JP2002073626A - Pharmocophor analysis method using decision tree - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a pharmocophor analysis method which can drastically improve prediction accuracy in a virtual screening technology. SOLUTION: A minimum unit 'N-center pharmocophor component' that represents the three-dimensional spatial arrangement information of combination factors and 'a structural descriptor' that represents the physico-chemical characteristic, etc., of a compound used generally in a quantitative structure activity correlation analysis are used for an explanatory variable, and a 'decision tree analysis' being one of a nonparametric and nonlinear statistical analysis technique is performed, thereby achieving the objective method.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a new pharmacophore analysis method for virtual screening performed on a computer using a large amount of pharmacological activity data obtained from high-throughput screening. More specifically, “decision tree” is used as an explanatory variable with “N-centered pharmacophore constituents” and “various structural descriptors” expressing the physicochemical properties of compounds generally used in quantitative structure-activity relationship analysis. The present invention relates to a statistical model that can identify the presence or absence of pharmacological activity with high accuracy by performing “analysis”.

[0002]

2. Description of the Related Art Recent advances in computer chemistry have created a virtual screening technique for predicting, on a computer, the presence or absence of pharmacological activity of a virtual library of easily obtainable or synthesizable compounds. This technology is expected to become a useful and indispensable technology for improving the efficiency and success probability of a series of drug discovery screening studies. As one of the conventional virtual screening techniques, a docking study with a three-dimensional structural model of an in-vivo macromolecule such as a target enzyme or receptor is known. Although this method achieves relatively high prediction accuracy, not only does it take a great deal of time for calculations, but also the three-dimensional structure of the target macromolecule in vivo is often unknown.

When the three-dimensional structure of a macromolecule in a living body is unknown, a pharmacophore model is created from the structure of the compound and known pharmacological activity data, and the degree of conformity to the model is evaluated. Is performed. For this pharmacophore analysis technique, a method of quantitatively analyzing pharmacological activity by multiple regression analysis using a structural descriptor representing the physicochemical properties of a compound as an explanatory variable has been known for a long time. Although this method is useful for analyzing the correlation between local partial structures and pharmacological activities, it is difficult to use it for global analysis. As another pharmacophore analysis technique, a technique for analyzing the spatial arrangement of factors contributing to intermolecular binding, which is shared by a plurality of compounds having the same pharmacological action, is known. Using a model obtained from this analysis, it is possible to find a compound having a specific pharmacological activity from a large number of compounds. However, this method was not appropriate for the correlation analysis between the local partial structure and the pharmacological activity, as opposed to the quantitative structure-activity relationship analysis.

The three-dimensional quantitative structure-activity relationship analysis method, which combines the above two methods in a stepwise manner, compensates for the disadvantages of both, and analyzes the correlation between the global and local structure of the compound and the pharmacological activity. Has been realized. However, it is difficult to say that sufficient prediction accuracy has been achieved because the commonality of spatial arrangement of factors contributing to binding and physicochemical properties are separately evaluated. In addition, the expression ability of the pharmacophore is poor,
As in the results obtained from high-throughput screening, it is not sufficient to analyze the correlation between diverse structural groups and pharmacological activities. In addition, due to a large amount of calculation and complicated operation, the data amount that can be handled is limited to several to several tens of compounds. As described above, the main analysis methods known so far include the quantitative structure-activity relationship method and the overlay analysis method, and although these have achieved certain results,
It was hard to say that the accuracy required for virtual screening was sufficiently satisfied. For example, it is a sufficient analysis method because it has several issues to be improved, such as improvement of prediction accuracy, improvement of model expression ability, faster calculation processing, and more efficient data management. Did not.

[0005]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a pharmacophore analysis technique which realizes a great improvement in prediction accuracy in virtual screening technology. In addition, this method is a method that exhibits high predictability with versatility in the target pharmacological activity, and is a method that realizes a fast and simple data management operation.

[0006]

The present inventor has conducted intensive studies on a pharmacophore analysis technique, which plays the most important role in the success or failure of virtual screening, and is one of the non-linear and non-parametric statistical analysis techniques. Using "decision tree analysis", we obtained the knowledge that pharmacophore analysis including both quantitative structure-activity relationship method and overlay analysis method can be realized. Specifically, the present inventor has proposed the “N-centered pharmacophore component” expressing the three-dimensional spatial arrangement information of a binding factor and the physicochemical properties of a compound generally used in a quantitative structure-activity relationship method. By performing "decision tree analysis" using the expressed "structural descriptor" as an explanatory variable, it has been found that a statistical model capable of identifying the presence or absence of pharmacological activity with high accuracy can be obtained, and the present invention has been completed.

