WO2022094873A1

WO2022094873A1 - Molecular force field quality control system and control method therefor

Info

Publication number: WO2022094873A1
Application number: PCT/CN2020/126782
Authority: WO
Inventors: 王果; 马健; 张佩宇; 方栋; 温书豪; 赖力鹏
Original assignee: 深圳晶泰科技有限公司
Priority date: 2020-11-05
Filing date: 2020-11-05
Publication date: 2022-05-12

Abstract

A molecular force field quality control system and a control method therefor. The molecular force field quality control system comprises: a data acquisition module for reading quantum mechanics optimization data for benchmarking and molecular mechanics optimization data of a force field to be tested, and consolidating said data into a data table and a molecular coordinate file that comprise specified items; an energy analysis module for calculating evaluation indexes in an index set that are related to molecular conformation energy and a potential energy surface, calculating an energy analysis index by means of parameters according to two test methods, and generating an original data table; a structural analysis module for calculating an evaluation index in the index set that is related to a molecular structure, calculating a structural analysis index, and generating an original data table; and a data statistics module for calculating a statistical index, printing statistical data, and drawing a statistical chart. By incorporating force field parameter items into a quality control process, a force field developer can accurately identify underperforming parameter items in a test set, and targetedly optimize parameters.

Description

Molecular force field quality control system and its control method

technical field

The invention belongs to the field of molecular mechanics, and in particular relates to a molecular force field quality control system and a control method thereof. As a method for testing the performance of intramolecular parameters in a force field, the invention is suitable for evaluating parameters in a molecular force field and for evaluating poor performance in a molecular force field. parameters for fitting correction.

Background technique

Molecular mechanics has a wide range of applications in various fields such as drug design and materials science due to its low computational cost and high computational accuracy. Molecular mechanics uses specific functions to calculate molecular properties (such as structure and energy), and the functional form and parameters in the function are called molecular force fields. The commonly used force field energy can be decomposed into two parts: bonding (intramolecular) and non-bonding (intermolecular), of which the bonding part describes the vibration and rotation of the chemical bond in the molecule, including bond length, bond angle, dihedral angle, surface The outer bending parameter, the non-bonding part, describes the attraction or repulsion of atoms with far distances between or within the molecule due to electrostatic and van der Waals forces, including the charge model and van der Waals parameters.

Since the molecular force field model itself contains approximation processing, and the parameters are derived from the fitting of quantum chemical calculation results of a large number of molecules or empirical laws, the coverage of the training set affects the scope of application of the parameters, and the similarity within the training set affects the accuracy of the parameters , so the generality and accuracy of the force field need to be balanced, which results in differences in the accuracy of most force fields for target molecules. In force field development and application, it is necessary to evaluate and validate specific test sets to obtain quality control data.

In the current literature reports on open source or commercial force fields, the evaluation indicators used are not uniform, the indicators are not clearly defined, the test sets are very different, the calculation process is usually coupled with the data analysis process, and there is a large amount of data that is not disclosed, lacking a complete evaluation method. and indicator system. Researchers and commercial users generally do not have the conditions to perform quantitative method verification on the selected force field and obtain data support under the specific usage scenario. A complete and independent quality control process helps to select and optimize force field parameters in different scenarios, and obtain more accurate molecular mechanics calculation results.

SUMMARY OF THE INVENTION

In view of the above technical problems, the present invention provides a molecular force field quality control system and a control method thereof. The molecular conformational energy and structure optimized by a high-precision quantum chemical method are used as standards to test and evaluate the molecular mechanics calculation results of different force fields. Energy and structural deviations between standards.

The specific technical solutions are:

Molecular force field quality control system, including four modules: data acquisition module, energy analysis module, structure analysis module and data statistics module;

In the data acquisition module, according to the molecular source of the test set, the source of the force field to be measured and user preference, the data acquisition selects corresponding computational simulation tools to generate the quantum mechanical optimization data for the target and the molecular mechanics optimization data of the force field to be measured, And organize it into a data table and molecular coordinate file containing the specified column, as the input data of the energy analysis module;

The energy analysis module calculates the evaluation indexes related to the molecular conformational energy and potential energy surface of the index set; the program reads the molecular name, the dihedral atomic number, angle and conformational energy provided by the user to limit optimization, and according to two test methods Parameter calculation energy analysis index, generating original data table;

The structure analysis module calculates the evaluation index related to the molecular structure of the index set; the program reads the molecular name, the dihedral atomic number and 3D coordinates provided by the user to limit optimization, calculates the structure analysis index, and generates a raw data table;

The data statistics module includes a series of general statistical and graphing programs for energy analysis and structural analysis, which can read the original data table of energy and structural analysis results, call corresponding programs, calculate statistical indicators, print statistical data, and draw statistics. chart.

