Method and device for processing molecular conformation generator with conditional constraints

Abstract

The embodiment of the invention relates to a processing method and device for a molecular conformation generator with conditional constraints. The method comprises the following steps: constructing a molecular conformation generator based on a plurality of Un -Mo < l > models and a property prediction model of which model training is completed; initializing model parameters of all Un -Mo < l > models in the molecular conformation generator on the basis of preset base model parameters; receiving a first molecule sequence, a first molecule property vector and a first generated molecule number input by a user; performing three-dimensional conformation conversion processing on the first molecular sequence based on a preset chemoinformatics tool to obtain a corresponding first conversion conformation; according to the first conversion conformation, the first molecule property vector, the first generated molecule number and the molecule conformation generator, batch molecule generation task processing is carried out, and a corresponding first batch molecule task report is obtained and fed back to a user. According to the method, the molecular design efficiency can be improved, and the exploration capability of unknown molecular structures is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for a molecular conformation generator with conditional constraints. Background Technology

[0002] In the fields of molecular design and drug development, generating novel molecules with specific properties is an important research direction. Currently, the conventional molecular design process can be simply divided into the following steps: a) First, design multiple new molecular structures based on expert knowledge and artificial structure adjustments, and pre-determine a set of molecular property indicators; b) Then, measure or calculate various properties of each new molecular structure (such as melting point, solubility, density, acid / basic type, stability fraction, and reactivity fraction) using experimental methods or molecular motion simulation methods; c) Compare the obtained molecular property data with the pre-determined molecular property indicators, and screen new molecular structures based on the comparison results. If no new molecular structure meets the standards, steps a and c above must be repeated until the properties of the obtained new molecular structure meet the standards. It is easy to see that this conventional molecular design method has shortcomings such as low design efficiency and insufficient ability to explore unknown chemical spaces. 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 a molecular conformation generator with conditional constraints. This invention first constructs an end-to-end molecular conformation generator based on multiple Uni-Mol models and a pre-trained property prediction model. The model parameters of all Uni-Mol models within the generator are initialized based on the model parameters of a pre-trained base Uni-Mol model. Upon receiving the user-input molecular sequence, molecular property vector, and number of molecules to be generated, cheminformatics tools (such as OpenBabel and RDKit) are used to perform a three-dimensional conformational transformation on the molecular sequence. Then, based on the transformed conformation, molecular property vector, number of generated molecules, and the molecular conformation generator, a batch molecular generation task is processed. The molecular conformation generator provided by this invention can generate a large number of molecular structures without experiments, molecular motion simulations, or expert knowledge limitations. Furthermore, it can impose directional conditional constraints on the generated molecular structures according to the set target molecular properties. This invention not only improves design efficiency but also enhances the ability to explore unknown chemical spaces.

[0004] To achieve the above objectives, a first aspect of the present invention provides a processing method for a molecular conformation generator with conditional constraints, the method comprising:

[0005] A molecular conformation generator is constructed based on multiple Uni-Mol models and a property prediction model that has already been trained; wherein, the molecular conformation generator is used to generate a molecular conformation X based on the input molecular conformation X. ini and condition constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X D ;

[0006] The model parameters of all Uni-Mol models in the molecular conformation generator are initialized based on the preset base model parameters; wherein, the base model parameters are the model parameters of a base Uni-Mol model that has already completed model pre-training.

[0007] The system receives user input of a first molecular sequence, a first molecular property vector, and a first number of generated molecules. The first molecular sequence is a one-dimensional SMIELS sequence. The first molecular property vector includes multiple first property data points. Each first property data point corresponds to a specified molecular property. The specified molecular property includes at least melting point, solubility, density, acid / basic type, stability fraction, and reactivity fraction. When each first property data point is 0, it indicates that the specified molecular property corresponding to the current property data is not constrained when generating a new molecular conformation. When the first property data is not 0, it indicates that the specified molecular property is constrained by the current property data as the corresponding conditional property data when generating a new molecular conformation.

[0008] The first molecular sequence is subjected to three-dimensional conformational transformation based on a preset cheminformatics tool to obtain the corresponding first transformed conformation; wherein the cheminformatics tool includes at least OpenBabel and RDKit;

[0009] Based on the first transformed conformation, the first molecular property vector, the first number of generated molecules, and the molecular conformation generator, a batch molecular generation task is processed to obtain a corresponding first batch molecular task report, which is then fed back to the user.

[0010] Preferably, the Uni-Mol model is used to perform molecular conformation optimization processing on the input three-dimensional molecular conformation and output the corresponding optimized molecular conformation;

[0011] The property prediction model consists of multiple parallel first prediction models; each first prediction model is based on a class of nonlinear regression models, including at least XGBoost, GBDT, Random Forest and MLP models; each first prediction model corresponds to a specific molecular property.

[0012] The property prediction model is used to send the currently input three-dimensional molecular conformation to each of the first prediction models to obtain corresponding property prediction data; and to form a corresponding property prediction vector from all the obtained property prediction data and output it; the property prediction vector is composed of multiple property prediction data.

[0013] The first prediction model in the property prediction model, which corresponds to melting point, solubility, density, acid / basic type, stability score, or reactivity score, is used to perform corresponding melting point, solubility, density, acid / basic type, stability score, or reactivity score prediction processing based on the currently input three-dimensional molecular conformation to obtain the corresponding predicted values ​​of melting point, solubility, density, acid / basic type, stability score, or reactivity score as a corresponding property prediction data and output.

[0014] Preferably, the first model input terminal of the molecular conformation generator is used to receive the molecular conformation X input from the model. ini The second model input terminal is used to receive the condition constraint vector C of the model input. T The model output is used to output the corresponding generated molecular conformation X. D ;

[0015] The molecular conformation X ini and the generated molecular conformation X D Each is a three-dimensional molecular conformation; the condition constraint vector C T It consists of multiple conditional property data, each of which corresponds to a specified molecular property;

[0016] The molecular conformation generator includes a conformation optimization model, an encoder, the property prediction model, a constraint comparison module, an encoding modulation module, and a decoder.

[0017] The input of the conformation optimization model is connected to the input of the first model, and its output is connected to the first input of the encoder; the second input of the encoder is connected to the output of the coding modulation module, its first output is connected to the input of the property prediction model, and its second output is connected to the first input of the constraint comparison module; the output of the property prediction model is connected to the second input of the constraint comparison module; the third input of the constraint comparison module is connected to the input of the second model, its first output is connected to the input of the coding modulation module, and its second output is connected to the input of the decoder; the output of the decoder is connected to the model output.

