Method for predicting antibacterial peptide in bacteria based on deep learning

Abstract

The invention provides a method for predicting antibacterial peptides in bacteria based on deep learning, and relates to the technical field of antibacterial peptide prediction of microorganisms, antibacterial peptide sequences are collected from a plurality of antibacterial peptide databases to serve as a positive data set, and protein sequences are collected from a protein database to serve as a negative data set for training; all amino acid sequences in the positive data set and the negative data set are converted into vector sequences according to a digital-amino acid conversion table to serve as model input; and constructing a multi-layer combination model, inputting the vector sequence, and outputting a prediction probability value so as to judge the antibacterial peptide sequence with antibacterial activity.

Description

Technical Field

[0001] The present invention relates to the technical field of microbial antimicrobial peptide prediction, and in particular to a method for predicting antimicrobial peptides in bacteria based on deep learning. Background Art

[0002] Antimicrobial peptides are usually short, positively charged amphipathic cationic peptides, usually less than 50 amino acids, and are important candidates for solving the problem of antibiotic failure caused by existing antibiotic-resistant bacterial infections. Therefore, the rapid and efficient prediction and screening of antimicrobial peptides plays an important role.

[0003] Limited by the number of antimicrobial peptides discovered, existing prediction methods have fewer training sets available, and the low sequence similarity of antimicrobial peptides may make it difficult for old methods or models to discover new antimicrobial peptides. In recent years, antimicrobial peptide data has shown a significant growth trend, but existing machine learning models often do not consider the accuracy of the data when constructing training sets, and do not distinguish whether the antimicrobial peptides exist naturally, which will lead to a decrease in model prediction accuracy and possible expression problems. Based on the above reasons, the existing prediction methods for bacterial antimicrobial peptides may have poor accuracy and are not conducive to subsequent expression and modification research. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method for predicting antimicrobial peptides in bacteria based on deep learning, comprising the following steps: S1. Collect antimicrobial peptide sequences from multiple antimicrobial peptide databases as positive data sets, and collect protein sequences from protein databases as negative data sets for training; S2, converting all amino acid sequences in the positive and negative data sets into vector sequences as model input according to the number-amino acid conversion table; S3. Construct a multi-layer combination model, input the vector sequence, and output a predicted probability value, so as to determine the antimicrobial peptide sequence with antimicrobial activity.

[0005] In a preferred embodiment, in step S1, antimicrobial peptide sequences are collected from multiple antimicrobial peptide databases such as APD3, DRAMP, DBAASP, and dbAMP as positive data sets, and amino acid sequences without antimicrobial, antibiotic, and bactericide annotations are downloaded from the Uniprot protein database as negative data sets.

[0006] In a preferred embodiment, in step S3, the relative mapping relationship adopts the following formula: in, are multiple embedding vectors of amino acid sequence A A set of, where i represents the position of the amino acid in the mapping, and n is the number of amino acid sequences; The antimicrobial peptide sequence feature extraction formula is as follows: in, represents the output value at position i after the convolution operation, represents the output value at position i after the pooling operation, x is the input vector, w is the convolution kernel, b is the bias term, j is the position index in the convolution kernel, k is the size of the convolution kernel, p is the maximum pooling window size, and s is the step size.

[0007] In a preferred embodiment, the forward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the backward activation vectors of the input gate, forget gate, and output gate, respectively. is the backward candidate memory unit, is the state of the backward memory unit at the current time step, is the backward hidden state of the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the backward input vector of the input sequence at time step t, Yes Input To the backward weight matrix of each gate, is the backward hidden state at time step t-1 To the backward weight matrix of each gate, is the back bias term of each gate.

[0008] In a preferred embodiment, the backward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the forward activation vectors of the input gate, forget gate, and output gate, respectively. is the forward candidate memory unit, is the state of the forward memory unit at the current time step, is the forward hidden state at the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the forward input vector of the input sequence at time step t, yes To the forward weight matrix of each gate, is the forward hidden state at time step t+1 To the forward weight matrix of each gate, is the forward bias term for each gate.

[0009] In a preferred embodiment, the forward and backward hidden states are concatenated as the output of a bidirectional LSTM network: .

[0010] In a preferred embodiment, the binary cross entropy loss function calculates the loss expression as: in Represents the loss value of a single sample, represents the average loss value of the entire data set, y and They represent the true label and predicted value of the sequence respectively, and m represents the number of samples.

[0011] In a preferred embodiment, the classification of antimicrobial peptide sequences is achieved based on the binary classification probability value output by the Sigmoid activation function: when the probability value>0.5, the sequence is determined to be an antimicrobial peptide sequence; when the probability value≤0.5, it is determined to be a non-antimicrobial peptide sequence.

[0012] In a preferred embodiment, in step S2, sequences 10 times the number of the positive data set are extracted from the collected negative data set and merged with the positive data set into a sample set, and distributed sampling is used to cut the sample set into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0013] In a preferred embodiment, a five-fold cross validation method is used to evaluate the multi-layer combination model and automatically search for the optimal hyperparameter combination.