That is, the gist of the present invention is as follows. (1) A method for selecting a lead candidate compound having pharmacological activity at high speed and high prediction accuracy from a database containing structural information of a compound, wherein “N center” is used to express three-dimensional spatial arrangement information of binding factors. “Pharmacophore constituents” and “structural descriptors” that represent the physicochemical properties of compounds commonly used in quantitative structure-activity relationship analysis
A pharmacophore analysis by performing "decision tree analysis", which is one of the non-parametric and non-linear statistical analysis methods, using as an explanatory variable. (2) CART (Classification a)
(1) The method according to the above (1), which comprises employing an (nd Regression Tree) method. (3) The description in (1) above, wherein the “N-centered pharmacophore component” is a “two-centered pharmacophore component” that is a minimum unit expressing three-dimensional spatial arrangement information of a binding factor. the method of. (4) As the “structural descriptor”, one reflecting the contribution of entropy in intermolecular bonding such as octanol-water partition coefficient, change in free energy due to change in hydration state, conformational freedom of ligand, or molecular weight, The method according to the above (1), which employs a shape and size reflecting the volume and surface area. (5) A method of automatically creating and updating structure-activity association rules from a database containing compound structure information and pharmacological activity information using a computer, and expressing three-dimensional spatial arrangement information of binding factors. Non-parametric and non-linear non-linear using the "N-centered pharmacophore component" and the "structural descriptor" that expresses the physicochemical properties of compounds commonly used in quantitative structure-activity relationship analysis as explanatory variables A method characterized by performing pharmacophore analysis by performing one of the statistical analysis methods "decision tree analysis". (6) A database of compound structure-activity relationship rules obtained by the method for creating structure-activity relationship rules described in (5) above. (7) A method of using a computer to search for a lead candidate compound having a specific pharmacological activity by registering the structural formula of a specific compound and pharmacological activity data in the database of the structure-activity relationship rule described in (5) above . (8) The database of the structure-activity relationship rules described in the above (5) is given a specific pharmacological activity by giving either a SQL-based search formula running on a computer or specific pharmacological activity information of the compound. A method for searching for a lead candidate compound having activity using a computer.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS "Decision tree analysis" used in the present invention is one of the non-linear and non-parametric statistical analysis methods, and performs optimal record classification according to the class of the objective variable. The statistical model is represented as a phylogenetic tree.
The decision tree analysis method can handle a large number of records and explanatory variables because the calculation is performed at high speed. “CART”, “C5.0 (ID3)”, “CHA”, depending on the branching method at the tree growth stage and the pruning method used to ensure robustness
Many methods such as “ID” have been proposed (Atsushi Otaki / Yoji Horie / Dan Steinberg, Nikkagiren, 1998).

The CART (Classificati) used in the present invention
on and Regression Tree method ”, written by Atsushi Otaki / Yuji Horie / Dan Steinberg, Nikkagiren, 19
98), wherein tree growth is limited to two branches,
It is characterized by active pruning and is a decision tree analysis method that emphasizes the robustness of statistical models. The CART analysis is executed by creating a program on its own or by using commercially available application software such as a decision tree analysis module “Answer Tree 2.1J” (SPSS) of the statistical analysis package software “SPSS 10.0J”.

The “pharmacophore analysis” of the present invention is
The purpose of the present invention is to clarify the mechanism of expressing a specific pharmacological activity, and the findings of pharmacophore obtained from the analysis can predict the presence or absence of a pharmacological activity for a compound whose pharmacological activity is unknown.

The "N-centered pharmacophore component" of the present invention represents relative arrangement information in a three-dimensional space of a binding factor having an important role in binding a polymer to a ligand, and is generally known. (Erin K. Bradley et al., J. Med. Chem. 2000, 4).
3, 2770-2774). Binding agents include hydrogen bond acceptors, hydrogen bond donors, hydrophobic centers, and the like. The relative arrangement information is represented as an arbitrary defined discrete value. Therefore,
The "N-centered pharmacophore component" is composed of N types of binding factors and discrete values of their relative distances. Similarly, the "two-centered pharmacophore component" is the minimum unit of the "N-centered pharmacophore component" and is composed of two types of binding factors and discrete values of their relative distances. As the binding factors, hydrogen bond acceptor, hydrogen bond donor, hydrophobic center, aromatic ring center, positive charge ability, negative charge ability are used, and by defining eight discrete values of relative distance, 180 kinds of `` A two-center pharmacophore component is defined.

The "structure descriptor" of the present invention is generally used in a quantitative structure-activity relationship technique. For example, a water-octanol partition coefficient reflecting a thermodynamic entropy change accompanying the binding of a ligand is used. And the number of free rotation bonds, and the molecular weight and molecular refractive index, etc., which represent the shape and size of the molecule.

