Neural network model training method and apparatus, electronic device, and storage medium

Abstract

The present disclosure provides a network model training method and device, electronic equipment and storage medium, and relates to the technical field of biological information. The specific implementation scheme comprises: obtaining a data set, the data in the data set comprising drug target protein sequence data and non-drug target protein sequence data; obtaining a target neural network model having a function of predicting a drug target protein based on an initial ensemble learning model according to the data set, the initial ensemble learning model comprising a feature extraction network, at least two base learners and a total classifier, wherein the feature extraction network comprises an ESM2 model. The present disclosure can train a target neural network model having a function of predicting a drug target protein based on an ensemble learning model comprising an EMS2 model, so that whether a protein sequence corresponding to a protein is a drug target protein can be quickly, efficiently and accurately predicted according to the neural network model, and the prediction cost is reduced.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of bioinformatics, and particularly relates to a neural network model training method and device, electronic equipment and storage medium. BACKGROUND

[0002] A drug target refers to a biological macromolecule, such as a certain protein and nucleic acid, in vivo that has a drug effect and can be acted on by a drug. The design of a modern drug often needs to effectively identify a drug target from a complex biological system.

[0003] At present, the prediction of a drug target is usually obtained by a biological experiment method and a calculation model method. The biological experiment method determines a drug target through a large number of biological experiments, and the calculation model method predicts a drug target through a pre-designed mathematical model.

[0004] However, in the prior art, the biological experiment method has high accuracy, but is time-consuming, high in cost and low in efficiency; the calculation model method is high in efficiency, but the accuracy and reliability are limited by the design of the mathematical model and the quality of the data. SUMMARY

[0005] The present disclosure provides a neural network model training method and device, electronic equipment and storage medium, which can train a target neural network model having the function of predicting a drug target protein based on an ensemble learning model including an ESM2 model, so that whether a protein corresponding to a protein sequence is a drug target protein can be quickly, efficiently and accurately predicted according to the neural network model, thereby reducing the prediction cost.

[0006] According to a first aspect of the present disclosure, a neural network model training method is provided, comprising:

[0007] obtaining a data set, wherein the data in the data set includes drug target protein sequence data and non-drug target protein sequence data; obtaining a target neural network model having the function of predicting a drug target protein based on an initial ensemble learning model according to the data set, wherein the initial ensemble learning model includes a feature extraction network, at least two base learners and a total classifier, and the feature extraction network includes an ESM2 model.

[0008] Further, before the target neural network model having the function of predicting a drug target protein is obtained based on the initial ensemble learning model according to the data set, the method further comprises:

[0009] dividing the data set into a training set and a test set;

[0010] The target neural network model with the function of predicting the drug target protein is obtained based on an initial ensemble learning model according to the data set, and the method comprises the steps of:

[0011] The target neural network model with the function of predicting the drug target protein is obtained based on an initial ensemble learning model according to the training set.

[0012] Further, the data set is divided into a training set and a test set, which comprises:

[0013] According to a preset proportion, the data set is divided into the training set and the test set.

[0014] Further, the base learner comprises at least two of a neural network (NN), a CatBoost model, an XGBoost model, and a support vector machine (SVM).

[0015] Further, the training set comprises a first training set and a second training set, the total classifier is a learning-based support vector machine (SVM), and the target neural network model with the function of predicting the drug target protein is obtained based on an initial ensemble learning model according to the training set, which comprises the steps of:

[0016] The at least two base learners are optimized according to the first training set, and the total classifier is optimized according to the second training set, so as to obtain the target neural network model with the function of predicting the drug target protein.

[0017] Further, the at least two base learners are optimized according to the first training set, and the total classifier is optimized according to the second training set, so as to obtain the target neural network model with the function of predicting the drug target protein.

[0018] The first training set is subjected to feature extraction according to the ESM2 model to obtain first features, and the first training set is subjected to feature extraction according to an amino acid composition ACC algorithm to obtain second features; the first features and the second features are subjected to fusion and normalization processing to obtain third features; the at least two base learners respectively output classification results of whether a protein corresponding to the current features is a drug target protein based on the third features; loss functions corresponding to the at least two base learners are obtained based on the classification results corresponding to the at least two base learners and the first training set; the at least two base learners are respectively optimized based on the loss functions corresponding to the at least two base learners to obtain optimized at least two base learners; the total classifier is optimized based on the second training set and classification results of the second training set by the optimized at least two base learners to obtain an optimized total trainer; and the target neural network is obtained based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer.

