deductive prediction model that predicts the function of a compound from known principles and regularities about the compound may be used. Even when a deductive prediction model is used, the MI server 1 does not need to have a function of generating a learning model (learning model generation unit). Furthermore, when encrypted molecular structure data of a function prediction target is obtained from outside the MI server 1, the MI server 1 does not need to have a function of encrypting the molecular structure data (encryption unit).

In addition, in the above embodiment, the MI server 1, the public DB server 2, the reagent DB server 3, and the client server 4 are each described as being composed of a single server device, but this is not limiting, and any of these server devices may be replaced with a system composed of multiple devices. For example, at least a portion of the programs and data stored in the memory unit 113 of the MI server 1 may be stored in a memory device external to the MI server 1.

Although several embodiments of the present invention have been described, the scope of the present invention is not limited to the above-described embodiments, but includes the scope of the invention described in the claims and their equivalents.

The present invention can be used in information processing devices, programs and information processing methods.

1. MI server (information processing device)
11 Control unit (encryption unit, learning model generation unit, information provision unit, first to third data acquisition units, prediction unit, structured data generation unit)
111 CPU
112 RAM
113 Storage unit 113a General data 113b Client-derived data 113c Program 12 Operation unit 13 Display unit 14 Communication unit 15 Bus 2 Public DB server (database)
3. Reagent DB server (second external device)
4. Client Server (first external device)
41 Control unit 411 CPU
412 RAM
413 Storage unit 413a Program 42 Operation unit 43 Display unit 44 Communication unit 45 Bus 100 Compound information processing system A1, B1, C1 Molecular structure data (structure data)
A2, B2, C2 Encrypted molecular structure data (encrypted structure data)
C2L Labeled encrypted molecular structure data (encrypted structure data)
C3 Function data D1 Encryption algorithm information D2 Learning model data N Communication network

Claims (13)

an information providing unit that provides the first external device with encryption algorithm information for performing encryption according to a predetermined encryption algorithm;
a first data acquisition unit that acquires encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction unit that predicts a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Equipped with
The prediction model represents a correlation between encrypted structure data obtained by encrypting structure data relating to a structure of a compound according to the encryption algorithm and function data relating to a function of the compound. an encryption unit that encrypts structure data relating to a compound structure according to the encryption algorithm to generate encrypted structure data;
a learning model generation unit that generates a learning model as the prediction model based on the encrypted structure data and function data related to the function of the compound;
The information processing device according to claim 1 . the first data acquisition unit acquires, from the first external device, encrypted structural data of a learning object encrypted according to the encryption algorithm, and functional data of a learning object related to a function of a compound corresponding to the encrypted structural data of the learning object;
The information processing device according to claim 2 , wherein the learning model generation unit generates the learning model using at least the encrypted structure data of the learning object and the functional data of the learning object acquired by the first data acquisition unit. an encryption unit that encrypts structure data relating to a compound structure according to a predetermined encryption algorithm to generate encrypted structure data;
a learning model generating unit that generates a learning model that represents a correlation between the encrypted structure data and the functional data, based on the encrypted structure data and functional data related to a function of the compound;
an information providing unit that provides a first external device with encryption algorithm information for performing encryption according to the encryption algorithm;
a first data acquisition unit that acquires, from the first external device, encrypted structure data of a learning object encrypted according to the encryption algorithm, and function data of a learning object related to a function of a compound corresponding to the encrypted structure data;
Equipped with
An information processing device, wherein the learning model generation unit generates the learning model using at least the encrypted structure data of the learning target and the functional data of the learning target acquired by the first data acquisition unit. The first data acquisition unit acquires encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device,
The information processing device according to claim 4 , further comprising a prediction unit configured to predict, based on the learning model, a function of a compound corresponding to the encrypted structure data of the function prediction target acquired by the first data acquisition unit. a structure data generating unit that generates the structure data,
The encryption unit encrypts the structure data generated by the structure data generation unit to generate the encrypted structure data of a function prediction target;
The information processing device according to claim 2 , wherein the prediction unit predicts a function of a compound corresponding to the encrypted structure data of the function prediction target generated by the encryption unit based on the learning model. a second data acquisition unit that acquires structural data relating to a structure of the compound from a second external device that publishes the structure of the compound;
The encryption unit encrypts the structure data acquired by the second data acquisition unit to generate the encrypted structure data of a function prediction target;
The information processing device according to claim 2 , wherein the prediction unit predicts a function of a compound corresponding to the encrypted structure data of the function prediction target generated by the encryption unit based on the learning model. a third data acquisition unit that acquires the structure data and the function data from an external predetermined database;
The encryption unit generates the encrypted structure data based on the structure data acquired by the third data acquisition unit,
The information processing device according to any one of claims 2 to 7, wherein the learning model generation unit generates the learning model using at least the encrypted structure data and the functional data acquired by the third data acquisition unit. The information processing device according to any one of claims 1 to 8, wherein the encryption algorithm is incapable of reverse conversion to the structure data before encryption. A computer installed in the information processing device,
an information providing means for providing the first external device with encryption algorithm information for performing encryption according to a predetermined encryption algorithm;
a data acquisition means for acquiring encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction means for predicting a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Function as a
The prediction model is a program that represents the correlation between encrypted structural data obtained by encrypting structural data relating to a structure of a compound according to the encryption algorithm, and functional data relating to a function of the compound. A computer installed in the information processing device,
an encryption means for encrypting structural data relating to a compound structure according to a predetermined encryption algorithm to generate encrypted structural data;
a learning model generating means for generating a learning model representing a correlation between the encrypted structure data and functional data relating to a function of the compound, based on the encrypted structure data and functional data relating to a function of the compound;
an information providing means for providing a first external device with encryption algorithm information for performing encryption according to the encryption algorithm;
a data acquisition means for acquiring, from the first external device, encrypted structure data of a learning subject encrypted according to the encryption algorithm, and function data of a learning subject relating to a function of a compound corresponding to the encrypted structure data;
Function as a
The learning model generation means is a program that generates the learning model using at least the encrypted structure data of the learning subject and the functional data of the learning subject acquired by the data acquisition means. An information processing method executed by an information processing device,
an information providing step of providing encryption algorithm information for performing encryption according to a predetermined encryption algorithm to a first external device;
a data acquisition step of acquiring encrypted structure data of a function prediction target encrypted according to the encryption algorithm from the first external device;
a prediction step of predicting a function of a compound corresponding to the encrypted structure data of the function prediction target based on a predetermined prediction model;
Including,
An information processing method, wherein the prediction model represents a correlation between encrypted structural data obtained by encrypting structural data relating to a structure of a compound according to the encryption algorithm, and functional data relating to a function of the compound. An information processing method executed by an information processing device,
an encryption step of encrypting structural data relating to the structure of the compound according to a predetermined encryption algorithm to generate encrypted structural data;
a learning model generating step of generating a learning model representing a correlation between the encrypted structure data and the functional data, based on the encrypted structure data and functional data relating to a function of the compound;
an information providing step of providing encryption algorithm information for performing encryption according to the encryption algorithm to a first external device;
a data acquisition step of acquiring, from the first external device, encrypted structure data of a learning subject encrypted according to the encryption algorithm, and function data of a learning subject relating to a function of a compound corresponding to the encrypted structure data;
Including,
An information processing method, in which the learning model generation step generates the learning model using at least the encrypted structure data of the learning subject and the functional data of the learning subject acquired in the data acquisition step.

Images