[0014] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention proposes a new deep neural network model. Compared with the existing multi-model combination prediction, only one model is used to realize the prediction of antimicrobial peptides. No complex manual feature extraction is required for the prediction sequence, which reduces the complexity of antimicrobial peptide prediction and the computational cost of prediction. Compared with the existing single-model prediction method, this model adopts a more rigorous and accurate data set screening, starting from the data set to reduce the potential misjudgment effect on the model. At the same time, this model constructs a deep structure model, fully utilizing the characteristics of different layers, realizing sequence feature extraction through convolutional layers, and further capturing potential correlation features of the sequence context through two layers of bidirectional long short-term memory networks. Finally, a layer of attention layer is used to complete the weighted scoring of important features, optimize the model's ability to identify the potential characteristics of antimicrobial peptides, and have good generalization capabilities for various types of antimicrobial peptides.

[0015] Compared with the existing prediction methods such as AMPActiPred, Macrel, amplify (imbalance), CAMP (ANN), and AMPScanner, the accuracy, precision, recall rate, F1-score, and MCC are improved by 3.9%-14.8%, 15.1%-58.9%, 4.6%-35.4%, 20.3%-41.1%, and 20.5%-42.0%, respectively. The present invention has better accuracy and versatility for the prediction of antimicrobial peptides. It is expected to promote the discovery and application of new antimicrobial peptides and accelerate the development of antimicrobial peptides. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0017] Figure 1 This is a schematic diagram of the process of the method for predicting antimicrobial peptides in bacteria based on deep learning of the present invention; Figure 2 It is a structural schematic diagram of the multi-layer combination model of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] In the drawings of the specific embodiments of the present invention, in order to better and more clearly describe the working principles of the various components in the system, the connection relationship of the various parts in the device is shown, which only clearly distinguishes the relative position relationship between the various components, and cannot constitute a limitation on the signal transmission direction, connection sequence and size, dimensions and shape of the components or structures.

[0020] like Figure 1 As shown, it is a schematic diagram of the process of the method for predicting antimicrobial peptides in bacteria based on deep learning of the present invention, and the method for predicting antimicrobial peptides in bacteria comprises the following steps: S1. Collect antimicrobial peptide sequences from the antimicrobial peptide database as the positive data set, and collect protein sequences from the protein database as the negative data set for training, such as Figure 1 Step ①.

[0021] Antimicrobial peptide sequences were collected from multiple antimicrobial peptide databases such as APD3, DRAMP, DBAASP, and dbAMP as positive data sets. Based on the length distribution of antimicrobial peptide sequences with antibacterial activity, the sequence length in the data set was limited to 5-40 amino acids. In addition, to prevent the rationality and interpretability between antimicrobial peptide sequences from uncommon sources, only sequences derived from animals, plants, bacteria, and fungi were retained. Sequences that were annotated as verified sequences or sequences with MIC measurement results in the database were manually retrieved from the sequences to reduce the interference of potentially inactive or low-activity non-verified sequences on the characteristics of effective antimicrobial peptide sequences. In addition, amino acid sequences without antimicrobial, antibiotic, bactericide, etc. activity annotations were downloaded from the Uniprot protein database as negative data sets.

[0022] S2. Convert all amino acid sequences in the positive and negative data sets into vector sequences as model input according to the number-amino acid conversion table.

[0023] To ensure the rationality and accuracy of the natural sequence source, peptide sequences containing anything other than the 20 natural amino acids were removed to ensure the biological significance of the dataset used. CD-HIT software (similarity threshold set to 1) was used to remove identical redundant sequences from all collected positive and negative datasets to ensure the uniqueness of the data and reduce noise in model training.

[0024] From the collected negative data set, 10 times the number of sequences in the positive data set are extracted and merged with the positive data set as the sample set. Figure 1 As shown in step ②, distributed sampling is used to divide the sample set into training set, validation set and test set in a ratio of 6:2:2. This ratio allocation helps the generalization ability and performance evaluation of the model.

[0025] In addition, the key processing step is to convert the amino acid sequence into a digital embedding format as the input of the model. We correspond each amino acid to a number and define the following correspondence for the 20 natural amino acids. The amino acid sequence is digitized to simplify the data, making the model training more efficient and reducing the computational complexity. At the same time, the sequence with less than 40 amino acids is padded to a uniform length with "0" on the right end of the sequence to standardize the input sequence.

[0026] encode = { 'A': "1", 'C': "2", 'D': "3", 'E': "4", 'F': "5", 'G': "6", 'H': "7", 'I': "8", 'K': "9", 'L': "10", 'M': "11", 'N': "12", 'P': "13", 'Q': "14", 'R': "15", 'S': "16", 'T': "17", 'V': "18", 'W': "19", 'Y': "20" }.

[0027] S3. Construct a multi-layer combination model, input the vector sequence, and output a predicted probability value, so as to determine the antimicrobial peptide sequence with antimicrobial activity.

[0028] like Figure 1 As shown in step ③, a multi-layer combination model is constructed. The input is a vector sequence of amino acids. The local features are extracted through the convolution layer, and the features are weighted summed in combination with the bidirectional long short-term memory layer and the attention mechanism to capture the global information of the sequence. Finally, the classification prediction is completed through the fully connected layer. The model loss calculation uses the binary cross entropy loss function, and the accuracy and other indicators are monitored during the training process to dynamically adjust the learning rate or optimizer parameters.