[0019] According to a second aspect of the present disclosure, a neural network model training device is provided, which comprises:

[0020] A data processing module is configured to obtain a data set, wherein data in the data set comprises drug target protein sequence data and non-drug target protein sequence data.

[0021] A model training module is configured to obtain a target neural network model having a function of predicting drug target proteins based on an initial ensemble learning model according to the data set, wherein the initial ensemble learning model comprises a feature extraction network, at least two base learners, and a total classifier, and the feature extraction network comprises an ESM2 model.

[0022] Further, the data processing module is further configured to divide the data set into a training set and a test set; and the model training module is specifically configured to obtain a target neural network model having a function of predicting drug target proteins based on an initial ensemble learning model according to the training set.

[0023] Further, the data processing module is specifically configured to divide the data set into the training set and the test set according to a preset ratio.

[0024] Further, the base learners comprise at least two of a neural network NN, a CatBoost model, an XGBoost model, and a support vector machine SVM.

[0025] Further, the training set includes a first training set and a second training set, and the total classifier is a learning-based support vector machine (SVM), and the model training module is specifically configured to: optimize the at least two base learners according to the first training set, and optimize the total classifier according to the second training set, to obtain a target neural network model having a function of predicting a drug target protein.

[0026] Further, the model training module is specifically configured to: perform feature extraction on the first training set according to the ESM2 model to obtain first features, and perform feature extraction on the first training set according to an amino acid composition ACC algorithm to obtain second features; perform fusion and normalization processing on the first features and the second features to obtain third features; the at least two base learners respectively output classification results of whether a protein corresponding to the current features is a drug target protein based on the third features; obtain loss functions corresponding to the at least two base learners based on the classification results corresponding to the at least two base learners and the first training set; respectively optimize the at least two base learners based on the loss functions corresponding to the at least two base learners to obtain optimized at least two base learners; and optimize the total classifier based on the second training set and classification results of the second training set by the optimized at least two base learners, to obtain an optimized total classifier; and obtain the target neural network based on the feature extraction network, the optimized at least two base learners, and the optimized total classifier.

[0027] According to a third aspect of the present disclosure, an electronic device is provided, including: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect.

[0028] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to enable a computer to perform the method according to the first aspect.

[0029] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program, and the computer program, when executed by a processor, implements the method according to the first aspect.

[0030] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings are used to better understand the present scheme and do not constitute a limitation on the present disclosure. Among them:

[0032] Figure 1 The flowchart of the neural network model training method provided by the embodiment of the present disclosure is shown in the figure;

[0033] Figure 2 Another flowchart of the neural network model training method provided by the embodiment of the present disclosure is shown in the figure;

[0034] Figure 3 The composition schematic diagram of the neural network model training device provided by the embodiment of the present disclosure is shown in the figure;

[0035] Figure 4 The composition schematic diagram of the electronic device provided by the embodiment of the present disclosure is shown in the figure. DETAILED DESCRIPTION

[0036] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of the present disclosure to help understanding, which should be considered only as exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, in order to be clear and concise, the description below omits the description of well-known functions and structures.

[0037] It should be understood that in various embodiments of the present disclosure, the character " / " generally indicates that the associated objects before and after are in an "or" relationship. The terms "first", "second", etc. are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

[0038] Drug target refers to a biological macromolecule in vivo with drug efficacy function and capable of being acted on by a drug, such as certain protein and nucleic acid biological macromolecules. The design of a modern drug often needs to effectively identify the drug target from a complex biological system.

[0039] At present, the prediction of drug targets is usually obtained by biological experimental method and computational model method. The biological experimental method determines the drug target through a large number of biological experiments, and the computational model method predicts the drug target through a pre-designed mathematical model.

[0040] However, in the prior art, the biological experimental method has high accuracy, but is time-consuming, high-cost and low-efficient; the computational model method has high efficiency, but the accuracy and reliability are limited by the design of the mathematical model and the quality of the data.

[0041] In this background art, the present disclosure provides a neural network model training method, which can train a target neural network model having a function of predicting a drug target protein based on an ensemble learning model including an ESM2 model, so that whether a protein corresponding to a protein sequence is a drug target protein can be quickly, efficiently and accurately predicted according to the neural network model, and the prediction cost is reduced.

