Drug Molecule Property Prediction and Classification Method and System Based on BiLSTM

Abstract

The present invention discloses a drug molecular property prediction and classification method and system based on BiLSTM; wherein the method includes: obtaining a drug molecule to be predicted, and converting the molecular structure of the drug analysis to obtain a drug molecular descriptor and a drug molecular fingerprint; inputting both the drug molecular descriptor and the drug molecular fingerprint into a trained drug molecular property prediction and classification network, and outputting a drug molecular property prediction result; the working principle of the trained drug molecular property prediction and classification network includes: respectively extracting features from the drug molecular descriptor and the drug molecular fingerprint to obtain two drug molecular feature vectors; performing feature fusion on the two drug molecular feature vectors; performing classification prediction on the fused feature vectors, and outputting a drug molecular property prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of drug screening, and particularly to a method and system for predicting and classifying drug molecular properties based on BiLSTM. Background Art

[0002] The statements in this section merely mention the background art related to the present invention and do not necessarily constitute prior art.

[0003] The discovery of new drugs or the re - utilization of drugs is a popular task in the field of biochemistry. Predicting the effectiveness or toxicity of drugs through molecular properties plays an important role in this task. In recent years, with the development of machine learning, especially the emergence of deep learning, many methods have achieved better performance in this task. Drug molecules or compounds are converted into computer - recognizable formats, such as molecular graphs, molecular fingerprints, and molecular descriptors. These readable forms of molecular information are extracted by various means, including deep learning, to form unique molecular features, which can be used to achieve subsequent classification or prediction tasks.

[0004] In the process of implementing the present disclosure, the inventors found the following technical problems in the prior art:

[0005] (1) Different types of molecular descriptors and molecular fingerprints are designed for specific tasks and do not have universality. They often have mediocre effects when dealing with complex and changeable tasks.

[0006] (2) In common deep - learning networks, the feature - extraction part cannot completely extract all information when facing drug molecules, which leads to a reduction in the accuracy during subsequent classification.

[0007] (3) When using mainstream complex classification models, overfitting occurs, but the classification effect of traditional classification models is generally mediocre. Summary of the Invention

[0008] To solve the deficiencies of the prior art, the present invention provides a method and system for predicting and classifying drug molecular properties based on BiLSTM; the drug molecular property classification results obtained by the solution are more accurate, which is helpful for subsequent biochemical experiments such as drug screening and has a certain practicality.

[0009] In the first aspect, the present invention provides a method for predicting and classifying drug molecular properties based on BiLSTM;

[0010] The method for predicting and classifying drug molecular properties based on BiLSTM includes:

[0011] Obtain the drug molecule to be predicted, and convert the molecular structure of the drug molecule to obtain a drug molecular descriptor and a drug molecular fingerprint;

[0012] Both the drug molecule descriptors and the drug molecule fingerprints are input into the trained drug molecule property prediction classification network, and the drug molecule property prediction results are output;

[0013] Among them, the network structure of the trained drug molecule property prediction classification network includes: a feature extraction part, a feature fusion part, and a feature classification part connected in sequence; the feature extraction part includes two parallel branches: the first branch and the second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from drug molecule descriptors; the second branch is used to extract features from drug molecule fingerprints; the first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence; the feature fusion part is used to splice and fuse the output features of the first branch and the output features of the second branch; the feature classification part is implemented using a support vector machine, and the parameters of the support vector machine are optimized using a particle swarm algorithm during the training phase.

[0014] In a second aspect, the present invention provides a drug molecule property prediction classification system based on BiLSTM;

[0015] The drug molecule property prediction classification system based on BiLSTM includes:

[0016] An acquisition module configured to: acquire a drug molecule to be predicted, and convert the molecular structure of the drug molecule to obtain drug molecule descriptors and drug molecule fingerprints;

[0017] A drug molecule property prediction module configured to: input both the drug molecule descriptors and the drug molecule fingerprints into the trained drug molecule property prediction classification network, and output drug molecule property prediction results;

[0018] Among them, the network structure of the trained drug molecule property prediction classification network includes: a feature extraction part, a feature fusion part, and a feature classification part connected in sequence; the feature extraction part includes two parallel branches: the first branch and the second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from drug molecule descriptors; the second branch is used to extract features from drug molecule fingerprints; the first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence; the feature fusion part is used to splice and fuse the output features of the first branch and the output features of the second branch; the feature classification part is implemented using a support vector machine, and the parameters of the support vector machine are optimized using a particle swarm algorithm during the training phase.