[0018] The conformation optimization model is implemented based on one of the Uni-Mol models;

[0019] The conformation optimization model is used to optimize the input molecular conformation X. ini Perform molecular conformation optimization and output the corresponding optimized molecular conformation X. O ; and the optimized molecular conformation X O Send to the encoder;

[0020] The encoder is composed of a preset first number N. E The Uni-Mol models are connected in series;

[0021] The encoder is used to receive the optimized molecular conformation X sent by the conformation optimization model. O At that time, the optimized molecular conformation X received this time... O Save it; and use N E The Uni-Mol model described above is used to evaluate the optimized molecular conformation X received in this study. O Perform stepwise optimization and use the optimized molecular conformation output by the last Uni-Mol model as the corresponding encoded output conformation X. E ; and the encoded output configuration X obtained this time E Send to the property prediction model and the constraint comparison module;

[0022] The encoder is also configured to reset its own encoder parameters using the received modulation model parameters when it receives modulation model parameters from the encoding and modulation module; and to reset its own encoder parameters again using N when the parameter reset is complete. E The Uni-Mol model is used to save the optimized molecular conformation X. O Perform step-by-step optimization and the Nth E The optimized molecular conformation output by the Uni-Mol model is used as the latest encoded output conformation X. E ; and the encoded output configuration X obtained this time E Send to the property prediction model and the constraint comparison module;

[0023] The property prediction model is used to construct X from the encoded output of the current input. E The corresponding property prediction data is sent to each of the first prediction models; and the corresponding property prediction vector C is composed of all the property prediction data obtained in this process. P ; and the property prediction vector C obtained this time P Send to the constraint comparison module;

[0024] The constraint comparison module is used to predict vector C based on the input property. P and the condition constraint vector C TConstraint comparison processing is performed to obtain the corresponding condition satisfaction states; and the condition satisfaction states are identified; if the condition satisfaction state is a condition non-satisfaction, then the latest property prediction vector C is... P and the condition constraint vector C T Send to the encoding and modulation module; if the condition is satisfied (the state is that the constraint condition is met), then send the latest encoded output configuration X. E Send to the decoder; wherein, the condition satisfaction state includes not satisfying the constraint condition and satisfying the constraint condition;

[0025] The coding and modulation module is used to predict vector C based on the property received in the current iteration. P and the condition constraint vector C T Perform a round of parameter modulation processing on the encoder to obtain the corresponding modulation model parameters; and send the obtained modulation model parameters to the encoder.

[0026] The decoder consists of a preset second number N. D The Uni-Mol models are connected in series;

[0027] The decoder is used to use N D The Uni-Mol model encodes the input and outputs the conformation X. E Perform step-by-step optimization and the Nth D The optimized molecular conformation output by the Uni-Mol model is used as the corresponding generated molecular conformation X. D And output it.

[0028] Furthermore, the prediction vector C based on the input property... P and the condition constraint vector C T The constraint comparison process is performed to obtain the corresponding condition satisfaction state, specifically including:

[0029] The constraint comparison module compares the constraint vector C. T Each non-zero conditional property data is denoted as the corresponding first alignment property data.

[0030] The system iterates through all the first alignment property data in one pass; during this pass, the currently passed first alignment property data is used as the corresponding current threshold data; the specified molecular property corresponding to the current threshold data is used as the corresponding current property; and the property prediction vector C is... PThe predicted data corresponding to the current property is used as the corresponding current predicted data; based on the current property, a preset first correspondence table is queried, and the first relative error range field of the first correspondence record matching the first property field with the current property in the first correspondence table is extracted as the corresponding current relative error range; the current predicted data and the current threshold data form a corresponding prediction-threshold data pair; the relative error between the current predicted data and the current threshold data is calculated as: current relative error = |current predicted data - current threshold data| / current threshold data; and it is identified whether the current relative error meets the corresponding current relative error range. If it does, the corresponding single-item error comparison result is set to "satisfied"; otherwise, the pair is set to "satisfied". The individual error comparison result is not satisfied; and at the end of this round of traversal, all the obtained prediction-threshold data are input into a preset error evaluation function to calculate the corresponding first evaluation value; wherein, the first correspondence table is a correspondence table reflecting the correspondence between property and relative error range, and is composed of multiple first correspondence records; each first correspondence record corresponds to a specified molecular property; the first correspondence record includes a first property field and a first relative error range field; the first property field is used to store a corresponding specified molecular property, and the first relative error range field is used to store a corresponding relative error range; the individual error comparison result includes satisfied and unsatisfied; the error evaluation function includes at least the RMSE function;

[0031] It identifies whether the first evaluation value meets the preset first evaluation value range. If it does, the corresponding full-item comparison result is set to "satisfied"; if it does not, the corresponding full-item comparison result is set to "unsatisfied". The full-item comparison result includes "satisfied" and "unsatisfied".

[0032] The system identifies all the individual error comparison results and the full comparison results. If any individual error comparison result or the full comparison result is not satisfied, the corresponding condition satisfaction state is set to not satisfying the constraint condition. If all individual error comparison results and the full comparison results are satisfied, the corresponding condition satisfaction state is set to satisfying the constraint condition.

[0033] Furthermore, the prediction vector C based on the property received at that time... P and the condition constraint vector C T The encoder undergoes one round of parameter modulation processing to obtain the corresponding modulation model parameters, specifically including:

[0034] Step 51, the coding and modulation module converts the condition constraint vector CT Each conditional property data that is 0 is recorded as the corresponding invalid property data;

[0035] Step 52, and then the property prediction vector C P The property prediction data corresponding to each of the invalid property data is reset to 0;

[0036] Step 53, and then reset the property prediction vector C. P and the condition constraint vector C T Substitute the preset model into the objective function for optimization;

[0037] The objective function for model optimization is:

[0038] θ represents the model parameters of the encoder;

[0039] L M The model loss function is a preset model loss function, L. M It should include at least the L1 loss function, the L2 loss function, and the cross-entropy loss function;

[0040] Step 54, and based on the preset model parameter optimizer, perform a round of parameter modulation processing on the model parameter θ of the encoder in the direction of minimizing the model optimization objective function, and at the end of this round of parameter modulation processing, use the model parameter θ obtained after this round of modulation as the corresponding modulation model parameter;

[0041] The model parameter optimizer includes at least the SGD optimizer, ADAM optimizer, RMSprop optimizer, AdamW optimizer, and LBGFS optimizer.