[0042] The execution subject of the neural network model training method provided by the embodiments of the present disclosure can be a computer or a server, or can also be other electronic devices with data processing capability; or the execution subject of the method can also be a processor (such as a central processing unit (CPU)) in the above-mentioned electronic devices; or the execution subject of the method can also be an application program (APP) installed in the above-mentioned electronic devices and capable of realizing the functions of the method; or the execution subject of the method can also be a functional module or unit with the functions of the method in the above-mentioned electronic devices. The execution subject of the method is not limited herein.

[0043] In some embodiments, the server can be a single server, or can also be a server cluster composed of multiple servers. In some embodiments, the server cluster can also be a distributed cluster. The specific implementation of the server is not limited by the present disclosure.

[0044] The neural network model training method will be described below with reference to the accompanying drawings.

[0045] Figure 1 A flowchart of the neural network model training method provided by the embodiments of the present disclosure is shown in FIG. 1. As shown in FIG. 1, the method can include: Figure 1

[0046] S101, obtaining a data set.

[0047] The data in the data set includes drug target protein sequence data and non-drug target protein sequence data.

[0048] For example, sequence data of proteins that can be drug targets and sequence data of proteins that cannot be drug targets can be collected from existing protein sequence databases to form a data set. For example, protein sequence data can be obtained from public protein sequence databases such as DRUGBANK and UniProt to obtain a data set. It can be understood that the protein sequence data in the data set can come from various species (such as plants, animals, humans and microorganisms), which will not be described herein.

[0049] ​As a possible implementation, after obtaining the data set, the data in the data set can also be preprocessed to ensure the quantity and high quality of the data in the data set.

[0050] For example, protein sequence data with a sequence length exceeding a preset number (such as 512) can be removed to improve the training efficiency of the model; the data set can be de-duplicated to remove redundant data in the data set; data augmentation and processing methods such as generating pseudo samples, adding noise, etc. can be used to expand the data set to increase the data quantity; oversampling, undersampling, SMOTE, etc. can be used to balance the proportion of drug target protein sequence data and non-drug target protein sequence data in the data set to avoid the neural network model trained by the data set from being biased towards the majority class and improve the accuracy of the model.

[0051] S102, obtaining a target neural network model for predicting the function of a drug target protein based on an initial ensemble learning model according to the data set, the initial ensemble learning model comprising a feature extraction network, at least two base learners and a total classifier.

[0052] The feature extraction network comprises an ESM2 model.

[0053] For example, other Transformer-based models (such as BERT, ProtBert, etc.) can also be used for protein sequence feature extraction.

[0054] For example, more protein feature information can be combined during feature extraction, such as secondary structure, tertiary structure and functional domain information, and features based on the physical and chemical properties of amino acids, such as hydrophobicity, charge and volume, etc., to improve the accuracy of the obtained target neural network model.

[0055] For example, the ESM2 model can be an esm2_t33_630M_UR30D model, and can also be other models in the ESM2 series, which are not limited here.

[0056] Optionally, the base learner can include at least two of a neural network NN, a CatBoost model, an XGBoost model, and a support vector machine SVM. It can be understood that the neural network NN can be any artificial neural network, such as a graph neural network GNN, an ESM2 model, etc., which is not limited.

[0057] For example, the base learner can also be a random forest, a decision tree or other machine learning classifier, which is not limited.

[0058] Exemplarily, the total classifier can be a support vector machine (SVM) with preset weights for each base learner, and can also implement a learning-based weighted voting or multi-model fusion technology, such as a hybrid expert system or a soft voting mechanism, which will not be described herein.

[0059] Exemplarily, based on the above data set, an ensemble learning model including an ESM2 model can be used as an initial model for training (for example, using a supervised learning related technology for training), so as to obtain the target neural network model.

[0060] By the neural network model training method of the present disclosure, the target neural network model with the function of predicting drug target proteins can be trained based on an ensemble learning model including an ESM2 model, so that the protein sequence corresponding to the protein whether it is a drug target protein can be quickly and efficiently and accurately predicted according to the neural network model, and the prediction cost is reduced.

[0061] In some possible embodiments, before the target neural network model with the function of predicting drug target proteins is obtained based on the initial ensemble learning model according to the data set, the method can further include:

[0062] Dividing the data set into a training set and a test set.

