A method and apparatus for analyzing property correlations of molecular substructures

Abstract

Embodiments of the present application relate to a method and device for analyzing the property correlation of molecular substructure, the method comprising: constructing and training a molecular property prediction model; receiving a current molecular structure after the training is completed; if the current molecular structure is a two-dimensional topological structure, generating a first molecular conformation based on the current structure, and if the current molecular structure is a three-dimensional molecular conformation, taking it as the first molecular conformation; identifying the substructure of the first molecular conformation; extracting the backbone conformation and each branch conformation in the sub-conformation set and performing branch cutting processing on the first molecular conformation based on each branch conformation to obtain a corresponding local conformation; predicting the N-class specified molecular property of the first molecular conformation, the backbone conformation and each local conformation based on the molecular property prediction model, and analyzing the positive / negative correlation of all substructures and the N-class specified molecular property according to the prediction result. Based on the present application, the analysis efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and apparatus for analyzing the correlation of properties of molecular substructures. Background Technology

[0002] In fields such as chemical industry, materials science, and drug development, understanding the correlation (positive correlation and its degree, negative correlation and its degree) between molecular substructures (such as the framework and branches) and properties (physical / chemical properties) is of great significance for guiding practical applications. Currently, most conventional analytical methods for analyzing the correlation between the properties of molecular substructures are based on experimental methods. However, experimental methods generally suffer from problems such as high experimental complexity and long experimental cycles, which naturally make it difficult to effectively improve the analytical efficiency of conventional methods. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for analyzing the property correlations of molecular substructures. This invention pre-constructs and trains a molecular property prediction model; after the model training is completed, it receives a user input of the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule as the corresponding current molecular structure; when the current molecular structure is a two-dimensional topological structure, it performs corresponding three-dimensional molecular conformation generation processing based on the current structure to obtain the corresponding current molecular conformation; when the current molecular structure is a three-dimensional molecular conformation, it directly uses it as the corresponding current molecular conformation; it uses conventional chemical information tools to identify substructures of the current molecular conformation to obtain the corresponding set of subconformations, and extracts the skeleton and each branch conformation from the set of subconformations as the corresponding skeleton and branch conformations; it performs a branch pruning process on the current molecular conformation based on each branch conformation (i.e., removing a branch conformation from the current molecular conformation) to obtain a corresponding local conformation; and based on the molecular property prediction model, it predicts multiple specified molecular properties of the current molecular conformation, skeleton conformation, and each local conformation, and analyzes the positive / negative correlation between all substructures and all specified molecular properties based on the prediction results to obtain a corresponding analysis report and provide feedback to the user. This invention shortens the property identification cycle and improves identification efficiency by introducing an artificial intelligence model, namely a molecular property prediction model. It uses correlation analysis to identify the correlation between all substructures and all specified molecular properties, and ranks the positive / negative correlations of all substructures for each specified molecular property. Based on this invention, not only can the complexity of analysis be reduced and the efficiency improved, but multi-dimensional analytical conclusions can also be provided.

[0004] To achieve the above objectives, a first aspect of the present invention provides a method for analyzing the property correlation of molecular substructures, the method comprising:

[0005] A molecular property prediction model is constructed; and the model is trained based on a preset first dataset; the input of the molecular property prediction model is a three-dimensional molecular conformation, and the output of the model is a corresponding molecular property prediction vector; the molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input to the model and output the corresponding molecular property prediction vector; the total number of specified molecular properties N is a positive integer; each specified molecular property corresponds to a preset class of molecular physical or chemical properties;

[0006] After the model training is completed, it receives a two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure.

[0007] If the current molecular structure is a two-dimensional topological structure, a preset three-dimensional structure modeling tool is used to generate a corresponding three-dimensional molecular conformation based on the current molecular structure to obtain a corresponding first molecular conformation; if the current molecular structure is a three-dimensional molecular conformation, the current molecular structure is used as the corresponding first molecular conformation; and a preset chemical information tool is used to perform substructure recognition processing on the first molecular conformation to obtain a corresponding first subconformation set.

[0008] The unique skeleton conformation in the first subconformation set is extracted as the corresponding first skeleton conformation; each branch conformation in the first subconformation set is extracted as the corresponding first branch conformation; and the first molecular conformation is pruned based on each first branch conformation to obtain the corresponding first local conformation.

[0009] Based on the molecular property prediction model, the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations are predicted. Based on the prediction results, the positive / negative correlation between all substructures and the N types of specified molecular properties is analyzed to obtain the corresponding first analysis report, which is then fed back to the current user.

[0010] Preferably, the three-dimensional molecular conformation input to the model consists of an atom set and a chemical bond set; the atom set consists of multiple atoms; the chemical bond set consists of multiple chemical bonds; each atom in the atom set corresponds to a set of atom object parameters, the atom object parameters including at least an atom identifier, an atom type, and an atom coordinates; the atom identifier is a unique identifier for the current atom; the atom type is the chemical element type of the current atom; the atom coordinates are three-dimensional coordinates; each chemical bond in the chemical bond set corresponds to a set of chemical bond object parameters, the chemical bond object parameters including at least a chemical bond identifier, a chemical bond type, and a chemical bond atom group; the chemical bond identifier is a unique identifier for the current chemical bond; the chemical bond type is the bonding type of the current chemical bond; the chemical bond atom group consists of the atom identifiers of the two atoms connected by the current chemical bond.

[0011] The molecular property prediction vector output by the model is formed by sequentially sorting N property prediction values; each property prediction value corresponds one-to-one with the specified molecular property.

[0012] The molecular property prediction model consists of a first feature extraction module, N first property prediction heads, and a first property splicing module; each of the first property prediction heads corresponds one-to-one with the specified molecular property.

[0013] The input of the first feature extraction module is connected to the input of the model, and the output is connected to the input of each of the first property prediction heads; the output of each of the first property prediction heads is connected to one input of the first property splicing module; the output of the first property splicing module is connected to the output of the model.

[0014] The first feature extraction module is implemented based on the encoder structure of the transformer model; the first feature extraction module is used to perform molecular structure feature extraction processing on the three-dimensional molecular conformation input by the model to obtain the corresponding extracted feature tensor and send it to each of the first property prediction heads;

[0015] Each of the first property prediction heads is a regression calculation model, implemented based on a type of deep learning model; the deep learning model includes at least an MLP model; each of the first property prediction heads is used to perform corresponding molecular property regression prediction processing based on the extracted feature tensor to obtain the corresponding property prediction value and send it to the first property splicing module;

[0016] The first property splicing module is used to sort the obtained N property prediction values ​​according to a preset molecular property arrangement order to form a corresponding molecular property prediction vector and output it.

[0017] Preferably, the first dataset includes multiple first data records; the first data record includes a first training structure and a first label vector; the first training structure is a three-dimensional molecular conformation, composed of the corresponding set of atoms and the set of chemical bonds; the first label vector is a molecular property label vector, composed of N first property label values; the first property label values ​​correspond one-to-one with the specified molecular property.

[0018] Preferably, the step of training the molecular property prediction model based on a preset first dataset specifically includes:

[0019] Step 41: Based on a preset first segmentation ratio, the first dataset is divided into two sub-datasets, denoted as the first training set and the first evaluation set.

[0020] Wherein, both the first training set and the first evaluation set are composed of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio;

[0021] Step 42: Extract the first data record of the first training set as the corresponding current training record;

[0022] Step 43: Input the first training structure of the current training record as the corresponding current three-dimensional molecular conformation into the molecular property prediction model. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation and uses the molecular property prediction vector output by the molecular property prediction model as the corresponding first prediction vector.