Among them, the data acquisition module selects the corresponding tool according to the specific situation, and the other modules are implemented with Python programs, involving numpy, scipy, pandas, matplotlib and rdkit libraries, integrating indicator calculation, data statistics and drawing processes into Jupyter notebook .

The control method of the molecular force field quality control system includes the following steps:

(1) Preparation of high-precision quantum chemical data, namely QM data: select the target test set, use quantum chemical calculation software, and use high-precision quantum chemical methods to scan the potential energy surface of the rotatable dihedral angle of the molecule, and the scan interval is generally 15 or 30 degrees, the resulting energy E _QM and coordinate r _QM are used for benchmarking;

(2) Preparation of molecular mechanics data of the force field to be measured, namely MM data: using the conformation in step (1) as the input conformation, one or more force fields to be measured are used to perform molecular mechanics optimization for more than 1000 steps respectively, and the obtained Energy data E _MM1 , E _MM2 , ... are saved in table format, energy unit is kcal/mol, and structure data is saved in .mol or .sdf format;

(3) Energy translation: input the data table, translate the E _QM and E _MMi of each rotatable dihedral angle, i=1, 2,..., so that the value of the lowest energy point of E is 0; E _MMi has two There are a variety of translation strategies to choose from, namely, the lowest point is 0 and the variance with E _QM is minimized;

(4) The energy analysis module calculates the single conformation energy related indexes and potential energy surface related indexes in turn, and can extract a specific potential energy surface for drawing and viewing;

(5) The structural analysis module calculates the RMSD, bond length, bond angle, and dihedral angle related indexes in turn. The force field developer can additionally input the parameter item data table to calculate the parameter item related indexes;

(6) The data statistics module reads the calculation results of steps (4) and (5), defines the colors and labels of pictures and tables, and outputs statistical data and reports.

The molecular force field quality control system and its control method provided by the present invention have the following technical advantages:

(1) Standardized set of standardized indicators for molecular force field quality control.

(2) A series of unreported indexes are proposed to evaluate the similarity between the shape of the dihedral angle potential energy surface and the QM potential energy surface.

(3) Incorporating the force field parameter items into the quality control process, the force field developers can accurately identify the parameter items that perform poorly in the test set, so as to optimize the parameters in a targeted manner.

(4) Indicator calculation and data analysis tools are decoupled from commercial software, which can process and analyze calculation data from a wide range of sources.

(5) The indicator calculation and data statistics and drawing process are integrated into the Jupyter notebook to realize visualization.

Description of drawings

Fig. 1 is the molecular force field quality control process flow of the present invention;

Fig. 2 is the potential energy surface diagram of embodiment;

Fig. 3 is the molecular structure diagram of the embodiment;

Fig. 4a is the conformational energy correlation of Example QM;

Fig. 4b is the energy correlation of the MM conformation of the embodiment;

Fig. 5a is the QM energy dependence of the potential energy surface minima of the embodiment;

Fig. 5b is the MM energy dependence of the potential energy surface minima of the embodiment;

Fig. 6 is the potential energy surface RMSE distribution of embodiment;

Fig. 7 is the distribution of potential energy surface minimum value correlation coefficient R of the embodiment;

Fig. 8 is the distribution of potential energy surface correlation coefficient R of the embodiment;

Fig. 9 is the position (angle) deviation of the extremum value of the potential energy surface of the embodiment;

Fig. 10a is the two-dimensional distribution of potential energy surface minimum value energy and position error of the GAFF2 force field of the embodiment;

Fig. 10b is the two-dimensional distribution of the minimum value energy and position error of the potential energy surface of the OpenFF force field of the embodiment;

Figure 11 is a root mean square displacement (RMSD) distribution for an embodiment;

Fig. 12 is the bond length deviation distribution of embodiment;

Fig. 13 is the bond angle deviation distribution of embodiment;

FIG. 14 is a non-fixed dihedral angle deviation distribution of the embodiment.

Detailed ways

The specific technical solutions of the present invention are described with reference to the embodiments.

The specific technical solutions are:

The quality control index set of molecular force field is shown in Table 1. The metric set defines a series of metrics that evaluate the performance of the force field parameters on the test set.