[0042] Preferably, the step of performing three-dimensional conformational transformation on the first molecule sequence based on a preset cheminformatics tool to obtain the corresponding first transformed conformation specifically includes:

[0043] The first molecular sequence is input into the first processing interface provided by the cheminformatics tool for molecular sequence standardization to obtain the corresponding first standard sequence; and the first standard sequence is input into the second processing interface provided by the cheminformatics tool for molecular sequence three-dimensional conformation transformation to obtain the corresponding first transformed conformation.

[0044] The first and second processing interfaces are two processing interfaces provided by the cheminformatics tool. The first processing interface is used to perform normalized sequence transformation on the molecular sequence input by the interface according to a preset normalized molecular sequence arrangement rule and output the normalized sequence as the interface processing result. The first processing interface outputs the same normalized sequence for different SMIELS sequences of the same molecule. The second processing interface is used to generate the corresponding three-dimensional molecular conformation according to the molecular sequence input by the interface and output the three-dimensional molecular conformation as the interface processing result.

[0045] Preferably, the step of processing the batch molecule generation task based on the first transformed conformation, the first molecular property vector, the first number of generated molecules, and the molecular conformation generator to obtain the corresponding first batch molecule task report and feeding it back to the user specifically includes:

[0046] Step 71: Initialize the first counter to 1;

[0047] Step 72: Use the first transformed conformation and the first molecular property vector as the corresponding molecular conformation X. ini and the condition constraint vector C T The molecular conformation generator is input to perform new molecular conformation generation processing to obtain the corresponding generated molecular conformation X. D As the corresponding first molecular conformation;

[0048] Step 73: Increment the first counter by 1;

[0049] Step 74: Identify whether the first counter is greater than the first number of generated molecules; if yes, proceed to step 75; if no, return to step 72.

[0050] Step 75: Identify the similarity between every two first generated molecular conformations to obtain the corresponding first similarity; and cluster all the obtained first generated molecular conformations based on the first similarity to obtain one or more corresponding sets of first conformations.

[0051] Wherein, each first conformation set consists of one or more first generated molecular conformations; when the total number of first generated molecular conformations in the first conformation set is unique, the first similarity of the unique first generated molecular conformation with any other first generated molecular conformation is greater than a preset first similarity threshold; when the total number of first generated molecular conformations in the first conformation set is not unique, the first similarity between any two first generated molecular conformations in the set is less than or equal to the first similarity threshold.

[0052] Step 76: The first batch molecular task report, consisting of the first molecular sequence, the first molecular property vector, the first number of generated molecules, and all the first conformations, is fed back to the user.

[0053] A second aspect of the present invention provides an apparatus for implementing the processing method of a molecular conformation generator with conditional constraints as described in the first aspect above. The apparatus includes: a model building module, a model initialization module, a user data receiving module, a data preprocessing module, and a batch generation task processing module.

[0054] The model building module is used to construct a molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has already been trained; wherein, the molecular conformation generator is used to generate a molecular conformation X based on the input molecular conformation X. ini and condition constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X D ;

[0055] The model initialization module is used to initialize the model parameters of all the Uni-Mol models in the molecular conformation generator based on the preset base model parameters; wherein, the base model parameters are the model parameters of a base Uni-Mol model that has already completed model pre-training;

[0056] The user data receiving module is used to receive the first molecular sequence, the first molecular property vector, and the first number of generated molecules input by the user; wherein, the first molecular sequence is a one-dimensional SMIELS sequence; the first molecular property vector includes multiple first property data; each first property data corresponds to a specified molecular property; the specified molecular property includes at least melting point, solubility, density, acid / basic type, stability fraction, and reactivity fraction; when each first property data is 0, it indicates that the specified molecular property corresponding to the current property data is not constrained when generating a new molecular conformation; when the first property data is not 0, it indicates that the specified molecular property is constrained by the current property data as the corresponding conditional property data when generating a new molecular conformation.

[0057] The data preprocessing module is used to perform three-dimensional conformational transformation on the first molecular sequence based on preset cheminformatics tools to obtain the corresponding first transformed conformation; wherein, the cheminformatics tools include at least OpenBabel and RDKit;

[0058] The batch generation task processing module is used to perform batch molecular generation task processing based on the first transformation conformation, the first molecular property vector, the first number of generated molecules, and the molecular conformation generator to obtain a corresponding first batch molecular task report and provide it to the user.

[0059] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0060] 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;

[0061] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0062] 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.

[0063] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for a molecular conformation generator with conditional constraints. As described above, this invention first constructs an end-to-end molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has already been trained. The generator initializes the model parameters of all Uni-Mol models within it based on the model parameters of a pre-trained base Uni-Mol model. Upon receiving the user-input molecular sequence, molecular property vector, and number of molecules to be generated, cheminformatics tools (such as OpenBabel and RDKit) are used to perform a three-dimensional conformational transformation on the molecular sequence. Then, based on the transformed conformation, molecular property vector, number of generated molecules, and the molecular conformation generator, a batch molecular generation task is processed. The molecular conformation generator provided by this invention can generate a large number of molecular structures without experiments, molecular motion simulations, or expert knowledge limitations. It can also impose directional conditional constraints on the generated molecular structures based on set target molecular properties. This invention not only effectively improves the design efficiency of molecular design but also significantly enhances the ability to explore unknown and novel molecular structures. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of a processing method for a molecular conformation generator with conditional constraints provided in Embodiment 1 of the present invention;

[0065] Figure 2This is a schematic diagram of the molecular conformation generator provided in Embodiment 1 of the present invention;

[0066] Figure 3 This is a module structure diagram of a processing device for a molecular conformation generator with conditional constraints provided in Embodiment 2 of the present invention;

[0067] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0068] 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.

[0069] Embodiment 1 of the present invention provides a processing method for a molecular conformation generator with conditional constraints, such as... Figure 1 The schematic diagram shows a processing method for a molecular conformation generator with conditional constraints provided in Embodiment 1 of the present invention. This method mainly includes the following steps:

[0070] Step 1: Construct a molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has already been trained.

[0071] Here, the molecular conformation generator in this embodiment of the invention is used to generate a molecular conformation X based on the input molecular conformation X. ini and condition constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X. D .