[0023] Step 44: Substitute the first prediction vector and the first label vector of the current training record into the preset first model loss function; and modulate the model parameters of the molecular property prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function;

[0024] Wherein, the first model loss function includes at least the L1 loss function and the L2 loss function; the first model optimizer includes at least the SGD optimizer and the Adam optimizer;

[0025] Step 45: Identify whether the current training record is the last first data record in the first training set; if yes, proceed to step 46; if no, extract the next first data record in the first training set as the new current training record and return to step 43.

[0026] Step 46: Perform a traversal of all the first data records in the first evaluation set; during this traversal, take the currently traversed first data record as the corresponding current evaluation record; input the first training structure of the current evaluation record as the corresponding current three-dimensional molecular conformation into the molecular property prediction model, and have the molecular property prediction model predict the N specified molecular properties of the current three-dimensional molecular conformation, and take the molecular property prediction vector output by the molecular property prediction model as the corresponding second prediction vector; form a corresponding first prediction-label pair with the second prediction vector and the first label vector of the current training record; and at the end of this traversal, input all the obtained first prediction-label pairs into a preset first model evaluation function to calculate the corresponding first evaluation value;

[0027] The first model evaluation function includes at least the RMSE function;

[0028] Step 47: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 42 to continue training; if it meets the range, stop training and confirm that the training of the molecular property prediction model has ended.

[0029] Preferably, the three-dimensional structure modeling tools include at least ChemDraw, Avogadro, Jmol, Spartan, Discovery Studio, and Collector; the chemical information tools include at least RDKIT, KNIME, and ChemAxon.

[0030] The first molecular conformation includes the first set of atoms and the first set of chemical bonds;

[0031] The first set of atoms includes multiple first atoms; the parameters of the first atom object corresponding to each first atom include at least a first atom identifier, a first atom type, and first atom coordinates; the first atom identifier is the unique identifier of the current first atom; the first atom type is the chemical element type of the current first atom; and the first atom coordinates are the three-dimensional coordinates of the current first atom.

[0032] The first set of chemical bonds includes multiple first chemical bonds; the first chemical bond object parameters corresponding to each first chemical bond include at least a first chemical bond identifier, a first chemical bond type, and a first chemical bond atom group; the first chemical bond identifier is a unique identifier for the current first chemical bond; the first chemical bond type is the bonding type of the current first chemical bond; the first chemical bond atom group is composed of the first atom identifiers of the two first atoms connected by the current first chemical bond;

[0033] The first subconformation set includes multiple first subconformations; each first subconformation includes a first subconformation identifier, a first subconformation type, a first subconformation atom set, and a first subconformation chemical bond set; the first subconformation identifier is a unique identifier for the current first subconformation; the first subconformation type includes skeletal conformations and branched conformations; the first subconformation atom set consists of all the first atoms within the current first subconformation; the first subconformation chemical bond set consists of all the first chemical bonds within the current first subconformation.

[0034] There is one and only one first sub-conformation in the first sub-conformation set whose first sub-conformation type is skeletal architecture; the first sub-conformation in the first sub-conformation set whose first sub-conformation type is skeletal architecture is denoted as the corresponding first skeletal architecture; the first sub-conformation in the first sub-conformation set whose first sub-conformation type is branched architecture is denoted as the corresponding first branched architecture.

[0035] Each of the first local conformations is obtained by shaving off one of the first branch conformations from the first molecular conformation; the first local conformation includes a first shave-off conformation identifier, a first local conformation atom set, and a first local conformation chemical bond set; the first shave-off conformation identifier matches the first subconformation identifier of the shave-off first branch conformation corresponding to the current first local conformation; the first local conformation atom set consists of all the first atoms in the current first local conformation; the first local conformation chemical bond set consists of all the first chemical bonds in the current first local conformation.

[0036] Preferably, the step of pruning the first molecular conformation based on each of the first branch conformations to obtain the corresponding first local conformation specifically includes:

[0037] A traversal is performed on all the first branch conformations. During this traversal, the currently traversed first branch conformation is taken as the corresponding current pruning conformation. The first sub-conformation atom set of the current pruning conformation is taken as the corresponding current pruning atom set. All first atoms in the first atom set of the first molecular conformation, except for the current pruning atom set, are extracted to form the corresponding current local atom set. All first atom identifiers in the current local atom set are used to form the corresponding current atom identifier set. The two first atom identifiers of each first chemical bond atom group in the first chemical bond set of the first molecular conformation satisfy the first chemical bond of the current atom identifier set and are extracted to form the corresponding current local chemical bond set. The first sub-conformation identifier corresponding to the current pruning conformation is taken as a corresponding first pruning conformation identifier. The current local atom set and the current local chemical bond set are taken as the corresponding first local conformation atom set and the first local conformation chemical bond set. The first pruning conformation identifier, the first local conformation atom set, and the first local conformation chemical bond set corresponding to the current pruning conformation are used to form a corresponding first local conformation.

[0038] Preferably, the step of predicting the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations based on the molecular property prediction model, and analyzing the positive / negative correlation between all substructures and the N types of specified molecular properties based on the prediction results to obtain a corresponding first analysis report and providing feedback to the current user, specifically includes:

[0039] Step 71: A corresponding current three-dimensional molecular conformation is formed by the first set of atoms and the first set of chemical bonds of the first molecular conformation. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation, and the predicted values ​​of each property in the molecular property prediction vector output by the molecular property prediction model are used as a corresponding global property value x. g,i 1 ≤ property index i ≤ N; and N global property values ​​x obtained. g,i Form the corresponding global property sequence;

[0040] Step 72, based on each of the global property values ​​x g,i The specified molecular properties and their corresponding molecular properties form a first property data set; and the N first property data sets formed a first molecular analysis record.

[0041] Step 73: A corresponding current three-dimensional molecular conformation is formed by the first sub-conformation atom set and the first sub-conformation chemical bond set of the first skeletal conformation. The molecular property prediction model predicts N types of specified molecular properties of the current three-dimensional molecular conformation, and each property prediction value of the molecular property prediction vector output by the molecular property prediction model is used as a corresponding skeleton property value x. b,i ; and the obtained N skeleton property values ​​x b,i Form the corresponding skeleton property sequence;

[0042] Step 74: The total number of the first local conformations is statistically analyzed to obtain the corresponding total number of local conformations M; a current three-dimensional molecular conformation is formed by the set of atoms and chemical bonds of each first local conformation; the molecular property prediction model predicts N types of specified molecular properties of the current three-dimensional molecular conformation; and each property prediction value of the molecular property prediction vector output by the molecular property prediction model is used as a corresponding local property value x. i,j 1 ≤ local conformation index j ≤ M; and N local property values ​​x corresponding to each of the first local conformations. i,j Form the corresponding sequence of local properties;

[0043] Step 75: Match the pair of global property values ​​x that have the same property index i in the global property sequence and the skeleton property sequence. g,i and the skeleton property value x b,i Form a corresponding first alignment data group; and perform one round of traversal on N first alignment data groups; during this round of traversal, take the currently traversed first alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and the skeleton property value x b,i The current global property value and the current skeleton property value are used as the corresponding values. The difference between the current skeleton property value and the current global property value is used as the corresponding first relative difference. The first relative difference is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive correlation. If it is, the corresponding first correlation is set to negative correlation. The specified molecular property, the first relative difference, and the first correlation corresponding to the current comparison data group are combined to form a corresponding first property analysis result. At the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence. The first substructure analysis record is formed by the first subconformation identifier, the first subconformation type, and the first property analysis result sequence corresponding to the first skeleton image.