[0063] Exemplarily, the data set can be randomly split into a training set and a test set. Of course, in some possible implementations, the data set can also be split in other ways to obtain the training set and the test set, which is not limited herein. For example, the data set can be evenly split according to the species corresponding to the protein sequences in the data set, so that the training set and the test set obtained both have protein sequences from various species.

[0064] Optionally, the dividing the data set into a training set and a test set can include:

[0065] According to a preset ratio, the data set is divided into the training set and the test set.

[0066] For example, the preset ratio can be 9:1, that is, 90% of the data in the data set can be used as the training set, and the remaining 10% of the data can be used as the test set. Of course, in actual application, other preset ratios can also be set according to actual needs or requirements, which is not limited herein.

[0067] In this way, the training of the neural network model according to the training set and the testing of the trained neural network model according to the test set can be facilitated, so as to improve the training quality of the target neural network model.

[0068] Correspondingly, the obtaining, based on the initial ensemble learning model, of the target neural network model having the function of predicting the drug target protein from the data set can include:

[0069] The obtaining, based on the initial ensemble learning model, of the target neural network model having the function of predicting the drug target protein from the training set.

[0070] In some possible implementation manners, the training set includes a first training set and a second training set, the total classifier is a learning-based support vector machine (SVM), and the obtaining, based on the initial ensemble learning model, of the target neural network model having the function of predicting the drug target protein from the training set can include:

[0071] The at least two base learners are optimized according to the first training set, and the total classifier is optimized according to the second training set, to obtain the target neural network model having the function of predicting the drug target protein.

[0072] For example, the base learner can output a classification result of whether a protein corresponding to a current feature is a drug target protein according to the protein sequence feature extracted by the first training set from the feature extraction network, obtain a loss function according to the classification result and an actual type (i.e., a drug target protein or a non-drug target protein) of the protein corresponding to the current feature, and optimize the base learner according to the loss function.

[0073] For example, it is assumed that , represents a classification probability of the i th base learner on the j th class, and the overall prediction result is . The number of base learners is represented by N. In a traditional weighted voting mechanism, the final prediction result can be represented by formula (1):

[0074] Formula (1)

[0075] wherein, w i represents a weight of the i th base model, and satisfies . is usually manually set in advance according to a preset rule or performance of the base learner. In this embodiment, the total classifier is a learning-based support vector machine (SVM), and one SVM is used as a voter in this embodiment, which automatically learns the weight through an optimization problem. It is assumed that

[0076] represents a combined prediction probability vector, and is defined as ​​​The learning-based support vector machine SVM in the embodiment determines the weight by solving an optimization problem of the following formula and the bias :

[0077] Formula (2)

[0078] wherein, represents the weight learned by the SVM, represents the bias.

[0079] Through this optimization problem, the support vector machine SVM can automatically adjust the weight and the bias of each base learner, avoiding the subjectivity and inaccuracy of manually setting the weight in the traditional voter.

[0080] Exemplarily, the total classifier can be optimized by the second training set and the classification results of the at least two base learners on the second training set, combined with formula (2), so that the total classifier automatically optimizes the weight in the training process, thereby avoiding the subjectivity and inaccuracy of manually setting the weight in the traditional voter, and ensuring the optimization of the performance of the target neural network model.

[0081] In some possible implementations, the optimization of the at least two base learners according to the first training set, the optimization of the total classifier according to the second training set, and the obtaining of the target neural network model having the function of predicting the drug target protein can include:

[0082] S201, according to the ESM2 model, the first training set is feature extracted to obtain the first feature, and according to the amino acid composition ACC algorithm, the first training set is feature extracted to obtain the second feature.

[0083] Exemplarily, taking the ESM2 model as an esm2_t33_630M_UR30D model, the self-attention mechanism of the model can be represented by formula (3):

[0084] Formula (3)

[0085] wherein, , , query, key and value matrices are represented by Q, K and V respectively, and d_k represents the dimension of the key vector.

[0086] According to the model, the data in the training set can be feature extracted to obtain a sequence representation of the protein sequence wherein, represents Positional characteristics.

[0087] Exemplarily, AAC (amino acid composition) is a common method for describing protein characteristics, which converts protein sequences into data representing the percentage of different amino acids, reflecting the amino acid composition of the protein, and the principle is as follows:

[0088]

[0089] wherein, represents the amino acid the frequency of occurrence in the sequence, represents the amino acid the total number of occurrences in the sequence, represents the length of the sequence.