[0019] In a third aspect, the present invention also provides an electronic device, including:

[0020] a memory for non-temporarily storing computer-readable instructions; and

[0021] a processor for running the computer-readable instructions,

[0022] wherein when the computer-readable instructions are run by the processor, the method described in the first aspect above is executed.

[0023] In a fourth aspect, the present invention also provides a storage medium that non-temporarily stores computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the instructions for executing the method described in the first aspect are executed.

[0024] In a fifth aspect, the present invention also provides a computer program product, including a computer program that, when run on one or more processors, is used to implement the method described in the first aspect above.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] The present invention uses an optimized model based on a one-dimensional convolutional neural network for feature extraction, and a bidirectional long short-term memory module and an attention module are added thereto to fully and accurately extract feature information. The present invention uses a particle swarm optimization algorithm to improve the support vector machine, which can fully combine the advantages of both and can classify feature information more precisely. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0028] Figure 1 is a network architecture diagram of a drug molecule property prediction and classification method based on a bidirectional long short-term memory model information feature extraction neural network in Embodiment 1;

[0029] Figure 2 is the extraction of molecular fingerprint features by the bidirectional long short-term memory feature extraction network in Embodiment 1;

[0030] Figure 3 is a structural diagram of the long short-term memory module in Embodiment 1;

[0031] Figure 4 is an overview of the data set used in Embodiment 1;

[0032] Figure 5 is a comparison of ROC curves of experimental results of a part of the data set in Embodiment 1;

[0033] Figure 6 It is the comparison effect diagram of the evaluation effect of the first embodiment and the current advanced classification model. Specific implementation manners

[0034] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0035] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0036] In the case of no conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0037] All data acquisition in this embodiment is based on compliance with laws and regulations and user consent, and is a legal application of the data.

[0038] The first embodiment

[0039] This embodiment provides a drug molecule property prediction and classification method based on BiLSTM;

[0040] As Figure 1 shown, the drug molecule property prediction and classification method based on BiLSTM includes:

[0041] S101: Obtain the drug molecule to be predicted, and convert the molecular structure of the drug molecule to obtain a drug molecule descriptor and a drug molecule fingerprint;

[0042] S102: Input both the drug molecule descriptor and the drug molecule fingerprint into the trained drug molecule property prediction and classification network, and output the drug molecule property prediction result;

[0043] Among them, the network structure of the trained drug molecule property prediction and classification network includes:

[0044] A feature extraction part, a feature fusion part and a feature classification part connected in sequence;

[0045] The feature extraction part includes two parallel branches: the first branch and the second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from drug molecule descriptors; the second branch is used to extract features from drug molecule fingerprints;

[0046] The first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence;

[0047] The feature fusion part is used to splice and fuse the output features of the first branch and the output features of the second branch;

[0048] The feature classification part is implemented by a support vector machine, and the parameters of the support vector machine are optimized by a particle swarm algorithm in the training stage.

[0049] The specific structure of the first branch is as Figure 2 shown.

[0050] Furthermore, the working principle of the trained drug molecule property prediction classification network includes:

[0051] Extract features from drug molecule descriptors and drug molecule fingerprints respectively to obtain two drug molecule feature vectors; perform feature fusion on the two drug molecule feature vectors; perform classification prediction on the fused feature vectors, and output the drug molecule property prediction result.

[0052] Furthermore, the embedding layer is used to transform discrete molecular information or molecular fingerprints into continuous vectors.

[0053] The one-dimensional convolutional layer is used to adjust the number of channels and modify the feature dimension.

[0054] The average pooling layer is used to reduce the model calculation amount and prevent overfitting.

[0055] The bidirectional long short-term memory network is used to solve the problem of gradient disappearance or gradient explosion in the convolutional layer and preserve the order information in the sequence.

[0056] The concatenation layer is used to concatenate the outputs of the bidirectional long short-term memory network.

[0057] The self-attention mechanism module is used to modify the weights of feature information and increase the weights of key information in the classification network.

[0058] It should be understood that the present invention proposes a Bidirectional short - term and long - term memory and particle swarm optimized support vector machine classification network (BiLSTM - PSO - SVMN) structure to classify the effectiveness of drug molecules against various diseases, such as Figure 1 as shown. The entire network structure includes two parts, namely the feature extraction part and the feature - based classification part.