[0044] The sequence of first property analysis results is formed by sequentially sorting N first property analysis results; each of the N first property analysis results corresponds one-to-one with one of the N specified molecular properties; each first property analysis result consists of a class of specified molecular properties and their corresponding first relative difference and first correlation; the first correlation includes positive and negative correlation; the first correlation is positive when the first relative difference is greater than or equal to 0, and negative when the first relative difference is less than 0;

[0045] Step 76: Each of the local property sequences is taken as the corresponding current local property sequence; and the pair of global property values ​​x that have the same property index i in the global property sequence and the current local property sequence are selected. g,i and the local property value x i,j Form a corresponding second alignment data group; and perform one round of traversal on N second alignment data groups; during this round of traversal, take the currently traversed second alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and the local property value x i,j The current global property value and the current local property value are used as the corresponding values; the difference between the current global property value and the current local property value is used as the corresponding first relative difference value; the first relative difference value is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive correlation; if so, the corresponding first correlation is set to negative correlation; the specified molecular property, the first relative difference value, and the first correlation value of the current comparison data group are combined to form a corresponding first property analysis result; at the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence; the first branch conformation corresponding to the first scissor conformation identifier of the first local conformation of the current local property sequence is used as the corresponding current branch conformation; and the first sub-conformation identifier, the first sub-conformation type, and the first property analysis result sequence of the current branch conformation are combined to form a corresponding first substructure analysis record;

[0046] Step 77: A corresponding set of first substructure analysis records is formed by the first skeleton image and 1+M first substructure analysis records corresponding to M first local images.

[0047] Step 78: Extract 1+M first property analysis results corresponding to the i-th type of specified molecular property from the first substructure analysis record set to form a corresponding current property analysis result set; extract the first property analysis results with positive first correlation from the current property analysis result set to form a corresponding positive correlation set, and extract the first property analysis results with negative first correlation to form a corresponding negative correlation set; sort all the first relative differences in the positive and negative correlation sets in descending order of relative difference to obtain corresponding positive correlation difference sequences and negative correlation difference sequences; sort the first subconformation identifiers corresponding to each first relative difference in the positive correlation difference sequence according to the current sorting order of the positive correlation difference sequence to obtain a corresponding positive correlation substructure sequence; sort the first subconformation identifiers corresponding to each first relative difference in the negative correlation difference sequence according to the current sorting order of the negative correlation difference sequence to obtain a corresponding negative correlation substructure sequence; and form a corresponding first property analysis record from the i-th specified molecular property and its corresponding positive and negative correlation substructure sequences.

[0048] Step 79: The obtained N first property analysis records are combined to form a corresponding first property analysis record set;

[0049] Step 80: The first analysis report, composed of the obtained first molecule analysis record, first substructure analysis record set, and first property analysis record set, is fed back to the current user.

[0050] A second aspect of the present invention provides an apparatus for implementing the method for analyzing the property correlation of molecular substructures as described in the first aspect above. The apparatus includes: a model building and training module, a molecular structure receiving module, a substructure identification module, a local conformation processing module, and a correlation analysis module.

[0051] The model building and training module is used to build a molecular property prediction model; and to train the molecular property prediction model based on a preset first dataset; the model input of the molecular property prediction model is a three-dimensional molecular conformation, and the model output is a corresponding molecular property prediction vector; the molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input to the model and output the corresponding molecular property prediction vector; the total number of specified molecular properties N is a positive integer; each specified molecular property corresponds to a preset class of molecular physical or chemical properties;

[0052] The molecular structure receiving module is used to receive the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure after the model training is completed.

[0053] The substructure recognition module is used to, when the current molecular structure is a two-dimensional topological structure, perform corresponding three-dimensional molecular conformation generation processing based on a preset three-dimensional structure modeling tool to obtain the corresponding first molecular conformation; when the current molecular structure is a three-dimensional molecular conformation, take the current molecular structure as the corresponding first molecular conformation; and use a preset chemical information tool to perform substructure recognition processing on the first molecular conformation to obtain the corresponding first subconformation set.

[0054] The local conformation processing module is used to extract the unique skeleton conformation in the first sub-conformation set as the corresponding first skeleton conformation; and to extract each branch conformation in the first sub-conformation set as the corresponding first branch conformation; and to perform branch pruning on the first molecular conformation based on each first branch conformation to obtain the corresponding first local conformation.

[0055] The correlation analysis module is used to predict the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations based on the molecular property prediction model, and to analyze the positive / negative correlation between all substructures and the N types of specified molecular properties based on the prediction results, and to obtain the corresponding first analysis report and provide feedback to the current user.

[0056] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0057] The processor is used to couple with the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect above;

[0058] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0059] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0060] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for analyzing the property correlation of molecular substructures. As described above, this embodiment of the invention pre-constructs and trains a molecular property prediction model; after the model training is completed, it receives a user input of the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule as the corresponding current molecular structure; when the current molecular structure is a two-dimensional topological structure, it performs corresponding three-dimensional molecular conformation generation processing based on the current structure to obtain the corresponding current molecular conformation; when the current molecular structure is a three-dimensional molecular conformation, it directly uses it as the corresponding current molecular conformation; it uses conventional chemical information tools to identify substructures of the current molecular conformation to obtain the corresponding set of subconformations, and extracts the skeleton and each branch conformation in the set of subconformations as the corresponding skeleton and branch conformations; it performs a branch pruning process on the current molecular conformation based on each branch conformation (i.e., pruning a branch conformation from the current molecular conformation) to obtain a corresponding local conformation; and it predicts multiple specified molecular properties of the current molecular conformation, skeleton conformation, and each local conformation based on the molecular property prediction model, and analyzes the positive / negative correlation between all substructures and all specified molecular properties based on the prediction results to obtain a corresponding analysis report and provide feedback to the current user. This invention shortens the property identification cycle and improves identification efficiency by introducing an artificial intelligence model, namely a molecular property prediction model. It uses correlation analysis to identify the correlation between all substructures and all specified molecular properties, and ranks the positive / negative correlations of all substructures for each specified molecular property. Based on this invention, not only is the analytical complexity reduced and the analytical efficiency improved, but the analytical dimensions of correlation analysis are also enriched, and the analytical quality is enhanced. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of a method for analyzing the property correlation of molecular substructures according to Embodiment 1 of the present invention;

[0062] Figure 2 This is a block diagram of the molecular property prediction model provided in Embodiment 1 of the present invention;

[0063] Figure 3 This is a module structure diagram of an apparatus for analyzing the correlation of properties of molecular substructures according to Embodiment 2 of the present invention;

[0064] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0066] Embodiment 1 of the present invention provides a method for analyzing the correlation of properties of molecular substructures, such as... Figure 1 The schematic diagram shows a method for analyzing the correlation of properties of molecular substructures according to Embodiment 1 of the present invention. The method mainly includes the following steps:

[0067] Step 1: Construct a molecular property prediction model; and train the molecular property prediction model based on a pre-set first dataset;

[0068] Specifically, this includes: Step 11, constructing a molecular property prediction model;

[0069] Here is as Figure 2 As shown in the module structure diagram of the molecular property prediction model provided in Embodiment 1 of the present invention, the model input of the molecular property prediction model of the present invention is a three-dimensional molecular conformation, and the model output is a corresponding molecular property prediction vector; the molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input by the model and output the corresponding molecular property prediction vector; here, the total number of specified molecular properties N is a positive integer; each specified molecular property corresponds to a preset type of molecular physical or chemical property, such as melting point, boiling point, density, mass, etc.