[0072] It should be noted that the Uni-Mol model in this embodiment of the invention is used to perform molecular conformation optimization processing on the input three-dimensional molecular conformation and output the corresponding optimized molecular conformation. This model is an end-to-end three-dimensional molecular conformation optimization model based on the transformer model. The details of its model design can be understood through the publicly available technical document "UNI-MOL: A UNIVERSAL 3D MOLECULAR REPRESENTATION LEARNING FRAMEWORK". As can be seen from this technical document, a new three-dimensional molecular conformation can be generated based on the input three-dimensional molecular conformation using the Uni-Mol model. The pre-training scheme of this model can also be completed with reference to the pre-training scheme given in the aforementioned technical document. In the following, the Uni-Mol model that has completed pre-training is referred to as the base Uni-Mol model, and the model parameters of the base Uni-Mol model are referred to as the base model parameters.

[0073] It should be noted that the property prediction model in this embodiment of the invention consists of multiple parallel first prediction models; each first prediction model is based on a class of nonlinear regression models, including at least XGBoost, GBDT, Random Forest, and MLP models. Each first prediction model in this embodiment corresponds to a specified molecular property; this specified molecular property includes at least melting point, solubility, density, acid / basic type, stability fraction, and reactivity fraction. The property prediction model in this embodiment is used to send the currently input three-dimensional molecular conformation to each first prediction model to obtain corresponding property prediction data; and to output a corresponding property prediction vector composed of all obtained property prediction data; here, the property prediction vector is composed of multiple property prediction data. In the property prediction model of this invention embodiment, the first prediction model corresponding to melting point, solubility, density, acid / basic type, stability fraction, or reactivity fraction is used to perform corresponding melting point, solubility, density, acid / basic type, stability fraction, or reactivity fraction prediction processing based on the currently input three-dimensional molecular conformation to obtain the corresponding predicted values ​​of melting point, solubility, density, acid / basic type, stability fraction, or reactivity fraction as corresponding property prediction data and output.

[0074] It should be noted that the property prediction model in this embodiment of the invention is a mature model that has already completed model training. Its training method adopts conventional supervised training. The technical solution for training this property prediction model is not part of the technical solution of this invention. Therefore, the training process of the property prediction model will not be described in detail.

[0075] It should be noted that, as Figure 2As shown in the schematic diagram of the molecular conformation generator provided in Embodiment 1 of the present invention, the first model input terminal of the molecular conformation generator of the present invention is used to receive the molecular conformation X input by the model. ini The second model input terminal is used to receive the condition constraint vector C of the model input. T The model output is used to output the corresponding generated molecular conformation X. D The molecular conformation X here ini and the generated molecular conformation X D Each represents a three-dimensional molecular conformation; the condition constraint vector C T It consists of multiple conditional property data, each of which corresponds to a specific molecular property.

[0076] like Figure 2 As shown, the molecular conformation generator of this embodiment of the invention consists of a conformation optimization model, an encoder, a property prediction model, a constraint comparison module, an encoding modulation module, and a decoder. The internal connections of the molecular conformation generator are as follows: the input of the conformation optimization model is connected to the first model input, and its output is connected to the first input of the encoder; the second input of the encoder is connected to the output of the encoding modulation module, its first output is connected to the input of the property prediction model, and its second output is connected to the first input of the constraint comparison module; the output of the property prediction model is connected to the second input of the constraint comparison module; the third input of the constraint comparison module is connected to the second model input, its first output is connected to the input of the encoding modulation module, and its second output is connected to the input of the decoder; the output of the decoder is connected to the model output.

[0077] The functions of each component in this molecular conformation generator are described below:

[0078] 1) Conformation optimization model:

[0079] The conformation optimization model of the molecular conformation generator in this embodiment of the invention is based on a Uni-Mol model;

[0080] This conformation optimization model is used to optimize the input molecular conformation X. ini Perform molecular conformation optimization and output the corresponding optimized molecular conformation X. O ; and will optimize the molecular conformation X O Send to the encoder;

[0081] 2) Encoder:

[0082] The encoder of the molecular conformation generator in this embodiment of the invention is composed of a preset first number N. E The Uni-Mol models are connected in series; the first quantity N here is... E It is a pre-set positive integer;

[0083] This encoder is used to process the optimized molecular conformation X sent by the conformation optimization model. O At that time, the optimized molecular conformation X received this time... O Save it; and use N E The Uni-Mol model was used to optimize the molecular conformation X received in this study. O Perform stepwise optimization and use the optimized molecular conformation output by the last Uni-Mol model as the corresponding encoded output conformation X. E ; and the encoded output configuration X obtained this time E Send to the property prediction model and constraint comparison module;

[0084] The encoder is also used to reset its own encoder parameters using the received modulation model parameters when it receives modulation model parameters from the encoding and modulation module; and to reset its own encoder parameters again using N when the parameter reset is complete. E A Uni-Mol model for the preserved optimized molecular conformation X O Perform step-by-step optimization and the Nth E The optimized molecular conformation output by the Uni-Mol model is used as the latest encoded output conformation X. E ; and the encoded output configuration X obtained this time E Send to the property prediction model and constraint comparison module;

[0085] 3) Property prediction model:

[0086] The property prediction model of the molecular conformation generator in this embodiment of the invention is used to encode the current input into the output conformation X. E The corresponding property prediction data is sent to each of the first prediction models; and the corresponding property prediction vector C is composed of all the property prediction data obtained in this process. P ; and the property prediction vector C obtained this time P Send to the constraint comparison module;

[0087] 4) Constraint comparison module:

[0088] The constraint comparison module of the molecular conformation generator in this embodiment of the invention is used to predict vector C based on the input properties. P and condition constraint vector C T Constraint comparison is performed to obtain the corresponding condition satisfaction states; and the condition satisfaction states are identified; if a condition satisfaction state is a condition non-satisfaction state, then the latest property prediction vector C is updated. P and condition constraint vector C T Send to the encoding and modulation module; if the condition is met and the state satisfies the constraints, then send the latest encoded output configuration X. E Send to the decoder;

[0089] Among them, the condition satisfaction state includes not satisfying the constraint condition and satisfying the constraint condition;

[0090] Furthermore, this constraint comparison module predicts vector C based on the properties of the input. P and condition constraint vector C T The specific processing steps for obtaining the corresponding condition satisfaction state by comparing constraints consist of the following steps A1-A4:

[0091] Step A1, the constraint comparison module compares the constraint vector C. T Each non-zero conditional property data is denoted as the corresponding first alignment property data;

[0092] Step A2 involves iterating through all the first alignment property data; during this iteration, the currently iterated first alignment property data is used as the corresponding current threshold data; the specified molecular property corresponding to the current threshold data is used as the corresponding current property; and the property prediction vector C is... P The property prediction data corresponding to the current property is used as the corresponding current prediction data. Based on the current property, a preset first correspondence table is queried, and the first relative error range field of the first correspondence record that matches the first property field with the current property is extracted as the corresponding current relative error range. A corresponding prediction-threshold data pair is formed by the current prediction data and the current threshold data. The relative error is calculated as |current prediction data - current threshold data| / current threshold data. It is then identified whether the current relative error meets the corresponding current relative error range. If it does, the corresponding single error comparison result is set to "satisfied"; otherwise, it is set to "unsatisfied". At the end of this round of traversal, all the obtained prediction-threshold data pairs are substituted into a preset error evaluation function to calculate the corresponding first evaluation value.