Table 1 Molecular force field quality control index set

The data acquisition module can choose the corresponding tool according to the specific situation, and the rest of the modules are implemented by Python programs, involving numpy, scipy, pandas, matplotlib and rdkit libraries, integrating indicator calculation and data statistics and drawing process into Jupyter notebook. This process architecture removes the reliance on commercial software, facilitates use by users and developers with scientific computing backgrounds, and enables visualization.

Data acquisition module Data acquisition can be based on the source of molecules in the test set, the source of the force field to be measured, and user preferences, and corresponding computational simulation tools can be selected to generate the target quantum mechanics optimization data and the molecular mechanics optimization data of the force field to be measured, and organize them as Contains the data table and molecular coordinate file of the specified column as the input data of the analysis module.

The energy analysis module calculates the evaluation indexes related to the molecular conformational energy and potential energy surface of the index set. The program reads the molecular name provided by the user, the atomic number of the dihedral angle, the angle and the conformational energy provided by the user, calculates the energy analysis index according to the two test method parameters, and generates a raw data table.

The structure analysis module calculates the evaluation indexes related to the molecular structure of the index set. The program reads the molecular name, the atomic number of the dihedral angle and the 3D coordinates provided by the user to limit the optimization, calculates the structural analysis index, and generates a raw data table.

The data statistics module includes a series of general statistical and graphing programs for energy analysis and structural analysis. It can read the original data table of energy and structural analysis results, call corresponding programs, calculate statistical indicators, print statistical data, and draw statistical charts.

As shown in Figure 1, the control method of the molecular force field quality control system includes the following steps:

(1) Preparation of high-precision quantum chemical data (QM data): select the target test set, use quantum chemical calculation software (such as PSI4), and use high-precision quantum chemical methods (such as B3LYP/6-31G(d)) to analyze molecules The potential energy surface is scanned at the rotatable dihedral angle of , and the scanning interval is generally 15 or 30 degrees, and the obtained energy E _QM and coordinate r _QM are used for benchmarking.

(2) Preparation of molecular mechanics data (MM data) of the force field to be measured: using the conformation in step (1) as the input conformation, one or more force fields to be measured are used to perform molecular mechanics optimization for more than 1000 steps respectively, and the obtained The energy data E _MM1 , E _MM2 , ... are saved in the format shown in Table 2, the energy unit is kcal/mol, and the structure data is saved in .mol or .sdf format.

Table 2 Input data table

(3) Energy translation: Enter the table as shown in Table 2, and translate the E _QM and E _MMi (i=1, 2,...) of each rotatable dihedral angle, so that the value of the lowest energy point of E is 0 . E _MMi has two translation strategies to choose from, which are nadir at 0 and minimization of variance with E _QM .

(4) The energy analysis module calculates the single conformation energy related indexes and potential energy surface related indexes in turn, and can extract a specific potential energy surface for drawing and viewing.

(5) The structural analysis module calculates the RMSD and the related indexes of bond length, bond angle and dihedral angle in turn. The force field developer can additionally input the parameter item data table to calculate the parameter item related indexes.

(6) The data statistics module reads the calculation results of steps (4) and (5), defines the colors and labels of pictures and tables, and outputs statistical data and reports, as shown in Table 3-Table 5, Figure 4a-Figure 14 .

This example takes the Roche test set as an example. The test set contains 459 molecules with rotatable dihedral angles, and evaluates the performance of two open source force fields, GAFF2 and OpenFF. First download the energy and structure of the QM potential energy surface scan from the publicly released web page, then use the Amber software to obtain the MM data of the GAFF2 force field, use the OpenMM software to calculate the MM data of the OpenFF force field, and make the energy table shown in Table 2, And store the molecular coordinates in .mol format.

Use the energy analysis module to analyze the energy table to obtain a QC energy analysis data table, including the QC index values of a single conformation.

Molecular coordinates were analyzed using the structural analysis module to obtain RMSD, bond length, bond angle and dihedral angle data tables.

Use the data statistics module to generate statistical data and reports, as shown in Table 3-Table 5, Figure 4a-Figure 14.