[0070] The 3D molecular conformation input to the model consists of an atom set and a chemical bond set. An atom set comprises multiple atoms; a chemical bond set comprises multiple chemical bonds. Each atom in the atom set corresponds to a set of atom object parameters, which include at least an atom identifier, atom type, and atom coordinates. The atom identifier is a unique identifier for the current atom; the atom type is the chemical element type of the current atom; and the atom coordinates are 3D coordinates. Each chemical bond in the chemical bond set corresponds to a set of chemical bond object parameters, which include at least a chemical bond identifier, chemical bond type, and a chemical bond atom group. The chemical bond identifier is a unique identifier for the current chemical bond; the chemical bond type is the bonding type of the current chemical bond; and the chemical bond atom group consists of the atom identifiers of the two atoms connected by the current chemical bond.

[0071] The molecular property prediction vector output by the model is composed of N property prediction values ​​ordered sequentially; each property prediction value corresponds one-to-one with a specified molecular property.

[0072] like Figure 2As shown, the molecular property prediction model of this embodiment consists of a first feature extraction module, N first property prediction heads, and a first property splicing module; here, the first property prediction head corresponds one-to-one with the specified molecular property.

[0073] The connection relationships of the various components of the molecular property prediction model are as follows: the input end of the first feature extraction module is connected to the model input end, and the output end is connected to the input end of each first property prediction head; the output end of each first property prediction head is connected to one input end of the first property splicing module; the output end of the first property splicing module is connected to the model output end.

[0074] The functions of each component of this molecular property prediction model are as follows:

[0075] 1) The first feature extraction module is implemented based on the encoder structure of the transformer model; the first feature extraction module is used to perform molecular structure feature extraction processing on the three-dimensional molecular conformation input by the model to obtain the corresponding extracted feature tensor and send it to each first property prediction head;

[0076] It should be noted that the first feature extraction module of this embodiment of the invention can be implemented not only based on the encoder structure of the transformer model described above, but also based on other encoder structures, such as an encoder structure built by a three-dimensional graph neural network.

[0077] 2) Each first property prediction head is a regression calculation model, implemented based on a type of deep learning model; the deep learning model includes at least an MLP model; each first property prediction head is used to perform corresponding molecular property regression prediction processing based on the extracted feature tensor to obtain the corresponding property prediction value and send it to the first property splicing module;

[0078] It should be noted that, in addition to being based on the MLP model described above, the various first property prediction heads in the embodiments of the present invention can also be based on other machine learning models or other deep learning models. Other machine learning models include XGBoost models, regression tree models, etc., and other deep learning models include regression calculation models constructed by CNN networks + linear networks, regression calculation models constructed based on LSTM models, etc.

[0079] 3) The first property splicing module is used to sort the obtained N property prediction values ​​according to the preset molecular property arrangement order to form the corresponding molecular property prediction vector and output it.

[0080] Step 12, and train the molecular property prediction model based on the preset first dataset;

[0081] The first dataset is a model training dataset pre-constructed through big data collection and labeling; the first dataset includes multiple first data records; each first data record includes a first training structure and a first label vector; the first training structure is a three-dimensional molecular conformation, composed of a corresponding set of atoms and a set of chemical bonds; the first label vector is a molecular property label vector, composed of N first property label values; each first property label value corresponds one-to-one with a specified molecular property.

[0082] The current step 12 specifically includes:

[0083] Step 121: Based on the preset first segmentation ratio, the first dataset is divided into two sub-datasets, denoted as the first training set and the first evaluation set.

[0084] Wherein, the first segmentation ratio is a pre-set ratio parameter, such as 8:2; both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio;

[0085] Step 122: Extract the first data record of the first training set as the corresponding current training record;

[0086] Step 123: The first training structure of the current training record is used as the corresponding current three-dimensional molecular conformation and input into the molecular property prediction model. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation and uses the molecular property prediction vector output by the molecular property prediction model as the corresponding first prediction vector.

[0087] Step 124: Input the first prediction vector and the first label vector of the current training record into the preset first model loss function; and modulate the model parameters of the molecular property prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function;

[0088] The first model loss function includes at least the L1 loss function and the L2 loss function; the first model optimizer includes at least the SGD optimizer and the Adam optimizer.

[0089] Step 125: Identify whether the current training record is the last first data record of the first training set; if yes, proceed to step 126; if no, extract the next first data record of the first training set as the new current training record and return to step 123.

[0090] Step 126: Perform a round of traversal on all first data records in the first evaluation set; during this round of traversal, take the currently traversed first data record as the corresponding current evaluation record; and take the first training structure of the current evaluation record as the corresponding current three-dimensional molecular conformation as input to the molecular property prediction model. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation, and takes the molecular property prediction vector output by the molecular property prediction model as the corresponding second prediction vector; and take the second prediction vector and the first label vector of the current training record as a corresponding first prediction-label pair; and at the end of this round of traversal, take all the obtained first prediction-label pairs into the preset first model evaluation function to calculate the corresponding first evaluation value.

[0091] The first model evaluation function includes at least the RMSE function;

[0092] Step 127: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 122 to continue training; if it meets the range, stop training and confirm that the training of the molecular property prediction model has ended.

[0093] Here, the first evaluation value range is a pre-set numerical range.

[0094] Step 2: After the model training is completed, the system receives a two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure.

[0095] Step 3: If the current molecular structure is a two-dimensional topological structure, the corresponding three-dimensional molecular conformation generation process is performed based on the preset three-dimensional structure modeling tool to obtain the corresponding first molecular conformation; if the current molecular structure is a three-dimensional molecular conformation, the current molecular structure is taken as the corresponding first molecular conformation; and the first molecular conformation is processed by the preset chemical information tool to obtain the corresponding first subconformation set.

[0096] Here, the three-dimensional structure modeling tools mentioned in the embodiments of the present invention include at least ChemDraw, Avogadro, Jmol, Spartan, Discovery Studio, and Collector; the chemical information tools include at least RDKIT, KNIME, and ChemAxon.

[0097] The first molecular conformation of this invention includes at least a first set of atoms and a first set of chemical bonds; wherein:

[0098] 1) The first atom set includes multiple first atoms; the parameters of the first atom object corresponding to each first atom include at least the first atom identifier, the first atom type, and the first atom coordinates; the first atom identifier is the unique identifier of the current first atom; the first atom type is the chemical element type of the current first atom; the first atom coordinates are the three-dimensional coordinates of the current first atom;

[0099] 2) The first chemical bond set includes multiple first chemical bonds; the first chemical bond object parameters corresponding to each first chemical bond include at least the first chemical bond identifier, the first chemical bond type, and the first chemical bond atom group; the first chemical bond identifier is the unique identifier of the current first chemical bond; the first chemical bond type is the bonding type of the current first chemical bond; the first chemical bond atom group is composed of the first atom identifiers of the two first atoms connected by the current first chemical bond.