[0059] It should be understood that the present invention adds a bidirectional long - short - term memory layer to the traditional one - dimensional convolutional network to prevent gradient vanishing and gradient explosion. Relying solely on unidirectional long - short - term memory cannot handle sequential information flow well, and the molecular orientation information used in the present invention in BiLSTMN has strict requirements on the order in the information sequence. Therefore, the present invention uses bidirectional long - short - term memory, which has achieved good results in the field of natural language processing, to optimize the structure of the present invention. The long - short - term memory module is a variant of the traditional recurrent neural network, and its specific structure is as Figure 3 shown.

[0060] As Figure 3 shown, the entire LSTM module includes an input gate i t , an output gate o t and a forget gate f t . The calculation process of the long - short - term memory module is as follows:

[0061] i t =σ(W i c t-1 +U i x t +β i ); (1)

[0062]

[0063]

[0064] o t =σ(W o c t-1 +U o x t +β o ); (4)

[0065] c t =o t ⊙tanh(m t ); (5)

[0066] f t = σ(W f c t-1 + U f x t + β f ); (6)

[0067] where W and U represent offset matrices, β indicates a bias, and m t indicates the storage cell of the LSTM. x t and c t are the input vector and the hidden state vector at time t.

[0068] To enable the long short-term memory module to fully capture context information, a bidirectional long short-term memory module structure is adopted, with bidirectional hidden information and

[0069] The update of the two hidden informations and the calculation process of the subsequent concatenation layer are as follows:

[0070]

[0071] After the bidirectional long short-term memory network is processed, it is sent to the self-attention mechanism module for final feature extraction; the calculation process of the self-attention mechanism module is as follows:

[0072]

[0073] where η t is the hidden representation of c t , represents matrix transpose, η w is the context vector continuously initialized during the training phase, and the importance of the molecular partial information is calculated through the similarity between η t and η w Finally, the obtained weighted feature information s LSTM is sent to the subsequent classification module for processing.

[0074] Regarding the selection of the classifier, since the complexity of the molecular features obtained from the studied drug molecular feature dataset is relatively low, after comparative experiments of various classification models, it is found that the accuracy is the highest when using a support vector machine in the classification layer.

[0075] Therefore, a support vector machine is used to perform the final classification on the features collected in the attention layer. The kernel function used is RBF, the penalty parameter is set to 100, and the kernel parameter is 0.01.

[0076] Furthermore, the training process of the trained drug molecular property prediction classification network includes:

[0077] Construct a training set and a test set; both the training set and the test set are drug molecule descriptors and drug molecule fingerprints with known drug molecule properties.

[0078] Input the training set into the drug molecule property prediction classification network to train the network. When the value of the loss function no longer decreases, stop the training to obtain the trained drug molecule property prediction classification network; among them, during the process of training the network, the particle swarm algorithm is used to optimize the parameters of the support vector machine.

[0079] Input the test set into the trained drug molecule property prediction classification network to test the network. If the classification accuracy meets the set threshold, it is determined that the training is qualified, and the trained drug molecule property prediction classification network is obtained; otherwise, replace the training set and retrain.

[0080] Furthermore, during the process of training the network, the particle swarm algorithm is used to optimize the parameters of the support vector machine, which specifically includes:

[0081] Use the particle swarm algorithm to optimize the kernel parameter g and the penalty term C in the SVM to improve the generalization ability of the entire SVM classifier and obtain better results. The calculation process is as follows:

[0082]

[0083] Among them, Fit(x) is the fitness function of the entire optimization algorithm. The larger its value, the better the optimization effect. f i (x) and y i (x) are the true label and the predicted label respectively, and n is the number of samples.

[0084] The particle swarm algorithm regards C and g as a kind of particle, which is abstracted as a set of points in the plane, and continuously optimizes their positions L and moving speeds v.

[0085] The specific optimization process is as follows:

[0086]

[0087] Among them, the momentum coefficient U is used to control the overall optimization speed, c is the learning factor, rand is an independent random number belonging to the interval [0, 1], d is the dimension of the solution vector, i is the number of particles that can be composed of all C and g, G best and p best are the global optimal solution and the partial optimal solution respectively. The optimized parameters will be sent to the SVM for final classification.

[0088] When using the SVM as the classification layer for final classification, the obtained result is the distance from the experimental sample to the hyperplane, which is the basis for the support vector machine to perform classification.