[0093] The first correspondence table is a correspondence table reflecting the correspondence between properties and relative error ranges, consisting of multiple first correspondence records; each first correspondence record corresponds to a specified molecular property; the first correspondence record includes a first property field and a first relative error range field; the first property field is used to store a corresponding specified molecular property, and the first relative error range field is used to store a corresponding relative error range; the single error comparison result includes satisfaction and non-satisfaction; the error evaluation function includes at least the RMSE function;

[0094] Step A3, and identify whether the first evaluation value meets the preset first evaluation value range. If it does, set the corresponding full comparison result to meet; if it does not, set the corresponding full comparison result to not meet.

[0095] The first evaluation value range is a pre-set evaluation value range; the full comparison result includes both "meets" and "does not meet";

[0096] Step A4 involves identifying all individual error comparison results and the overall comparison results. If any individual error comparison result or the overall comparison result is not satisfied, the corresponding condition satisfaction state is set to "not satisfied constraint condition". If all individual error comparison results and the overall comparison results are satisfied, the corresponding condition satisfaction state is set to "satisfied constraint condition".

[0097] 5) Encoding and modulation module:

[0098] The encoding and modulation module of the molecular conformation generator in this embodiment of the invention is used to predict the vector C based on the property received at that time. P and condition constraint vector C T Perform one round of parameter modulation processing on the encoder to obtain the corresponding modulation model parameters; and send the obtained modulation model parameters to the encoder.

[0099] Furthermore, the coding and modulation module predicts the vector C based on the properties received at that time. P and condition constraint vector C T The specific processing steps for obtaining the corresponding modulation model parameters by performing one round of parameter modulation processing on the encoder consist of the following steps B1-B4:

[0100] Step B1, the coding and modulation module converts the condition constraint vector C T Each conditional property data that is 0 is recorded as the corresponding invalid property data;

[0101] Step B2, and the property prediction vector C P The property prediction data corresponding to each invalid property data is reset to 0;

[0102] Step B3, and complete the reset of the property prediction vector C. P and condition constraint vector C T Substitute the preset model into the objective function for optimization;

[0103] Here, the objective function for model optimization in this embodiment of the invention is:

[0104]

[0105] Where θ represents the encoder's model parameters; L MThe model loss function is a preset value, L. M It should include at least the L1 loss function, the L2 loss function, and the cross-entropy loss function;

[0106] Step B4, and based on the preset model parameter optimizer, perform a round of parameter modulation processing on the encoder model parameter θ in the direction of minimizing the model optimization objective function, and at the end of this round of parameter modulation processing, use the model parameter θ obtained after this round of modulation as the corresponding modulation model parameter;

[0107] Among them, the model parameter optimizers include at least the SGD optimizer, ADAM optimizer, RMSprop optimizer, AdamW optimizer, and LBGFS optimizer;

[0108] 6) Decoder:

[0109] The decoder of the molecular conformation generator in this embodiment of the invention is composed of a preset second number N. D The Uni-Mol models are connected in series; the second quantity N here is... D Let N be a pre-set positive integer. D <N E ;

[0110] This decoder is used to use N D A Uni-Mol model encodes the input and outputs the conformation X. E Perform step-by-step optimization and the Nth D The optimized molecular conformation output by each Uni-Mol model is used as the corresponding generated molecular conformation X. D And output it.

[0111] Step 2: Initialize the model parameters of all Uni-Mol models in the molecular conformation generator based on the preset base model parameters.

[0112] Here, the base model parameters of this embodiment are the model parameters of a base Uni-Mol model that has already completed model pre-training.

[0113] Step 3: Receive the user input of the first molecule sequence, the first molecule property vector, and the first number of generated molecules.

[0114] Here, the first molecular sequence is a one-dimensional SMIELS sequence; the first molecular property vector includes multiple first property data; each first property data corresponds to a specified molecular property; when each first property data is 0, it means that the specified molecular property corresponding to the current property data is not constrained when generating a new molecular conformation; when the first property data is not 0, it means that the specified molecular property is constrained by the current property data as the corresponding conditional property data when generating a new molecular conformation; the number of first generated molecules is a positive integer.

[0115] Step 4: Based on the preset cheminformatics tools, perform three-dimensional conformational transformation on the first molecule sequence to obtain the corresponding first transformed conformation;

[0116] Among them, cheminformatics tools include at least OpenBabel and RDKit;

[0117] Specifically, this includes: inputting the first molecular sequence into the first processing interface provided by the cheminformatics tool for molecular sequence standardization to obtain the corresponding first standard sequence; and inputting the first standard sequence into the second processing interface provided by the cheminformatics tool for molecular sequence three-dimensional conformation transformation to obtain the corresponding first transformed conformation.

[0118] Here, the first and second processing interfaces of this embodiment are two processing interfaces provided by cheminformatics tools; wherein: 1) the first processing interface is used to perform normalized sequence conversion on the molecular sequence input by the interface according to a preset normalized molecular sequence arrangement rule and output the obtained normalized sequence as the interface processing result. It should be noted that the first processing interface will output the same normalized sequence for different SMIELS sequence expressions of the same molecule; 2) the second processing interface is used to generate the corresponding three-dimensional molecular conformation according to the molecular sequence input by the interface and output the obtained three-dimensional molecular conformation as the interface processing result.

[0119] Step 5: Based on the first transformation conformation, the first molecular property vector, the first number of generated molecules, and the molecular conformation generator, perform batch molecular generation task processing to obtain the corresponding first batch molecular task report and provide feedback to the user;

[0120] Specifically, this includes: Step 51, initializing the first counter to 1;

[0121] Step 52: Take the first transformation conformation and the first molecular property vector as the corresponding molecular conformation X. ini and condition constraint vector C T The input molecular conformation generator performs a new molecular conformation generation process to obtain the corresponding generated molecular conformation X. D As the corresponding first molecular conformation;

[0122] Step 53: Increment the first counter by 1;

[0123] Step 54: Identify whether the first counter is greater than the number of first generated molecules; if yes, proceed to step 55; if no, return to step 52.