[0100] The first subconformation set obtained by chemical information tools in this embodiment of the invention includes multiple first subconformations; each first subconformation corresponds to a substructure; the first subconformation includes a first subconformation identifier, a first subconformation type, a first subconformation atom set, and a first subconformation chemical bond set; the first subconformation identifier is a unique identifier for the current first subconformation; the first subconformation type includes skeletal conformations and branched conformations; the first subconformation atom set consists of all first atoms within the current first subconformation; the first subconformation chemical bond set consists of all first chemical bonds within the current first subconformation.

[0101] It should be noted that there is only one first sub-conformation in the set of first sub-conformations whose first sub-conformation type is skeletal; the first sub-conformation in the set of first sub-conformations whose first sub-conformation type is skeletal is denoted as the corresponding first skeletal; and the first sub-conformations in the set of first sub-conformations whose first sub-conformation type is branched are denoted as the corresponding first branched.

[0102] Step 4: Extract the unique skeleton conformation from the first sub-conformation set as the corresponding first skeleton conformation; extract each branch conformation from the first sub-conformation set as the corresponding first branch conformation; and perform branch pruning on the first molecular conformation based on each first branch conformation to obtain the corresponding first local conformation.

[0103] Specifically, this includes: Step 41, extracting the unique skeleton image from the first sub-conformation set as the corresponding first skeleton image;

[0104] Specifically, the first sub-conformation with the unique first sub-conformation type of skeleton in the first sub-conformation set is extracted as the corresponding first skeleton image.

[0105] Step 42, and extract each branch conformation in the first subconformation set as the corresponding first branch conformation;

[0106] Specifically, the first sub-conformation of the type of branched conformation in the first sub-conformation set is extracted as the corresponding first branched conformation.

[0107] Step 43, and based on each first branch conformation, the first molecule conformation is pruned to obtain the corresponding first local conformation;

[0108] Each first local conformation is obtained by shaving a first branch conformation from a first molecular conformation; the first local conformation includes a first shave-branch conformation identifier, a first local conformation atom set, and a first local conformation chemical bond set; the first shave-branch conformation identifier matches the first sub-conformation identifier of the shave-branch conformation corresponding to the current first local conformation; the first local conformation atom set consists of all first atoms within the current first local conformation; the first local conformation chemical bond set consists of all first chemical bonds within the current first local conformation;

[0109] Specifically, this includes: performing a traversal of all first-branch conformations; during this traversal, taking the currently traversed first-branch conformation as the corresponding current pruning conformation; taking the first subconformation atom set of the current pruning conformation as the corresponding current pruning atom set; extracting all first atoms from the first atom set of the first molecular conformation except for the current pruning atom set to form the corresponding current local atom set; and forming the corresponding current atom identifier set from all the first atom identifiers of the current local atom set; extracting the first chemical bonds from the first chemical bond atom sets of the first chemical bond set of the first molecular conformation where both first atom identifiers satisfy the current atom identifier set to form the corresponding current local chemical bond set; taking the first subconformation identifier corresponding to the current pruning conformation as a corresponding first pruning conformation identifier; taking the current local atom set and the current local chemical bond set as the corresponding first local conformation atom set and first local conformation chemical bond set; and forming a corresponding first local conformation from the first pruning conformation identifier, the first local conformation atom set, and the first local conformation chemical bond set corresponding to the current pruning conformation.

[0110] Step 5: Based on the molecular property prediction model, predict the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each first local conformation, and analyze the positive / negative correlation between all substructures and the N types of specified molecular properties according to the prediction results to obtain the corresponding first analysis report and provide feedback to the current user.

[0111] Specifically, this includes: Step 51, where the first set of atoms and the first set of chemical bonds of the first molecular conformation are combined to form a corresponding current three-dimensional molecular conformation, and the molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation, and the predicted values ​​of each property in the molecular property prediction vector output by the molecular property prediction model are used as a corresponding global property value x.g,i 1 ≤ property index i ≤ N; and N global property values ​​x are obtained. g,i Form the corresponding global property sequence;

[0112] Step 52, based on the various global property values ​​x g,i The N first property data sets are composed of the N first property data sets and their corresponding specified molecular properties; and the N first property data sets are composed of a corresponding first molecular analysis record.

[0113] Step 53: A corresponding current three-dimensional molecular conformation is formed by the first sub-conformation atom set and the first sub-conformation chemical bond set of the first skeleton conformation. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation, and the predicted values ​​of each property in the molecular property prediction vector output by the molecular property prediction model are used as corresponding skeleton property values ​​x. b,i ; and the obtained N skeleton property values ​​x b,i Form the corresponding skeleton property sequence;

[0114] Step 54: Count the total number of first local conformations to obtain the corresponding total number of local conformations M; and form a corresponding current three-dimensional molecular conformation by the first local conformation atom set and first local conformation chemical bond set of each first local conformation. Then, use the molecular property prediction model to predict the N types of specified molecular properties of the current three-dimensional molecular conformation, and use each property prediction value of the molecular property prediction vector output by the molecular property prediction model as a corresponding local property value x. i,j 1 ≤ local conformation index j ≤ M; and N local property values ​​x corresponding to each first local conformation. i,j Form the corresponding sequence of local properties;

[0115] Step 55: Connect the pairs of global property values ​​x with the same property index i in the global property sequence and the skeleton property sequence. g,i and skeleton property value x b,i Form a corresponding first alignment data group; perform one round of traversal on the N first alignment data groups; during this round of traversal, take the currently traversed first alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and skeleton property value x b,iThe current global property value and the current skeleton property value are used as the corresponding values. The difference between the current skeleton property value and the current global property value is taken as the corresponding first relative difference. The first relative difference is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive correlation. If it is, the corresponding first correlation is set to negative correlation. The specified molecular property, the first relative difference, and the first correlation of the current comparison data group are combined to form a corresponding first property analysis result. At the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence. The first substructure analysis record is formed by the first subconformation identifier, the first subconformation type, and the first property analysis result sequence corresponding to the first skeleton image.

[0116] The sequence of first property analysis results consists of N first property analysis results ordered sequentially; each of the N first property analysis results corresponds one-to-one with N specified molecular properties; each first property analysis result consists of a class of specified molecular properties and their corresponding first relative difference and first correlation; the first correlation includes positive and negative correlation; the first correlation is positive when the first relative difference is greater than or equal to 0, and negative when the first relative difference is less than 0;

[0117] Step 56: Take each local property sequence as the corresponding current local property sequence; and assign the global property values ​​x with the same property index i in both the global property sequence and the current local property sequence. g,i and local property value x i,j Form a corresponding second alignment data group; perform one round of traversal on the N second alignment data groups; during this round of traversal, take the currently traversed second alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and local property value x i,j The current global property value and the current local property value are used as the corresponding values. The difference between the current global property value and the current local property value is taken as the corresponding first relative difference. The first relative difference is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive; if so, the corresponding first correlation is set to negative. The specified molecular property, the first relative difference, and the first correlation of the current comparison data group are combined to form a corresponding first property analysis result. At the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence. The first branch conformation corresponding to the first scissor conformation identifier of the first local conformation of the current local property sequence is taken as the corresponding current branch conformation. The first sub-conformation identifier, the first sub-conformation type, and the first property analysis result sequence of the current branch conformation are combined to form a corresponding first substructure analysis record.

[0118] Step 57: A corresponding set of first substructure analysis records is formed by the first skeleton image and the 1+M first substructure analysis records corresponding to the M first local images.