In the data preparation of the present invention, first, a stable and metastable conformation is generated from the structural formula information of a compound. The conformational generation can be performed by creating an original program based on generally known methods, or by using Catalyst ver.4.5 (MS
It is executed using commercially available application software such as I Company. The data creation of "two-center pharmacophore component" can be easily obtained from the coordinate data of the conformation.
It is executed using commercially available application software such as st ver.4.5 (MSI). “Structural descriptors” can be easily obtained from the coordinate data of the structural formula or conformation of a compound.
It is executed using commercially available application software such as Cerius2 ver.4.0 (MSI). Since these data are used in common for the analysis of pharmacological activity, they need not be recalculated because they are held inside the compound database or linked thereto.

Next, Table 1 shows the setting items of the CART method.

[0015]

[Table 1] These setting conditions are suitable for various pharmacophore analysis and are considered to be versatile conditions.
Used in all of the following examples. The procedure of the pharmacophore analysis method using the decision tree of the present invention is shown in the flowchart of FIG.

[0016]

The present invention will be described below in detail with reference to examples.
However, the present invention is not limited by these Examples.
Not. Example 1Analysis and validation of pharmacological activity of 5-HT1a antagonist  Registered in the marketed drug database MDDR (MDL) 5
-Randomized from 429 compounds of HT1a antagonist to 100
Compounds were selected, of which 75 were used for analysis and the remaining 25
Compounds were used for validation. On the other hand, 5-HT1a antagonist
As an inactive compound, for convenience, for commercial screening
10,000 compounds were randomly selected from the compounds,
500 compounds for analysis, remaining 2500 compounds for verification
Was. The "prior probability parameter" is defined as active and inactive
The settings were made equal in both cases. "Cost parameter
Was not used. As a result of analysis by CART,
The result was as follows. The prediction rate of the active group is high, and the inactive group
5-HT1a antagonist activity due to low misclassification rate
Could be identified with very high accuracy. In addition, analysis data
Data and verification data have little difference in prediction rate and misclassification rate
And a robust robust statistical model was obtained. Table 2 shows the results
Shown in

[0017]

[Table 2]

Embodiment 2Analysis of pharmacological activity of SSRI (5-HT Reuptake Inhibitor) and
Verification  S registered in the over-the-counter drug database MDDR (MDL)
100 compounds were randomly selected from 393 SRI compounds, and
Of which 75 for analysis and the remaining 25 for validation
Using. On the other hand, as an SSRI-inactive compound,
10,000 random compounds from sales screening compounds
And select 7500 compounds for analysis and the remaining 2500
Compounds were used for validation. The "prior probability parameter"
Set equal for both active and inactive cases
Was. No "cost parameters" were used. By CART
As a result of the analysis, the results are as shown in the table below. Conclusion of Example 1
As with the fruits, the prediction rate of the active group was high and the inactive group was misclassified.
Because of the low rate, the presence or absence of SSRI activity can be determined with extremely high accuracy.
Could be identified. In addition, the prediction rate of analysis data and verification data and
Robust statistical model with little difference in
Dell got. Table 3 shows the results.

[0019]

[Table 3]

Embodiment 3Inhibition activity and verification of COX-1 (Cyclooxygenase-1) inhibitor  C registered in the marketed drug database MDDR (MDL)
Random selection of 100 compounds from 688 compounds of OX-1 inhibitor
Of which 75 were analyzed for analysis and the remaining 25 were tested.
Used for proof. On the other hand, COX-1 inhibitory inactive compounds
For convenience, random from commercial screening compounds
10,000 compounds were selected, and 7,500 of them were used for analysis.
Then, the remaining 2500 compounds were used for verification. `` A priori probability
Parameters are equal in both active and inactive cases.
It was set so that. Do not use "cost parameters"
Was. As a result of analysis by CART, the results are as shown in the table below.
Was. Similar to the results of Examples 1 and 2, the prediction rate of the active group
Is high and the rate of misclassification in the inactive group is low,
The presence or absence of harmful activity could be identified with very high accuracy. Table of results
It is shown in FIG.

[0021]

[Table 4]

[0022]