[0089] The expected classification result is the probability of the effectiveness of a drug molecule against a certain disease. Therefore, the output of the support vector machine is processed, and the SigMoid activation function fitting method is used to convert the output of the support vector machine into the posterior probability P(y = 1|μ):

[0090]

[0091] where A and B are the parameters to be fitted, μ is the threshold-free output of the sample x, and f is the feature vector.

[0092] The advantage of the SigMoid fitting method is that it can well estimate the posterior probability while maintaining the sparsity of the support vector machine.

[0093] The present invention has conducted a large number of experiments on publicly available datasets such as HIV, BACE, and BBBP that are widely used in this field, as Figure 4 shown. The results obtained by the model are evaluated according to the evaluation index AUC (Area Under Curve). The evaluation effect is compared with that of the current advanced classification models, as Figure 5 and Figure 6 shown, indicating that the present invention has higher classification accuracy and better classification effect, and has a certain practicality. This shows that the drug molecule property prediction and classification method based on the deep learning convolutional neural network established by the present invention is effective, improves the calculation efficiency for screening effective target drugs in subsequent large-scale drug molecule libraries, and has a certain practicality.

[0094] Furthermore, the drug molecule descriptor specifically refers to a tool of chemometrics, which is a sequence or graph that uses mathematical statistics methods to explain the quantitative change law between the activity or physicochemical characteristics of a compound and its molecular structure.

[0095] Common molecular fingerprints include Daylight fingerprints, MDL, and public keys.

[0096] Furthermore, the drug molecule fingerprint refers to the Morgan molecular fingerprint.

[0097] The Morgan fingerprint is a circular fingerprint and also belongs to the topological fingerprint, which is obtained by adjusting the standard Morgan algorithm. It can be roughly equivalent to the extended connectivity fingerprint. These fingerprints have many advantages, such as fast calculation speed, no predefined (can represent infinitely many different molecular features), can contain chiral information, each element in the fingerprint represents a specific substructure, can be easily analyzed and interpreted, and can be modified accordingly according to different needs, etc. The specific Morgan fingerprint was initially developed for searching for molecular features related to activity rather than substructure search.

[0098] Furthermore, for the generation process of the drug molecular fingerprint, specifically:

[0099] S101-1: Atom initialization, assign an integer identifier to each heavy atom of the drug molecule, and the integer identifier is used to label the order of atoms in the molecule;

[0100] S101-2: Iterative update, with each heavy atom as the center, divide the heavy atoms near the center and the central heavy atom into a circle until the circle reaches the specified radius;

[0101] S101-3: For the sub-structure formed by the connection bonds between all heavy atoms in the circle, perform a sorting operation on the sub-structure and generate a feature list, and use the feature list as the drug molecular fingerprint; the sorting operation means that each heavy atom collects its own identifier and the identifiers of neighboring heavy atoms into an array, and sorts the heavy atoms in ascending order of the identifiers.

[0102] The present invention sets the number of bits of the required molecular fingerprint to 2048 bits.

[0103] The present invention is a classification algorithm for molecular fingerprint information data, belonging to the field of drug screening. The main steps are as follows: extract molecular fingerprint features through a bidirectional long short-term memory feature extraction network, which adds a bidirectional long short-term memory module and an attention module to a one-dimensional convolutional neural network to obtain more complete feature information. Then enter the PSO-SVM to output the classification result. Classify the drug data set through the trained network, and use AUC as an evaluation index to evaluate the performance of the network at this time. A better classification result is obtained by using the drug molecular property prediction and classification method based on the bidirectional long short-term memory model information feature extraction neural network, providing a method with obvious advantages for the field of drug molecular classification and screening.

[0104] Embodiment 2

[0105] This embodiment provides a drug molecular property prediction and classification system based on BiLSTM;

[0106] The drug molecular property prediction and classification system based on BiLSTM includes:

[0107] An acquisition module, configured to: acquire the drug molecule to be predicted, and convert the molecular structure of the drug molecule to obtain a drug molecular descriptor and a drug molecular fingerprint;

[0108] A drug molecular property prediction module, configured to: input both the drug molecular descriptor and the drug molecular fingerprint into the trained drug molecular property prediction and classification network, and output the drug molecular property prediction result;

[0109] Among them, the trained drug molecule property prediction classification network has a network structure including: a feature extraction part, a feature fusion part, and a feature classification part connected in sequence; the feature extraction part includes two parallel branches: a first branch and a second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from drug molecule descriptors; the second branch is used to extract features from drug molecule fingerprints; the first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence; the feature fusion part is used to splice and fuse the output features of the first branch and the output features of the second branch; the feature classification part is implemented using a support vector machine, and the parameters of the support vector machine are optimized using a particle swarm algorithm during the training phase.