[0119] Step 58: Extract the 1+M first property analysis results corresponding to the i-th specified molecular property from the first substructure analysis record set to form the corresponding current property analysis result set; extract the first property analysis results with positive first correlation from the current property analysis result set to form a corresponding positive correlation set, and extract the first property analysis results with negative first correlation to form a corresponding negative correlation set; sort all the first relative differences in the positive and negative correlation sets in descending order of relative difference to obtain the corresponding positive correlation difference sequence and negative correlation difference sequence; sort the first subconformation identifiers corresponding to each first relative difference in the positive correlation difference sequence according to the current sorting order of the positive correlation difference sequence to obtain a corresponding positive correlation substructure sequence; sort the first subconformation identifiers corresponding to each first relative difference in the negative correlation difference sequence according to the current sorting order of the negative correlation difference sequence to obtain a corresponding negative correlation substructure sequence; and form a corresponding first property analysis record from the i-th specified molecular property and its corresponding positive and negative correlation substructure sequences.

[0120] Step 59: The obtained N first-property analysis records form a corresponding first-property analysis record set;

[0121] Step 60: The first analysis report, composed of the obtained first molecule analysis record, first substructure analysis record set, and first property analysis record set, is fed back to the current user.

[0122] Figure 3 This is a module structure diagram of an apparatus for analyzing the correlation of properties of molecular substructures according to Embodiment 2 of the present invention. This apparatus can be a terminal device or server implementing the aforementioned method embodiments, or it can be an apparatus that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the apparatus can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building and training module 201, a molecular structure receiving module 202, a substructure recognition module 203, a local conformation processing module 204, and a correlation analysis module 205.

[0123] The model building and training module 201 is used to build a molecular property prediction model and train the model based on a preset first dataset. The input of the molecular property prediction model is a three-dimensional molecular conformation, and the output of the model is a corresponding molecular property prediction vector. The molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input by the model and output the corresponding molecular property prediction vector. The total number of specified molecular properties N is a positive integer. Each specified molecular property corresponds to a preset class of molecular physical or chemical properties.

[0124] The molecular structure receiving module 202 is used to receive the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure after the model training is completed.

[0125] The substructure recognition module 203 is used to generate a corresponding first molecular conformation based on a preset three-dimensional structure modeling tool when the current molecular structure is a two-dimensional topological structure; when the current molecular structure is a three-dimensional molecular conformation, the current molecular structure is used as the corresponding first molecular conformation; and a preset chemical information tool is used to perform substructure recognition processing on the first molecular conformation to obtain a corresponding set of first subconformations.

[0126] The local conformation processing module 204 is used to extract the unique skeleton conformation in the first sub-conformation set as the corresponding first skeleton conformation; and to extract each branch conformation in the first sub-conformation set as the corresponding first branch conformation; and to perform branching processing on the first molecular conformation based on each first branch conformation to obtain the corresponding first local conformation.

[0127] The correlation analysis module 205 is used to predict the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each first local conformation based on the molecular property prediction model, and to analyze the positive / negative correlation between all substructures and the N types of specified molecular properties based on the prediction results, and to obtain the corresponding first analysis report to be fed back to the current user.

[0128] The present invention provides an apparatus for analyzing the property correlation of molecular substructures, which can execute the method steps in the above method embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0129] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing elements; they can be fully implemented in hardware; or some modules can be implemented by processing elements calling software, while others are implemented in hardware. For example, the model building and training module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0130] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0131] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0132] Figure 4 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. Figure 4 As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the processing steps described in the foregoing embodiments. Preferably, the electronic device according to the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0133] exist Figure 4The system bus 305 mentioned can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0134] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0135] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0136] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for analyzing the property correlation of molecular substructures. As described above, this embodiment of the invention pre-constructs and trains a molecular property prediction model; after the model training is completed, it receives a user input of the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule as the corresponding current molecular structure; when the current molecular structure is a two-dimensional topological structure, it performs corresponding three-dimensional molecular conformation generation processing based on the current structure to obtain the corresponding current molecular conformation; when the current molecular structure is a three-dimensional molecular conformation, it directly uses it as the corresponding current molecular conformation; it uses conventional chemical information tools to identify substructures of the current molecular conformation to obtain the corresponding set of subconformations, and extracts the skeleton and each branch conformation in the set of subconformations as the corresponding skeleton and branch conformations; it performs a branch pruning process on the current molecular conformation based on each branch conformation (i.e., pruning a branch conformation from the current molecular conformation) to obtain a corresponding local conformation; and it predicts multiple specified molecular properties of the current molecular conformation, skeleton conformation, and each local conformation based on the molecular property prediction model, and analyzes the positive / negative correlation between all substructures and all specified molecular properties based on the prediction results to obtain a corresponding analysis report and provide feedback to the current user. This invention shortens the property identification cycle and improves identification efficiency by introducing an artificial intelligence model, namely a molecular property prediction model. It uses correlation analysis to identify the correlation between all substructures and all specified molecular properties, and ranks the positive / negative correlations of all substructures for each specified molecular property. Based on this invention, not only is the analytical complexity reduced and the analytical efficiency improved, but the analytical dimensions of correlation analysis are also enriched, and the analytical quality is enhanced.

[0137] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0138] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for analyzing the correlation of properties of molecular substructures, characterized in that, The method includes: A molecular property prediction model is constructed; and the model is trained based on a preset first dataset; the input of the molecular property prediction model is a three-dimensional molecular conformation, and the output of the model is a corresponding molecular property prediction vector; the molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input to the model and output the corresponding molecular property prediction vector; the total number of specified molecular properties N is a positive integer; each specified molecular property corresponds to a preset class of molecular physical or chemical properties; After the model training is completed, it receives a two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure. If the current molecular structure is a two-dimensional topological structure, a preset three-dimensional structure modeling tool is used to generate a corresponding three-dimensional molecular conformation based on the current molecular structure to obtain a corresponding first molecular conformation; if the current molecular structure is a three-dimensional molecular conformation, the current molecular structure is used as the corresponding first molecular conformation; and a preset chemical information tool is used to perform substructure recognition processing on the first molecular conformation to obtain a corresponding first subconformation set. The unique skeleton conformation in the first subconformation set is extracted as the corresponding first skeleton conformation; each branch conformation in the first subconformation set is extracted as the corresponding first branch conformation; and the first molecular conformation is pruned based on each first branch conformation to obtain the corresponding first local conformation. Based on the molecular property prediction model, the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations are predicted. Based on the prediction results, the positive / negative correlation between all substructures and the N types of specified molecular properties is analyzed to obtain the corresponding first analysis report, which is then fed back to the current user.