[0124] Step 55: Identify the similarity between every two first generated molecular conformations to obtain the corresponding first similarity; and cluster all the obtained first generated molecular conformations based on the first similarity to obtain one or more corresponding sets of first conformations.

[0125] Each first conformation set consists of one or more first generated molecular conformations. When the total number of first generated molecular conformations in the first conformation set is unique, the first similarity of the unique first generated molecular conformation with any other first generated molecular conformation is greater than a preset first similarity threshold. When the total number of first generated molecular conformations in the first conformation set is not unique, the first similarity between any two first generated molecular conformations in the set is less than or equal to the first similarity threshold.

[0126] Here, in this embodiment of the invention, when identifying the similarity between every two first generated molecular conformations, the conformation vector of each first generated molecular conformation is first transformed to obtain the corresponding first conformation vector, and then the similarity between every two first conformation vectors is calculated based on conventional vector similarity algorithms (such as cosine similarity algorithm) to obtain the corresponding first similarity.

[0127] Step 56: The first batch of molecular task reports, consisting of the first molecular sequence, the first molecular property vector, the first number of generated molecules, and the set of all first conformations, is fed back to the user.

[0128] Figure 3 This is a module structure diagram of a processing device for a molecular conformation generator with conditional constraints provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiments, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building module 201, a model initialization module 202, a user data receiving module 203, a data preprocessing module 204, and a batch generation task processing module 205.

[0129] Model building module 201 is used to construct a molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has already been trained; wherein, the molecular conformation generator is used to generate a molecular conformation X based on the input molecular conformation X. ini and condition constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X D .

[0130] The model initialization module 202 is used to initialize the model parameters of all Uni-Mol models in the molecular conformation generator based on the preset base model parameters; wherein, the base model parameters are the model parameters of a base Uni-Mol model that has already completed model pre-training.

[0131] The user data receiving module 203 is used to receive the first molecular sequence, the first molecular property vector, and the first number of generated molecules input by the user. The first molecular sequence is a one-dimensional SMIELS sequence. The first molecular property vector includes multiple first property data. Each first property data corresponds to a specified molecular property. The specified molecular property includes at least melting point, solubility, density, acid / basic type, stability fraction, and reactivity fraction. When each first property data is 0, it means that the specified molecular property corresponding to the current property data is not constrained when generating a new molecular conformation. When the first property data is not 0, it means that the specified molecular property is constrained by the current property data as the corresponding conditional property data when generating a new molecular conformation.

[0132] The data preprocessing module 204 is used to perform three-dimensional conformational transformation on the first molecular sequence based on preset cheminformatics tools to obtain the corresponding first transformed conformation; wherein, the cheminformatics tools include at least OpenBabel and RDKit.

[0133] The batch generation task processing module 205 is used to process batch molecular generation tasks based on the first transformation conformation, the first molecular property vector, the first number of generated molecules, and the molecular conformation generator to obtain the corresponding first batch molecular task report and provide feedback to the user.

[0134] The processing device for a molecular conformation generator with conditional constraints provided in this embodiment of the invention can execute the method steps in the above method embodiment. Its implementation principle and technical effect are similar, and will not be repeated here.

[0135] 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 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.

[0136] 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).

[0137] 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)).

[0138] 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 involved in 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.

[0139] 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.

[0140] 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.

[0141] 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.

[0142] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for a molecular conformation generator with conditional constraints. As described above, this invention first constructs an end-to-end molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has already been trained. The generator initializes the model parameters of all Uni-Mol models within it based on the model parameters of a pre-trained base Uni-Mol model. Upon receiving the user-input molecular sequence, molecular property vector, and number of molecules to be generated, cheminformatics tools (such as OpenBabel and RDKit) are used to perform a three-dimensional conformational transformation on the molecular sequence. Then, based on the transformed conformation, molecular property vector, number of generated molecules, and the molecular conformation generator, a batch molecular generation task is processed. The molecular conformation generator provided by this invention can generate a large number of molecular structures without experiments, molecular motion simulations, or expert knowledge limitations. It can also impose directional conditional constraints on the generated molecular structures based on set target molecular properties. This invention not only effectively improves the design efficiency of molecular design but also significantly enhances the ability to explore unknown and novel molecular structures.

[0143] 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.

[0144] 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 processing method for a molecular conformation generator with conditional constraints, characterized in that: The method comprises: A molecular conformation generator is constructed based on multiple Uni-Mo l models and a property prediction model that has completed model training; wherein the molecular conformation generator is used to generate a molecular conformation X according to the input molecular conformation X. ini and the conditional constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X D ; Initializing the model parameters of all the Uni-Mol models in the molecular conformation generator based on the preset base model parameters; wherein the base model parameters are the model parameters of a base Uni-Mol model that has completed model pre-training; Receive a first molecular sequence, a first molecular property vector and a first number of generated molecules input by a user; wherein the first molecular sequence is a one-dimensional SMIELS sequence; the first molecular property vector includes a plurality of first property data; each of the first property data corresponds to a specified molecular property; the specified molecular properties at least include melting point, solubility, density, acid / alkaline type, stability score and reactivity score; when each of the first property data is 0, it indicates that when generating a new molecular conformation, no constraint is imposed on the specified molecular property corresponding to the current property data; when the first property data is not 0, it indicates that when generating a new molecular conformation, the current property data is used as the corresponding conditional property data to constrain the corresponding specified molecular property; Based on a preset chemical informatics tool, a three-dimensional conformation conversion process is performed on the first molecular sequence to obtain a corresponding first conversion conformation; wherein the chemical informatics tool at least includes OpenBabel and RDKit; Batch molecule generation task processing is performed according to the first conversion conformation, the first molecular property vector, the first number of generated molecules and the molecular conformation generator to obtain a corresponding first batch molecule task report and feed it back to the user.