[0110] It should be noted here that the above acquisition module and drug molecule property prediction module correspond to steps S101 to S102 in Embodiment 1. The examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0111] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0112] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the above module division is only a logical function division. In actual implementation, there can be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0113] Embodiment 3

[0114] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the above one or more computer programs are stored in the memory. When the electronic device runs, the processor executes the one or more computer programs stored in the memory so that the electronic device executes the method described in Embodiment 1 above.

[0115] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, 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, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0116] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0117] In the implementation process, each step of the above method may be completed by the integrated logic circuit in the hardware of the processor or the instructions in the form of software.

[0118] The method in Embodiment 1 may be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of the hardware and software modules in the processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0119] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with this embodiment can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0120] Embodiment 4

[0121] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method described in Embodiment 1 is completed.

[0122] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A drug molecule property prediction and classification method based on BiLSTM, characterized in that Including: Obtain the drug molecule to be predicted, and transform the molecular structure of the drug molecule to obtain a drug molecule descriptor and a drug molecule fingerprint; Input both the drug molecule descriptor and the drug molecule fingerprint into the trained drug molecule property prediction classification network, and output the drug molecule property prediction result; Among them, the network structure of the trained drug molecule property prediction classification network includes: a feature extraction part, a feature fusion part, and a feature classification part connected in sequence; the feature extraction part includes two parallel branches: a first branch and a second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from the drug molecule descriptor; the second branch is used to extract features from the drug molecule fingerprint; the first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence; the feature fusion part is used to splice and fuse the output features of the first branch and the output features of the second branch; the feature classification part is implemented using a support vector machine, and the parameters of the support vector machine are optimized using a particle swarm algorithm during the training stage; Construct a training set, input the training set into the drug molecule property prediction classification network, train the network, and stop training when the loss function value no longer decreases to obtain the trained drug molecule property prediction classification network; among them, during the process of training the network, use a particle swarm algorithm to optimize the parameters of the support vector machine; specifically including: Use a particle swarm algorithm to optimize the kernel parameter g and the penalty term C in the SVM to improve the generalization ability of the entire SVM classifier and obtain better results. The calculation process is as follows: Among them, Fit(x) is the fitness function of the entire optimization algorithm; f i (x) and y i (x) are the true label and the predicted label respectively, and n is the number of samples; The particle swarm algorithm regards C and g as a kind of particle, which is abstracted as a set of points in a plane, and continuously optimizes their positions L and moving speeds v; The specific optimization process is as follows: Among them, the momentum coefficient U is used to control the speed of overall optimization, c is the learning factor, rand is an independent random number belonging to the interval [0, 1], d is the dimension of the solution vector, i is the number of particles that can be composed of all C and g, G best and p best are the global optimal solution and the partial optimal solution respectively; the optimized parameters will be sent to the SVM for final classification; Process the output of the support vector machine, and use the SigMoid activation function fitting method to convert the support vector machine output into a posterior probability P(y = 1|μ): Among them, A and B are the parameters to be fitted, μ is the threshold-free output of the sample x, and f is the feature vector.

2. The method for predicting and classifying drug molecule properties based on BiLSTM according to claim 1, characterized in that The working principle of the trained drug molecule property prediction classification network includes: Extract features from the drug molecule descriptor and the drug molecule fingerprint respectively to obtain two drug molecule feature vectors; perform feature fusion on the two drug molecule feature vectors; perform classification prediction on the fused feature vectors, and output the drug molecule property prediction result.

3. The method for predicting and classifying drug molecule properties based on BiLSTM according to claim 1, characterized in that, The embedding layer is used to transform discrete molecular information or molecular fingerprints into continuous vectors; The one-dimensional convolutional layer is used to adjust the number of channels and modify the feature dimension; The average pooling layer is used to reduce the model calculation amount and prevent overfitting; The bidirectional long short-term memory network is used to solve the problem of gradient disappearance or gradient explosion in the convolutional layer and preserve the order information in the sequence; The concatenation layer is used to concatenate the output of the bidirectional long short-term memory network; The self-attention mechanism module is used to modify the weights of the feature information.