[0090] S202, according to the first feature and the second feature, fusion and normalization processing are performed to obtain a third feature.

[0091] Exemplarily, the first feature and the second feature can be fused (such as splicing fusion, convolution fusion, attention fusion, etc.), and the fused feature can be normalized to obtain the third feature.

[0092] Exemplarily, min-max normalization can be applied to the feature data to ensure consistency and comparability of the feature data, and the principle is as follows:

[0093]

[0094] wherein, represents the normalized feature, and respectively represent the maximum feature value and the minimum feature value, represents the feature data to be normalized.

[0095] S203, the at least two base learners respectively output the classification results of whether the protein corresponding to the current feature is a drug target protein based on the third feature.

[0096] S204, based on the classification results corresponding to the at least two base learners and the first training set, the loss function corresponding to the at least two base learners is obtained.

[0097] Exemplarily, the loss function of the at least two base learners can be obtained according to the classification results of the at least two base learners and the actual type of the protein corresponding to the at least two classification results in the first training set.

[0098] S205, respectively optimize the at least two base learners based on the loss functions corresponding to the at least two base learners, to obtain optimized at least two base learners.

[0099] For example, the corresponding base learners can be respectively optimized based on the aforementioned loss functions, to obtain optimized at least two base learners.

[0100] S206, optimize the total classifier according to the second training set and the classification results of the optimized at least two base learners on the second training set, to obtain an optimized total trainer.

[0101] For example, the embodiment can refer to the specific implementation in the foregoing embodiments, which will not be described here.

[0102] S207, obtain the target neural network based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer.

[0103] For example, the neural network model can be constructed according to the feature extraction network, the optimized at least two base learners, and the optimized total trainer.

[0104] The embodiment extracts features from the first training set according to the ESM2 model to obtain first features, and extracts features from the first training set according to the amino acid composition ACC algorithm to obtain second features; performs fusion and normalization processing on the first features and the second features to obtain third features; the at least two base learners respectively output classification results of whether the protein corresponding to the current features is a drug target protein based on the third features; based on the classification results corresponding to the at least two base learners and the first training set, obtain the loss functions corresponding to the at least two base learners; respectively optimize the at least two base learners based on the loss functions corresponding to the at least two base learners, to obtain optimized at least two base learners; optimize the total classifier according to the second training set and the classification results of the optimized at least two base learners on the second training set, to obtain an optimized total trainer; obtain the target neural network based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer, which can quickly and accurately obtain a target neural network model with the function of predicting drug target proteins.

[0105] It should be noted that many current models are weak in the interpretability of the predicted results, and it is difficult for researchers to provide in-depth model analysis. The present disclosure can enhance the interpretability of the model by using t-SNE (t-distributed Stochastic Neighbor Embedding) and SHAP (Shapley Additive ex Planations) technology, so that researchers can more clearly understand the model, and improve the transparency of the model and the basis for scientific decision-making. The t-SNE and SHAP technologies are specifically introduced below.

[0106] t-SNE first calculates the conditional probability between any two points and in the high-dimensional space , which reflects the importance of point to point :

[0107]

[0108] wherein, is the square of the Euclidean distance between points , and is the standard deviation of the Gaussian distribution around point , which is dynamically determined by the perplexity of the data.

[0109] In order to simplify the model, t-SNE defines a symmetric joint probability P by averaging the conditional probability:

[0110]

[0111] wherein, is the total number of points in the data set.

[0112] In the low-dimensional space, the similarity between point and point is represented by a t-distribution, which has a heavier tail at a smaller distance:

[0113]

[0114] wherein, is the square of the Euclidean distance between point and point in the low-dimensional space.

[0115] The goal of the algorithm is to minimize the Kullback-Leibler divergence between the high-dimensional and low-dimensional probability distributions:

[0116]

[0117] wherein, is the loss function, and Represent the probability distribution in high-dimensional and low-dimensional space respectively.

[0118] The SHAP value measures the contribution of each feature to a single prediction by treating the model prediction task as a cooperative game in which each feature is a player. This approach can reveal the absolute contribution of each feature to the prediction and maintain consistency and local accuracy, which means that for any given prediction, the sum of the SHAP values ​​is equal to the difference between the actual model prediction value and the baseline prediction value. Calculating SHAP Values , in addition to the feature set Except All possible subsets except The specific calculation formula is as follows:

[0119]

[0120] in, is a subset The size of is the total number of features, Does not contain features The predicted value of Contains features The prediction results of . This formula considers different subsets Effect added to the prediction output to quantify features marginal contribution.