Table 3 Test parameters, conformational energy deviation, dihedral angle RMSE and correlation coefficient R statistics

Table 4 Matching rate of extreme points of potential energy surface

Table 5 Example of print result of structure index

It can be seen from the statistical report that OpenFF's deviation is generally smaller than that of GAFF2 on most of the energy indicators shown in Table 3 and Figure 4a-Figure 8, and the performance is better; but in terms of the description of the shape of the potential energy surface, the potential energy in Table 3 The surface correlation coefficient R, GAFF2 is comparable to OpenFF, the matching rate of the potential energy surface minimum and global minimum shown in Table 4 and the two-dimensional distribution of the error of the minimum energy and position shown in Figure 10a and Figure 10b, in In terms of exact matching of potential energy surface shapes, GAFF2 performs even slightly better than OpenFF. In terms of structure, according to the statistical indicators in Table 5 and the distribution of indicators in Figures 11-14, the performance of OpenFF is still slightly lower in the two indicators of RMSD and dihedral angle deviation, which are more important to the quality of the force field. better than GAFF2.

After sorting the correlation coefficient of the potential energy surface, it is found that the correlation coefficient R of the GAFF2 potential energy surface of the 006-C12H12N2 molecule and the QM results is less than 0, showing a negative correlation, and the GAFF2 force field does not perform well. Extracting the potential energy surface data to draw the potential energy surface diagram shown in Figure 2 and the molecular structure shown in Figure 3, it can be determined that the molecule where the dihedral angle is located belongs to the secondary aromatic amines, in this case the angle C3-N4- C11-C10. The energy barrier of the potential energy surface is less than 5kcal/mol, and the energy barrier of the dihedral angle rotation is lower, but the description of the extreme point position by the GAFF2 force field is wrong. Therefore, when users need to calculate the molecules of secondary aromatic amines, they can choose other force fields, and force field developers can consider optimizing the parameters of such molecules.

Claims

The molecular force field quality control system is characterized in that it includes four modules: a data acquisition module, an energy analysis module, a structural analysis module and a data statistics module;

In the data acquisition module, according to the molecular source of the test set, the source of the force field to be measured and user preference, the data acquisition selects corresponding computational simulation tools to generate the quantum mechanical optimization data for the target and the molecular mechanics optimization data of the force field to be measured, And organize it into a data table and molecular coordinate file containing the specified column, as the input data of the energy analysis module;

The energy analysis module calculates the evaluation indexes related to the molecular conformational energy and potential energy surface of the index set; the program reads the molecular name, the dihedral atomic number, angle and conformational energy provided by the user to limit optimization, and according to two test methods Parameter calculation energy analysis index, generating original data table;

The structure analysis module calculates the evaluation index related to the molecular structure of the index set; the program reads the molecular name, the dihedral atomic number and 3D coordinates provided by the user to limit optimization, calculates the structure analysis index, and generates a raw data table;

The data statistics module includes a series of general statistical and graphing programs for energy analysis and structural analysis, which can read the original data table of energy and structural analysis results, call corresponding programs, calculate statistical indicators, print statistical data, and draw statistics. chart.
The molecular force field quality control system according to claim 1, wherein the data acquisition module selects corresponding tools according to specific conditions, and the remaining modules are implemented with Python programs, involving numpy, scipy, pandas, matplotlib and rdkit Libraries that integrate metrics calculation and data statistics and graphing processes into Jupyter notebooks.
The control method of the molecular force field quality control system according to claim 1 or 2, is characterized in that, comprises the following steps:

(1) Preparation of high-precision quantum chemical data, namely QM data: select the target test set, use quantum chemical calculation software, and use high-precision quantum chemical methods to scan the potential energy surface of the rotatable dihedral angle of the molecule, and the scan interval is generally 15 or 30 degrees, the resulting energy E QM and coordinate r QM are used for benchmarking;

(2) Preparation of molecular mechanics data of the force field to be measured, namely MM data: using the conformation in step (1) as the input conformation, one or more force fields to be measured are used to perform molecular mechanics optimization for more than 1000 steps respectively, and the obtained Energy data E MM1 , E MM2 , ... are saved in table format, energy unit is kcal/mol, and structure data is saved in .mol or .sdf format;

(3) Energy translation: input the data table, translate the E QM and E MMi of each rotatable dihedral angle, i=1, 2,..., so that the value of the lowest energy point of E is 0; E MMi has two There are a variety of translation strategies to choose from, namely, the lowest point is 0 and the variance with E QM is minimized;

(4) The energy analysis module calculates the single conformation energy-related index and the potential energy surface related index in turn, and can extract a specific potential energy surface for drawing and viewing;

(5) The structural analysis module calculates the RMSD, bond length, bond angle, and dihedral angle related indexes in turn. The force field developer can additionally input the parameter item data table to calculate the parameter item related indexes;

(6) The data statistics module reads the calculation results of steps (4) and (5), defines the colors and labels of pictures and tables, and outputs statistical data and reports.