2. The method for analyzing the correlation of properties of molecular substructures according to claim 1, characterized in that, The three-dimensional molecular conformation input to the model consists of an atom set and a chemical bond set; the atom set consists of multiple atoms; the chemical bond set consists of multiple chemical bonds; each atom in the atom set corresponds to a set of atom object parameters, which include at least an atom identifier, an atom type, and atom coordinates; The atom identifier is a unique identifier for the current atom; The atom type is the chemical element type of the current atom; The atomic coordinates are three-dimensional coordinates; each chemical bond in the chemical bond set corresponds to a set of chemical bond object parameters, which include at least a chemical bond identifier, a chemical bond type, and a chemical bond atom group; the chemical bond identifier is a unique identifier for the current chemical bond; the chemical bond type is the bonding type of the current chemical bond; the chemical bond atom group is composed of the atom identifiers of the two atoms connected by the current chemical bond; The molecular property prediction vector output by the model is formed by sequentially sorting N property prediction values; each property prediction value corresponds one-to-one with the specified molecular property. The molecular property prediction model consists of a first feature extraction module, N first property prediction heads, and a first property splicing module; each of the first property prediction heads corresponds one-to-one with the specified molecular property. The input of the first feature extraction module is connected to the input of the model, and the output is connected to the input of each of the first property prediction heads; the output of each of the first property prediction heads is connected to one input of the first property splicing module; the output of the first property splicing module is connected to the output of the model. The first feature extraction module is implemented based on the encoder structure of the transformer model; the first feature extraction module is used to perform molecular structure feature extraction processing on the three-dimensional molecular conformation input by the model to obtain the corresponding extracted feature tensor and send it to each of the first property prediction heads; Each of the first property prediction heads is a regression calculation model, implemented based on a type of deep learning model; the deep learning model includes at least an MLP model; each of the first property prediction heads is used to perform corresponding molecular property regression prediction processing based on the extracted feature tensor to obtain the corresponding property prediction value and send it to the first property splicing module; The first property splicing module is used to sort the obtained N property prediction values ​​according to a preset molecular property arrangement order to form a corresponding molecular property prediction vector and output it.

3. The method for analyzing the correlation of properties of molecular substructures according to claim 2, characterized in that, The first dataset includes multiple first data records; the first data record includes a first training structure and a first label vector; the first training structure is a three-dimensional molecular conformation, composed of the corresponding set of atoms and the set of chemical bonds; the first label vector is a molecular property label vector, composed of N first property label values; the first property label values ​​correspond one-to-one with the specified molecular properties.

4. The method for analyzing the correlation of properties of molecular substructures according to claim 3, characterized in that, The process of training the molecular property prediction model based on a preset first dataset specifically includes: Step 41: Based on a preset first segmentation ratio, the first dataset is divided into two sub-datasets, denoted as the first training set and the first evaluation set. Wherein, both the first training set and the first evaluation set are composed of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio; Step 42: Extract the first data record of the first training set as the corresponding current training record; Step 43: Input the first training structure of the current training record as the corresponding current three-dimensional molecular conformation into the molecular property prediction model. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation and uses the molecular property prediction vector output by the molecular property prediction model as the corresponding first prediction vector. Step 44: Substitute the first prediction vector and the first label vector of the current training record into the preset first model loss function; and modulate the model parameters of the molecular property prediction model in one round based on the preset first model optimizer in the direction of minimizing the first model loss function; Wherein, the first model loss function includes at least the L1 loss function and the L2 loss function; the first model optimizer includes at least the SGD optimizer and the Adam optimizer; Step 45: Identify whether the current training record is the last first data record in the first training set; if yes, proceed to step 46; if no, extract the next first data record in the first training set as the new current training record and return to step 43. Step 46: Perform a traversal of all the first data records in the first evaluation set; during this traversal, take the currently traversed first data record as the corresponding current evaluation record; input the first training structure of the current evaluation record as the corresponding current three-dimensional molecular conformation into the molecular property prediction model, and have the molecular property prediction model predict the N specified molecular properties of the current three-dimensional molecular conformation, and take the molecular property prediction vector output by the molecular property prediction model as the corresponding second prediction vector; form a corresponding first prediction-label pair with the second prediction vector and the first label vector of the current training record; and at the end of this traversal, input all the obtained first prediction-label pairs into a preset first model evaluation function to calculate the corresponding first evaluation value; The first model evaluation function includes at least the RMSE function; Step 47: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 42 to continue training; if it meets the range, stop training and confirm that the training of the molecular property prediction model has ended.

5. The method for analyzing the correlation of properties of molecular substructures according to claim 1, characterized in that, The three-dimensional structural modeling tools include at least ChemDraw, Avogadro, Jmo l, Spartan, DiscoveryStudio, and Collector; the chemical information tools include at least RDK IT, KNIME, and ChemAxon. The first molecular conformation includes the first set of atoms and the first set of chemical bonds; The first set of atoms includes multiple first atoms; the parameters of the first atom object corresponding to each first atom include at least a first atom identifier, a first atom type, and first atom coordinates; the first atom identifier is the unique identifier of the current first atom; the first atom type is the chemical element type of the current first atom; and the first atom coordinates are the three-dimensional coordinates of the current first atom. The first set of chemical bonds includes multiple first chemical bonds; the first chemical bond object parameters corresponding to each first chemical bond include at least a first chemical bond identifier, a first chemical bond type, and a first chemical bond atom group; the first chemical bond identifier is a unique identifier for the current first chemical bond; the first chemical bond type is the bonding type of the current first chemical bond; the first chemical bond atom group is composed of the first atom identifiers of the two first atoms connected by the current first chemical bond; The first subconformation set includes multiple first subconformations; each first subconformation includes a first subconformation identifier, a first subconformation type, a first subconformation atom set, and a first subconformation chemical bond set; the first subconformation identifier is a unique identifier for the current first subconformation; the first subconformation type includes skeletal conformations and branched conformations; the first subconformation atom set consists of all the first atoms within the current first subconformation; the first subconformation chemical bond set consists of all the first chemical bonds within the current first subconformation. There is one and only one first sub-conformation in the first sub-conformation set whose first sub-conformation type is skeletal architecture; the first sub-conformation in the first sub-conformation set whose first sub-conformation type is skeletal architecture is denoted as the corresponding first skeletal architecture; the first sub-conformation in the first sub-conformation set whose first sub-conformation type is branched architecture is denoted as the corresponding first branched architecture. Each of the first local conformations is obtained by shaving off one of the first branch conformations from the first molecular conformation; the first local conformation includes a first shave-off conformation identifier, a first local conformation atom set, and a first local conformation chemical bond set; the first shave-off conformation identifier matches the first subconformation identifier of the shave-off first branch conformation corresponding to the current first local conformation; the first local conformation atom set consists of all the first atoms in the current first local conformation; the first local conformation chemical bond set consists of all the first chemical bonds in the current first local conformation.

6. The method for analyzing the correlation of properties of molecular substructures according to claim 5, characterized in that, The step of pruning the first molecular conformation based on each of the first branch conformations to obtain the corresponding first local conformation specifically includes: A traversal is performed on all the first branch conformations. During this traversal, the currently traversed first branch conformation is taken as the corresponding current pruning conformation. The first sub-conformation atom set of the current pruning conformation is taken as the corresponding current pruning atom set. All first atoms in the first atom set of the first molecular conformation, except for the current pruning atom set, are extracted to form the corresponding current local atom set. All first atom identifiers in the current local atom set are used to form the corresponding current atom identifier set. The two first atom identifiers of each first chemical bond atom group in the first chemical bond set of the first molecular conformation satisfy the first chemical bond of the current atom identifier set and are extracted to form the corresponding current local chemical bond set. The first sub-conformation identifier corresponding to the current pruning conformation is taken as a corresponding first pruning conformation identifier. The current local atom set and the current local chemical bond set are taken as the corresponding first local conformation atom set and the first local conformation chemical bond set. The first pruning conformation identifier, the first local conformation atom set, and the first local conformation chemical bond set corresponding to the current pruning conformation are used to form a corresponding first local conformation.