2. The processing method of the molecular conformation generator with conditional constraints according to claim 1, characterized in that: The Uni-Mol model is used to perform molecular conformation optimization processing on the input three-dimensional molecular conformation and output the corresponding optimized molecular conformation; The property prediction model is composed of a plurality of parallel first prediction models; each of the first prediction models is implemented based on a type of nonlinear regression model, and the nonlinear regression model includes at least an XGBoost model, a GBDT model, a RandomForest model and an MLP model; each of the first prediction models corresponds to one of the specified molecular properties; The property prediction model is used to send the currently input three-dimensional molecular conformation to each of the first prediction models to obtain corresponding property prediction data; and all the obtained property prediction data are used to form a corresponding property prediction vector and output it; the property prediction vector is composed of a plurality of the property prediction data; The first prediction model corresponding to the melting point, solubility, density, acid / alkalinity type, stability score or reaction activity score in the property prediction model is used to perform corresponding melting point, solubility, density, acid / alkalinity type, stability score or reaction activity score prediction processing according to the currently input three-dimensional molecular conformation to obtain the corresponding melting point, solubility, density, acid / alkalinity type, stability score or reaction activity score prediction value as a corresponding property prediction data and output it.

3. The processing method of the molecular conformation generator with conditional constraints according to claim 2, characterized in that: The first model input terminal of the molecular conformation generator is used to receive the molecular conformation X input by the model. ini The second model input terminal is used to receive the conditional constraint vector C of the model input T The model output terminal is used to output the corresponding generated molecular conformation X D ; The molecular conformation X ini and the generated molecular conformation X D Each is a three-dimensional molecular conformation; the conditional constraint vector C T It is composed of a plurality of the conditional property data, each of which corresponds to one of the specified molecular properties; The molecular conformation generator includes a conformation optimization model, an encoder, the property prediction model, a constraint comparison module, a coding modulation module and a decoder; The input end of the conformation optimization model is connected to the input end of the first model, and the output end is connected to the first input end of the encoder; the second input end of the encoder is connected to the output end of the coding and modulation module, the first output end is connected to the input end of the property prediction model, and the second output end is connected to the first input end of the constraint comparison module; the output end of the property prediction model is connected to the second input end of the constraint comparison module; the third input end of the constraint comparison module is connected to the second model input end, the first output end is connected to the input end of the coding and modulation module, and the second output end is connected to the input end of the decoder; the output end of the decoder is connected to the output end of the model; The conformation optimization model is implemented based on the Uni-Mol model; The conformation optimization model is used to input the molecular conformation X ini Perform molecular conformation optimization and output the corresponding optimized molecular conformation X O ; and the optimized molecular conformation X O Sending to the encoder; The encoder is composed of a preset first number N E The Uni-Mol model is connected in series; The encoder is used to receive the optimized molecular conformation X sent by the conformation optimization model O When the optimized molecular conformation X received this time is O to save; and use N E The Uni-Mol model is used to optimize the molecular conformation X received this time. O Perform step-by-step optimization and use the optimized molecular conformation output by the last Uni-Mol model as the corresponding coded output conformation X E ; And the encoding output conformation X obtained this time E Sending to the property prediction model and the constraint condition comparison module; The encoder is also used to reset its own encoder parameters using the modulation model parameters received this time when receiving the modulation model parameters sent by the coding modulation module; and to use N again when the parameter reset is completed. E The Uni-Mol model is used to save the optimized molecular conformation X O Perform step-by-step optimization and set the Nth E The optimized molecular conformation output by the Uni-Mol model is used as the latest encoded output conformation X E ; And the encoding output conformation X obtained this time E Sending to the property prediction model and the constraint condition comparison module; The property prediction model is used to convert the encoding output conformation X of the current input E Send the corresponding property prediction data to each of the first prediction models; and form a corresponding property prediction vector C from all the property prediction data obtained this time P ; and the property prediction vector C obtained this time P Sending to the constraint comparison module; The constraint condition comparison module is used to predict the vector C according to the input property P and the conditional constraint vector C T Perform constraint condition comparison processing to obtain the corresponding condition satisfaction state; and identify the condition satisfaction state; if the condition satisfaction state is that the constraint condition is not satisfied, the latest property prediction vector C P and the conditional constraint vector C T Send to the coding and modulation module; if the condition is satisfied, the latest coding output constellation X E Sending to the decoder; wherein the condition satisfaction state includes not satisfying the constraint condition and satisfying the constraint condition; The coding and modulation module is used to predict the vector C according to the property received at that time P and the conditional constraint vector C T Performing a round of parameter modulation processing on the encoder to obtain the corresponding modulation model parameters; and sending the modulation model parameters obtained this time to the encoder; The decoder is composed of a preset second number N D The Uni-Mo l models are connected in series; The decoder is used to use N D The Uni-Mol model encodes the input output conformation X E Perform step-by-step optimization and set the Nth D The optimized molecular conformation output by the Uni-Mol model is used as the corresponding generated molecular conformation X D And output.

4. The processing method of the molecular conformation generator with conditional constraints according to claim 3, characterized in that: The prediction vector C according to the property of the input P and the conditional constraint vector C T The constraint condition comparison process is performed to obtain the corresponding condition satisfaction status, which specifically includes: The constraint condition comparison module converts the condition constraint vector C T Each conditional property data that is not 0 is recorded as the corresponding first comparison property data; And perform a round of traversal on all the first comparison property data; and in this round of traversal, use the first comparison property data currently traversed as the corresponding current threshold data; and use the specified molecular property corresponding to the current threshold data as the corresponding current property; and use the property prediction vector C P The property prediction data corresponding to the current property is used as the corresponding current prediction data; and based on the current property, a preset first correspondence table is queried, and the first relative error range field of the first correspondence record in which the first property field in the first correspondence table matches the current property is extracted as the corresponding current relative error range; and a corresponding prediction-threshold data pair is formed by the current prediction data and the current threshold data; and a relative error calculation is performed on the current prediction data and the current threshold data, and the corresponding current relative error = |current prediction data-current threshold data| / current threshold data; and whether the current relative error satisfies the corresponding current relative error range is identified, and if so, the corresponding single error comparison result is set to be satisfied, and if not, the single error comparison result is set to be satisfied. The corresponding single error comparison result is not satisfied; and at the end of this round of traversal, all the obtained prediction-threshold data pairs are brought into the preset error evaluation function to calculate the corresponding first evaluation value; wherein, the first correspondence table is a correspondence table reflecting the property-relative error range correspondence, which is composed of multiple first correspondence records; each first correspondence record corresponds to a specified molecular property; the first correspondence record includes the first property field and the first relative error range field; the first property field is used to store a corresponding specified molecular property, and the first relative error range field is used to store a corresponding relative error range; the single error comparison results include satisfied and not satisfied; the error evaluation function at least includes the RMSE function; and identifying whether the first evaluation value satisfies a preset first evaluation value range, and if so, setting the corresponding full comparison result as satisfied, and if not, setting the corresponding full comparison result as unsatisfied; wherein the full comparison result includes satisfied and unsatisfied; And all the obtained single error comparison results and the full comparison results are identified; if any of the single error comparison results is not satisfied or the full comparison result is not satisfied, the corresponding condition satisfaction state is set to not satisfying the constraint condition; if all the single error comparison results and the full comparison results are satisfied, the corresponding condition satisfaction state is set to satisfying the constraint condition.