[0121] In addition to SHAP and t-SNE, other interpretability and visualization tools such as LIME (Local Interpretable Model-Effect Explanations), PCA (Principal Component Analysis), or UMAP (Uniform Manifold Approximation and Projection) can be used.

[0122] The above mainly introduces the solution provided by the embodiment of the present disclosure from the perspective of method. In order to realize the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. It should be easy to realize that the technical goals in this field, combined with the units and algorithm steps of each example described in the embodiments disclosed in this article, can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven hardware manner depends on the specific application and design constraints of the technical solution. Professional technical goals 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 this disclosure.

[0123] In an example embodiment, the present disclosure also provides a neural network model training apparatus, which can be used to implement the neural network model training method as described in the foregoing embodiments.

[0124] Figure 3 A constituent schematic diagram of the neural network model training apparatus provided by the present disclosure is shown in FIG. 3. As shown in FIG. 3, the apparatus can include a data processing module 301 and a model training module 302. Figure 3 The data processing module 301 is configured to obtain a data set, wherein the data in the data set includes drug target protein sequence data and non-drug target protein sequence data.

[0125] The data processing module 301 is configured to obtain a data set, wherein the data in the data set includes drug target protein sequence data and non-drug target protein sequence data.

[0126] The model training module 302 is configured to obtain a target neural network model having a function of predicting a drug target protein based on an initial ensemble learning model according to the data set, wherein the initial ensemble learning model includes a feature extraction network, at least two base learners, and a total classifier, and the feature extraction network includes an ESM2 model.

[0127] In some possible implementation manners, the data processing module 301 is further configured to divide the data set into a training set and a test set, and the model training module 302 is specifically configured to obtain a target neural network model having a function of predicting a drug target protein based on an initial ensemble learning model according to the training set.

[0128] In some possible implementation manners, the data processing module 301 is specifically configured to divide the data set into the training set and the test set according to a preset ratio.

[0129] In some possible implementation manners, the base learners include at least two of a neural network (NN), a CatBoost model, an XGBoost model, and a support vector machine (SVM).

[0130] In some possible implementation manners, the training set includes a first training set and a second training set, the total classifier is a learning-based support vector machine (SVM), and the model training module 302 is specifically configured to optimize the at least two base learners according to the first training set, optimize the total classifier according to the second training set, and obtain a target neural network model having a function of predicting a drug target protein.

[0131] In some possible implementation manners, the model training module 302 is specifically configured to: perform feature extraction on the first training set according to the ESM2 model to obtain first features, perform feature extraction on the first training set according to an amino acid composition ACC algorithm to obtain second features; perform fusion and normalization processing on the first features and the second features to obtain third features; the at least two base learners respectively output classification results of whether proteins corresponding to the current features are drug target proteins based on the third features; loss functions corresponding to the at least two base learners are obtained based on the classification results corresponding to the at least two base learners and the first training set; the at least two base learners are respectively optimized based on the loss functions corresponding to the at least two base learners to obtain optimized at least two base learners; the total classifier is optimized based on the second training set and classification results of the second training set by the optimized at least two base learners to obtain an optimized total trainer; and the target neural network is obtained based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer.

[0132] It should be noted that, Figure 3 The division of the modules in the above embodiments is illustrative, and is merely a logical function division. In actual implementation, another division manner can be used. For example, two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of a software function module.

[0133] In the technical solutions of the present disclosure, the acquisition, storage and application of the protein sequence data comply with relevant laws and regulations and do not violate public order and good customs.

[0134] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0135] In an example embodiment, the electronic device includes at least one processor and a memory connected to the at least one processor in communication. The memory stores instructions executable by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the above embodiments. The electronic device can be the computer or the server described above.

[0136] In an example embodiment, the readable storage medium can be a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method according to the above embodiments.

[0137] In an exemplary embodiment, a computer program product includes a computer program, which implements the method according to the above embodiments when executed by a processor.

[0138] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0139] like Figure 4 As shown, electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. Various programs and data required for the operation of electronic device 400 may also be stored in RAM 403. Computing unit 401, ROM 402, and RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to bus 404.