7. The method for analyzing the correlation of properties of molecular substructures according to claim 5, characterized in that, The process involves predicting the N specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations based on the molecular property prediction model, and analyzing the positive / negative correlation between all substructures and the N specified molecular properties based on the prediction results to obtain a corresponding first analysis report, which is then fed back to the current user. Specifically, this includes: Step 71: A corresponding current three-dimensional molecular conformation is formed by the first set of atoms and the first set of chemical bonds of the first molecular conformation. The molecular property prediction model predicts the N types of specified molecular properties of the current three-dimensional molecular conformation, and the predicted values ​​of each property in the molecular property prediction vector output by the molecular property prediction model are used as a corresponding global property value x. g,i 1 ≤ property index i ≤ N; and N global property values ​​x obtained. g,i Form the corresponding global property sequence; Step 72, based on each of the global property values ​​x g,i The specified molecular properties and their corresponding molecular properties form a first property data set; and the N first property data sets formed a first molecular analysis record. Step 73: A corresponding current three-dimensional molecular conformation is formed by the first sub-conformation atom set and the first sub-conformation chemical bond set of the first skeletal conformation. The molecular property prediction model predicts N types of specified molecular properties of the current three-dimensional molecular conformation, and each property prediction value of the molecular property prediction vector output by the molecular property prediction model is used as a corresponding skeleton property value x. b,i ; and the obtained N skeleton property values ​​x b,i Form the corresponding skeleton property sequence; Step 74: The total number of the first local conformations is statistically analyzed to obtain the corresponding total number of local conformations M; a current three-dimensional molecular conformation is formed by the set of atoms and chemical bonds of each first local conformation; the molecular property prediction model predicts N types of specified molecular properties of the current three-dimensional molecular conformation; and each property prediction value of the molecular property prediction vector output by the molecular property prediction model is used as a corresponding local property value x. i,j 1 ≤ local conformation index j ≤ M; and N local property values ​​x corresponding to each of the first local conformations. i,j Form the corresponding sequence of local properties; Step 75: Match the pair of global property values ​​x that have the same property index i in the global property sequence and the skeleton property sequence. g,i and the skeleton property value x b,i Form a corresponding first alignment data group; and perform one round of traversal on N first alignment data groups; during this round of traversal, take the currently traversed first alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and the skeleton property value x b,i The current global property value and the current skeleton property value are used as the corresponding values. The difference between the current skeleton property value and the current global property value is used as the corresponding first relative difference. The first relative difference is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive correlation. If it is, the corresponding first correlation is set to negative correlation. The specified molecular property, the first relative difference, and the first correlation corresponding to the current comparison data group are combined to form a corresponding first property analysis result. At the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence. The first sub-structure analysis record is formed by the first sub-conformation identifier, the first sub-conformation type, and the first property analysis result sequence corresponding to the first skeleton image. The sequence of first property analysis results is formed by sequentially sorting N first property analysis results; each of the N first property analysis results corresponds one-to-one with one of the N specified molecular properties; each first property analysis result consists of a class of specified molecular properties and their corresponding first relative difference and first correlation; the first correlation includes positive and negative correlation; the first correlation is positive when the first relative difference is greater than or equal to 0, and negative when the first relative difference is less than 0; Step 76: Each of the local property sequences is taken as the corresponding current local property sequence; and the pair of global property values ​​x that have the same property index i in the global property sequence and the current local property sequence are selected. g,i and the local property value x i,j Form a corresponding second alignment data group; and perform one round of traversal on N second alignment data groups; during this round of traversal, take the currently traversed second alignment data group as the corresponding current alignment data group; and set the global property value x of the current alignment data group. g,i and the local property value x i,j The current global property value and the current local property value are used as the corresponding values; the difference between the current global property value and the current local property value is used as the corresponding first relative difference value; the first relative difference value is checked to see if it is less than 0. If not, the corresponding first correlation is set to positive correlation; if so, the corresponding first correlation is set to negative correlation; the specified molecular property, the first relative difference value, and the first correlation value of the current comparison data group are combined to form a corresponding first property analysis result; at the end of this round of traversal, all the first property analysis results obtained in this round of traversal are sorted in order to form a corresponding first property analysis result sequence; the first branch conformation corresponding to the first scissor conformation identifier of the first local conformation of the current local property sequence is used as the corresponding current branch conformation; and the first sub-conformation identifier, the first sub-conformation type, and the first property analysis result sequence of the current branch conformation are combined to form a corresponding first substructure analysis record; Step 77: A corresponding set of first substructure analysis records is formed by the first skeleton image and 1+M first substructure analysis records corresponding to M first local images. Step 78: Extract 1+M first property analysis results corresponding to the i-th type of specified molecular property from the first substructure analysis record set to form a corresponding current property analysis result set; extract the first property analysis results with positive first correlation from the current property analysis result set to form a corresponding positive correlation set, and extract the first property analysis results with negative first correlation to form a corresponding negative correlation set; sort all the first relative differences in the positive and negative correlation sets in descending order of relative difference to obtain corresponding positive correlation difference sequences and negative correlation difference sequences; sort the first subconformation identifiers corresponding to each first relative difference in the positive correlation difference sequence according to the current sorting order of the positive correlation difference sequence to obtain a corresponding positive correlation substructure sequence; sort the first subconformation identifiers corresponding to each first relative difference in the negative correlation difference sequence according to the current sorting order of the negative correlation difference sequence to obtain a corresponding negative correlation substructure sequence; and form a corresponding first property analysis record from the i-th specified molecular property and its corresponding positive and negative correlation substructure sequences. Step 79: The obtained N first property analysis records are combined to form a corresponding first property analysis record set; Step 80: The first analysis report, composed of the obtained first molecule analysis record, first substructure analysis record set, and first property analysis record set, is fed back to the current user.

8. An apparatus for performing the method for analyzing the correlation of properties of molecular substructures as described in any one of claims 1-7, characterized in that, The device includes: a model building and training module, a molecular structure receiving module, a substructure recognition module, a local conformation processing module, and a correlation analysis module; The model building and training module is used to build a molecular property prediction model; and to train the molecular property prediction model based on a preset first dataset; the model input of the molecular property prediction model is a three-dimensional molecular conformation, and the model output is a corresponding molecular property prediction vector; the molecular property prediction model is used to predict N types of specified molecular properties of the three-dimensional molecular conformation input to the model and output the corresponding molecular property prediction vector; the total number of specified molecular properties N is a positive integer; each specified molecular property corresponds to a preset class of molecular physical or chemical properties; The molecular structure receiving module is used to receive the two-dimensional topological structure or three-dimensional molecular conformation of an arbitrary molecule input by the user as the corresponding current molecular structure after the model training is completed. The substructure recognition module is used to, when the current molecular structure is a two-dimensional topological structure, perform corresponding three-dimensional molecular conformation generation processing based on a preset three-dimensional structure modeling tool to obtain the corresponding first molecular conformation; when the current molecular structure is a three-dimensional molecular conformation, take the current molecular structure as the corresponding first molecular conformation; and use a preset chemical information tool to perform substructure recognition processing on the first molecular conformation to obtain the corresponding first subconformation set. The local conformation processing module is used to extract the unique skeleton conformation in the first sub-conformation set as the corresponding first skeleton conformation; and to extract each branch conformation in the first sub-conformation set as the corresponding first branch conformation; and to perform branch pruning on the first molecular conformation based on each first branch conformation to obtain the corresponding first local conformation. The correlation analysis module is used to predict the N types of specified molecular properties of the first molecular conformation, the first skeletal conformation, and each of the first local conformations based on the molecular property prediction model, and to analyze the positive / negative correlation between all substructures and the N types of specified molecular properties based on the prediction results, and to obtain the corresponding first analysis report and provide feedback to the current user.

9. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-7; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1-7.

Images