4. The method for predicting and classifying drug molecular properties based on BiLSTM according to claim 1, characterized in that The trained drug molecule property prediction classification network has a training process that includes: Constructing a test set; the test set is the drug molecule descriptors and drug molecule fingerprints of known drug molecule properties. Inputting the test set into the trained drug molecule property prediction classification network to test the network. If the classification accuracy meets the set threshold, it is determined that the training is qualified, and the trained drug molecule property prediction classification network is obtained; otherwise, replace the training set and retrain.

5. The method for predicting and classifying drug molecular properties based on BiLSTM according to claim 4, wherein The drug molecule fingerprint refers to the Morgan molecular fingerprint.

6. The method for predicting and classifying drug molecular properties based on BiLSTM according to claim 4, characterized in that, The specific generation process of the drug molecule fingerprint includes: Atom initialization, assigning an integer identifier to each heavy atom of the drug molecule, and the integer identifier is used to label the order of atoms in the molecule. Iterative update, with each heavy atom as the center, dividing the heavy atoms near the center and the central heavy atom into a circle until the circle reaches the specified radius. Sorting the sub-structures formed by the connection bonds between all the heavy atoms in the circle and generating a feature list, and using the feature list as the drug molecule fingerprint; the sorting operation means that each heavy atom collects its own identifier and the identifiers of neighboring heavy atoms into an array, and sorts the heavy atoms in ascending order of the identifiers.

7. A drug molecule property prediction and classification system based on BiLSTM, characterized in that, It includes: An acquisition module configured to: acquire the drug molecule to be predicted and convert the molecular structure of the drug molecule to obtain the drug molecule descriptor and the drug molecule fingerprint. A drug molecule property prediction module configured to: input both the drug molecule descriptor and the drug molecule fingerprint into the trained drug molecule property prediction classification network and output the drug molecule property prediction result. Among them, the network structure of the trained drug molecule property prediction classification network includes: a feature extraction part, a feature fusion part, and a feature classification part connected in sequence; the feature extraction part includes two parallel branches: the first branch and the second branch; the internal structures of the first branch and the second branch are the same; among them, the first branch is used to extract features from the drug molecule descriptor; the second branch is used to extract features from the drug molecule fingerprint; the first branch includes: an embedding layer, a one-dimensional convolutional layer, an average pooling layer, a bidirectional long short-term memory network BiLSTM, a concatenation layer, and a self-attention mechanism module connected in sequence; the feature fusion part is used to splice and fuse the output features of the first branch and the second branch; the feature classification part is implemented using a support vector machine, and the parameters of the support vector machine are optimized using a particle swarm algorithm during the training phase. Constructing a training set, inputting the training set into the drug molecule property prediction classification network to train the network, and stopping the training when the value of the loss function no longer decreases to obtain the trained drug molecule property prediction classification network; among them, during the process of training the network, the parameters of the support vector machine are optimized using a particle swarm algorithm; specifically including: The particle swarm optimization algorithm is used to optimize the kernel parameter g and the penalty term C in the SVM, improve the generalization ability of the entire SVM classifier, and obtain better results. The calculation process is as follows: Among them, Fit(x) is the fitness function of the entire optimization algorithm; f i (x) and y i (x) are the true label and the predicted label respectively, and n is the number of samples; The particle swarm optimization algorithm takes C and g as a kind of particle, abstracts them as a set of points in the plane, and continuously optimizes their positions L and moving speeds v; The specific optimization process is as follows: Among them, the momentum coefficient U is used to control the speed of overall optimization, c is the learning factor, rand is an independent random number belonging to the interval [0, 1], d is the dimension of the solution vector, i is the number of particles that can be composed of all C and g, G best and p best are the global optimal solution and the partial optimal solution respectively; the optimized parameters will be sent to the SVM for final classification; Process the output of the support vector machine, and use the SigMoid activation function fitting method to convert the support vector machine output into the posterior probability P(y = 1|μ): Where A and B are the parameters to be fitted, μ is the threshold-free output of the sample x, and f is the feature vector.

8. An electronic device, characterized in that it comprises: A memory for non-temporarily storing computer-readable instructions; And A processor for running the computer-readable instructions, Wherein, when the computer-readable instructions are run by the processor, the method described in any one of the above claims 1-6 is executed.

9. A storage medium, characterized in that it non - transiently stores computer - readable instructions, wherein, When the non-temporary computer-readable instructions are executed by a computer, instructions for executing the method described in any one of claims 1-6.

Images