[0140] Multiple components in the electronic device 400 are connected to the I / O interface 405, including an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a magnetic disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 400 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0141] The computing unit 401 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 401 performs various methods and processes described above, such as the neural network model training method. For example, in some embodiments, the neural network model training method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded onto the RAM 403 and executed by the computing unit 401, one or more steps of the neural network model training method described above can be performed. Alternatively, in other embodiments, the computing unit 401 can be configured to perform the neural network model training method by any other appropriate means, such as by means of firmware.

[0142] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0143] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0144] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0145] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0146] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0147] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server can arise by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0148] It should be understood that the various forms of flow shown above can be used to reorder, add, or delete steps. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technology disclosed in the present disclosure can be achieved, which is not limited herein.

[0149] The above detailed description does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A neural network model training method, characterized in that: The method comprises: Acquiring a data set, wherein the data in the data set includes drug target protein sequence data and non-drug target protein sequence data; Dividing the data set into a training set and a test set; Obtaining a target neural network model having a function of predicting drug target proteins based on an initial ensemble learning model according to the training set, wherein the initial ensemble learning model includes a feature extraction network, at least two base learners, and a total classifier, wherein the feature extraction network includes an ESM2 model; The training set includes a first training set and a second training set, the overall classifier is a learning-based support vector machine (SVM), and the target neural network model having the function of predicting drug target proteins is obtained based on the initial ensemble learning model according to the training set, including: Performing feature extraction on the first training set according to the ESM2 model to obtain a first feature, and performing feature extraction on the first training set according to the amino acid composition ACC algorithm to obtain a second feature; Performing a fusion and normalization process on the first feature and the second feature to obtain a third feature; The at least two base learners respectively output, based on the third feature, a classification result of whether the protein corresponding to the current feature is a drug target protein; Obtaining loss functions corresponding to the at least two base learners based on the classification results corresponding to the at least two base learners and the first training set; Optimizing the at least two base learners based on the loss functions corresponding to the at least two base learners to obtain the at least two optimized base learners; Optimizing the overall classifier according to the second training set and the classification results of the at least two optimized base learners on the second training set to obtain an optimized overall classifier; The target neural network is obtained based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer.

2. The method according to claim 1, characterized in that The step of dividing the data set into a training set and a test set comprises: The data set is divided into the training set and the test set according to a preset ratio.

3. The method according to claim 2, characterized in that The base learner includes at least two of a neural network NN, a CatBoost model, an XGBoost model, and a support vector machine SVM.

4. A neural network model training device, characterized in that: The device comprises: A data processing module is used to obtain a data set, wherein the data in the data set includes drug target protein sequence data and non-drug target protein sequence data; and divide the data set into a training set and a test set; A model training module is used to obtain a target neural network model with a function of predicting drug target proteins based on the training set based on an initial integrated learning model, wherein the initial integrated learning model includes a feature extraction network, at least two base learners and a total classifier, wherein the feature extraction network includes an ESM2 model; wherein the training set includes a first training set and a second training set, and the total classifier is a support vector machine SVM based on learning, and the model training module is specifically used to perform feature extraction on the first training set according to the ESM2 model to obtain a first feature, perform feature extraction on the first training set according to the amino acid composition ACC algorithm to obtain a second feature; perform fusion and normalization processing on the first feature and the second feature to obtain a third feature. feature; the at least two base learners respectively output, based on the third feature, a classification result of whether the protein corresponding to the current feature is a drug target protein; based on the classification results corresponding to the at least two base learners and the first training set, the loss functions corresponding to the at least two base learners are obtained; based on the loss functions corresponding to the at least two base learners, the at least two base learners are respectively optimized to obtain the optimized at least two base learners; according to the second training set and the classification results of the optimized at least two base learners for the second training set, the total classifier is optimized to obtain the optimized total trainer; based on the feature extraction network, the optimized at least two base learners, and the optimized total trainer, the target neural network is obtained.

5. The device according to claim 4, characterized in that The data processing module is specifically used to: The data set is divided into the training set and the test set according to a preset ratio.

6. The device according to claim 5, characterized in that The base learner includes at least two of a neural network NN, a CatBoost model, an XGBoost model, and a support vector machine SVM.

7. An electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method according to any one of claims 1 to 3.

8. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 3.

Images