According to the present invention, the minimum unit "N-centered pharmacophore component" expressing the three-dimensional spatial arrangement information of a binding factor and the physicochemical properties of compounds generally used for quantitative structure-activity relationship analysis Pharmacophore analysis with high speed and high prediction accuracy by using one of the nonparametric and non-linear statistical analysis methods, "decision tree analysis", using "structural descriptors" that express statistical characteristics as explanatory variables. Clarified that it can be realized. That is, among several decision tree analysis methods, the CART (Classification and Reg
The statistical analysis using the "response tree" method and the "2-center pharmacophore component" and the limited "structural descriptor" as explanatory variables showed that pharmacological activity could be predicted with high accuracy. As described above, according to the analysis method of the present invention, the prediction accuracy can be significantly improved in the virtual screening technology. Further, not only the prediction of pharmacological activity but also the prediction of physical properties, kinetics, toxicity, and the like can improve the accuracy. Furthermore, this method can also be used as a knowledge discovery tool, and can provide highly reliable knowledge on the direction of research. Furthermore, the method of the present invention realizes versatility for pharmacological action while maintaining high prediction accuracy equal to or higher than that of the prior art. Realizes high-speed calculation of descriptors and statistical analysis. be able to. These features of the method of the present invention can be summarized as follows. A pharmacophore statistical analysis model can be obtained from a large amount of pharmacological activity data obtained from high-throughput screening. Virtual screening can be performed to find compounds that exhibit a particular pharmacological activity. The above services can be provided by using this method. To construct a "compound database system" that allows non-statistical experts such as synthetic chemists to search for the presence or absence of a specific pharmacological activity simply by performing ordinary registration work such as structural formulas and pharmacological activity values of compounds. Can be. The above-described system that operates at high speed can be constructed in both a client-server type and a stand-alone type that operates on a personal computer. The above-described system incorporated in an existing compound database system such as ISIS (MDL) or Accord (MSI) can be constructed. Register the compounds that can be supplied to the above system and
It is possible to realize an environment where the information is published on W and a general user can search using the index of the presence or absence of a specific pharmacological activity as an index, and order a hit compound as it is. Furthermore, by giving either an SQL-format search formula provided by the above system running on a computer owned by a general user or a pharmacological activity information of a compound registered in a public database, Search can be performed. By registering the compound data and pharmacological activity data accumulated in the company in the above-mentioned system, it becomes possible to automatically create and update a database of structure-activity association rules. This rule database can be applied to search for new drug targets, predict side effects, analyze action mechanisms, and the like.

[Brief description of the drawings]

FIG. 1 is a flowchart of an analysis method according to the present invention. That is, the structural formula or the pharmacological activity data is input to the database including the structural formula of the compound, the pharmacological activity data and the explanatory variables used in the analysis as needed (S2).
When a structural formula is input, a conformation that can actually exist is automatically generated (S4), and explanatory variables required for analysis are calculated (S6), and are stored in the database. In the database, a decision tree analysis is performed for each pharmacological activity item, and the analysis result is stored in an SQL format (S10). This calculation process is executed periodically or manually, and the data is updated. The user specifies the pharmacological activity item and, based on the relevant SQL,
A lead candidate compound is selected from the compounds registered in the database (S12).

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06F 17/50 638 G06F 17/50 638

Claims (8)

[Claims] Claims: 1. In a database containing structural information of a compound, a lead candidate compound having pharmacological activity is identified as
This is a fast and highly accurate prediction method that uses the “N-centered pharmacophore component” to represent the three-dimensional spatial arrangement information of binding factors and the physical chemistry of compounds commonly used in quantitative structure-activity relationship analysis. Pharmacophore analysis by performing "decision tree analysis", one of the non-parametric and non-linear statistical analysis methods, using "structural descriptors" that express statistical characteristics as explanatory variables. Method. 2. As a decision tree analysis method, CART (Classifi
2. The method of claim 1, comprising employing a cation and regression tree) method. 3. The method according to claim 1, wherein the “N-centered pharmacophore component” is a “two-centered pharmacophore component” which is a minimum unit expressing three-dimensional spatial arrangement information of a binding factor. The described method. 4. Octanol-
Reflects the contribution of entropy in intermolecular bonding, such as water partition coefficient, free energy change due to changes in hydration state, and conformational freedom of ligand, or reflects shape and size of molecular weight, volume, surface area, etc. 2. The method of claim 1, comprising employing one. 5. A method for automatically creating and updating a structure-activity association rule from a database containing compound structure information and pharmacological activity information by using a computer, comprising: information on three-dimensional spatial arrangement of binding factors. The non-parametric and non-parametric non-structural variables are expressed using the “N-centered pharmacophore component” that expresses the phenomena and the “structure descriptor” that expresses the physicochemical properties of compounds commonly used in quantitative structure-activity relationship analysis. One of the linear statistical analysis methods "decision tree analysis"
Performing a pharmacophore analysis. 6. A database of compound structure-activity relationship rules obtained by the method for creating a structure-activity relationship rule according to claim 5. 7. A method for searching for a compound having a specific pharmacological activity using a computer by registering the structural formula and pharmacological activity data of a specific compound in the database of the structure-activity relationship rule according to claim 5. 8. A database for a structure-activity relationship rule according to claim 5, which is provided with a search formula in the form of SQL running on a computer or a specific pharmacological activity information of a compound, thereby giving a specific pharmacological activity information. A method for searching for a lead candidate compound having pharmacological activity using a computer.