5. The processing method of the molecular conformation generator with conditional constraints according to claim 3, characterized in that: The prediction vector C based on the property received at that time P and the conditional constraint vector C T Performing a round of parameter modulation processing on the encoder to obtain the corresponding modulation model parameters specifically includes: Step 51: the coding and modulation module converts the conditional constraint vector C T Each of the conditional property data that is 0 is recorded as the corresponding invalid property data; Step 52, and transform the property prediction vector C P The property prediction data corresponding to each of the invalid property data is reset to 0; Step 53, and complete the reset of the property prediction vector C P and the conditional constraint vector C T Bring in the preset model optimization objective function; Among them, the model optimization objective function is: θ is the model parameter of the encoder; L M is a preset model loss function, wherein the model loss function L M At least including L1 loss function, L2 loss function and cross entropy loss function; Step 54, based on a preset model parameter optimizer, a round of parameter modulation processing is performed on the model parameter θ of the encoder in a direction in which the model optimization objective function reaches a minimum value, and at the end of this round of parameter modulation processing, the model parameter θ obtained after this round of modulation is used as the corresponding modulation model parameter; Among them, the model parameter optimizer includes at least SGD optimizer, ADAM optimizer, RMSprop optimizer, AdamW optimizer, and LBGFS optimizer.

6. The processing method of the molecular conformation generator with conditional constraints according to claim 1, characterized in that: The performing three-dimensional conformation conversion processing on the first molecular sequence based on a preset chemical informatics tool to obtain a corresponding first conversion conformation specifically includes: Inputting the first molecular sequence into a first processing interface provided by the chemical informatics tool to perform molecular sequence standardization processing to obtain a corresponding first standard sequence; and inputting the first standard sequence into a second processing interface provided by the chemical informatics tool to perform molecular sequence three-dimensional conformation conversion processing to obtain the corresponding first conversion conformation; Among them, the first and second processing interfaces are two processing interfaces provided by the chemical informatics tool; the first processing interface is used to perform standardized sequence conversion on the molecular sequence input into the interface according to a preset standardized molecular sequence arrangement rule and output the obtained standardized sequence as the interface processing result; the first processing interface outputs the same standardized sequence for different SMIELS sequence expressions of the same molecule; the second processing interface is used to generate a corresponding three-dimensional molecular conformation according to the molecular sequence input into the interface and output the obtained three-dimensional molecular conformation as the interface processing result.

7. The processing method of the molecular conformation generator with conditional constraints according to claim 1, characterized in that: The performing batch molecule generation task processing according to the first conversion conformation, the first molecular property vector, the first number of generated molecules and the molecular conformation generator to obtain a corresponding first batch molecule task report and feeding back to the user specifically includes: Step 71, initializing the first counter to 1; Step 72: taking the first conversion conformation and the first molecular property vector as the corresponding molecular conformation X ini and the conditional constraint vector C T Input the molecular conformation generator to perform new molecular conformation generation processing to obtain the corresponding generated molecular conformation X D As the corresponding first generated molecular conformation; Step 73, adding 1 to the first counter; Step 74, identifying whether the first counter is greater than the first generated molecule quantity; if so, go to step 75; if not, return to step 72; Step 75, identifying the similarity between every two of the first generated molecular conformations to obtain a corresponding first similarity; and clustering all the first generated molecular conformations obtained based on the first similarity to obtain one or more corresponding first conformation sets; Wherein, each first conformation set is composed of one or more first generated molecular conformations; when the total number of the first generated molecular conformations in the first conformation set is unique, the first similarity between the unique first generated molecular conformation and any other first generated molecular conformations is greater than a preset first similarity threshold; when the total number of the first generated molecular conformations in the first conformation set is not unique, the first similarity between every two first generated molecular conformations in the set is less than or equal to the first similarity threshold; Step 76: Feedback to the user the first batch of molecular task report corresponding to the first molecular sequence, the first molecular property vector, the first generated molecular quantity and all the first conformation sets.

8. A device for executing the processing method of the molecular conformation generator with conditional constraints according to any one of claims 1 to 7, characterized in that: The device comprises: a model building module, a model initialization module, a user data receiving module, a data preprocessing module and a batch generation task processing module; The model building module is used to build a molecular conformation generator based on multiple Uni-Mol models and a property prediction model that has completed model training; wherein the molecular conformation generator is used to generate a molecular conformation X according to the input molecular conformation X. ini and the conditional constraint vector C T Perform new molecular conformation generation processing and output the corresponding generated molecular conformation X D ; The model initialization module is used to initialize the model parameters of all the Uni-Mol models in the molecular conformation generator based on the preset base model parameters; wherein the base model parameters are the model parameters of a base Uni-Mol model that has completed model pre-training; The user data receiving module is used to receive a first molecular sequence, a first molecular property vector and a first number of generated molecules input by a user; wherein the first molecular sequence is a one-dimensional SMIELS sequence; the first molecular property vector includes a plurality of first property data; each of the first property data corresponds to a specified molecular property; the specified molecular properties include at least melting point, solubility, density, acid / alkaline type, stability score and reactivity score; when each of the first property data is 0, it indicates that when generating a new molecular conformation, the specified molecular property corresponding to the current property data is not constrained, and when the first property data is not 0, it indicates that when generating a new molecular conformation, the corresponding specified molecular property is constrained with the current property data as the corresponding conditional property data; The data preprocessing module is used to perform three-dimensional conformation conversion processing on the first molecular sequence based on a preset chemical informatics tool to obtain a corresponding first conversion conformation; wherein the chemical informatics tool at least includes OpenBabel and RDK it; The batch generation task processing module is used to perform batch molecule generation task processing according to the first conversion conformation, the first molecular property vector, the first number of generated molecules and the molecular conformation generator to obtain a corresponding first batch molecule task report and feedback it to the user.

9. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 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, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 7.

Images