[0029] Construct a multi-layer combination model consisting of an embedding layer, a convolutional layer, a bidirectional long short-term memory layer, a dropout layer, an attention layer, and a fully connected layer, such as Figure 2 The structural schematic diagram of the multi-layer combination model of the present invention is shown as follows: The input sequence of the multi-layer combination model is a digital embedding format of a standardized candidate antimicrobial peptide sequence.

[0030] First, the input sequence is converted into a digital embedding vector format through the embedding layer so that it can be used as the input of the convolution layer. The relative mapping relationship uses the following formula: In this formula, is the embedding representation of the entire amino acid sequence A, which is a set of multiple embedding vectors. Each digitized amino acid is mapped to a vector, where i represents the position of the amino acid in the mapping and n is the number of amino acids. The integer index in the input sequence is mapped to the corresponding high-dimensional embedding vector by creating a two-dimensional weight matrix with the dimension of the antimicrobial peptide sequence length multiplied by the embedding vector dimension. Similar categories are relatively close in the vector space, thereby capturing the relationship between them.

[0031] After the sequence is embedded, the one-dimensional convolution kernel of the one-dimensional convolution layer slides on the sequence to extract the local features of the antimicrobial peptide. The convolution kernel performs a weighted summation of the adjacent amino acid embedding vectors to generate a feature map to capture the interactions and local patterns between amino acids. The convolution result is then processed using a nonlinear activation function (such as ReLU) to introduce nonlinear characteristics. The feature map is then reduced in dimensionality through a maximum pooling layer to reduce the amount of computation and retain important features. To realize the feature extraction of antimicrobial peptide sequences, the formula is as follows: in, represents the output value at position i after the convolution operation, Represents the output value at position i after the pooling operation, x is the input vector, w is the convolution kernel, b is the bias term, i is the position index in the sequence, j is the position index in the convolution kernel, k is the size of the convolution kernel, p is the maximum pooling window size, and s is the step size.

[0032] After the pooling layer, in order to prevent overfitting caused by too many model parameters and reduce the model's dependence on specific neurons, so that the model can learn more general features, a dropout layer is further introduced. By randomly "dropping" a certain proportion of neurons during the training process, they do not participate in the calculation during forward propagation, simplifying the model without adding too many additional parameters, making the model easier to train and less prone to overfitting, prompting the model to learn more robust features of antimicrobial peptides.

[0033] After the dropout layer, two bidirectional long short-term memory layers are introduced. It combines two long short-term memory layers and can process the forward information (from left to right) and reverse information (from right to left) of the sequence at the same time. The output of each time step considers both directions of context information, which can solve the gradient vanishing and explosion problems of traditional RNN in long sequence learning. The forward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the backward activation vectors of the input gate, forget gate, and output gate, respectively. is the backward candidate memory unit, is the state of the backward memory unit at the current time step, is the backward hidden state of the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the backward input vector of the input sequence at time step t, yes To the backward weight matrix of each gate, is the backward hidden state at time step t-1 To the backward weight matrix of each gate, is the back bias term of each gate.

[0034] The backward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the forward activation vectors of the input gate, forget gate, and output gate, respectively. is the forward candidate memory unit, is the state of the forward memory unit at the current time step, is the forward hidden state at the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the forward input vector of the input sequence at time step t, yes To the forward weight matrix of each gate, is the forward hidden state at time step t+1 To the forward weight matrix of each gate, is the forward bias term for each gate.

[0035] By combining the forward and backward features, the bidirectional long short-term memory layer can learn more comprehensive and accurate sequence features, further extract sequence context information and sequence potential features from the important features retained by the convolutional pooling layer, and improve the performance of the model. Finally, the forward and backward hidden states are concatenated into the output of the bidirectional long short-term memory network: The attention layer is introduced to assign weights to different features of the output sequence information of the bidirectional long short-term memory layer, so that the model can dynamically focus on the most relevant information in the current context, thereby improving the ability to process complex data. The calculation process of the attention layer is as follows: Where Q is the query vector, K is the key vector, and V is the value vector. is the length of the key vector. The value vector is weighted and summed according to the attention weights normalized by Softmax.

[0036] Finally, a three-layer fully connected neural network is constructed to connect the attention layer to form a complete multi-layer combination model. The number of neurons is 64, 20, and 1 respectively. The first two layers use the ReLU activation function to accelerate convergence and improve performance, while the last layer uses the Sigmoid activation function to output the model's probability prediction of whether the input sequence is an antimicrobial peptide. During the training process of the multi-layer combination model, the binary cross entropy loss function is used to calculate the loss value of the model. By calculating the logarithmic loss between the predicted probability and the actual category for each sample, the model can accurately quantify the difference between the predicted result and the true value. At the same time, the model accuracy is used as a monitor to evaluate and quantify the quality of the model output: in Represents the loss value of a single sample, represents the average loss value of the entire data set, y and They represent the true label and predicted value of the sequence respectively, and m represents the number of samples.

[0037] Finally, the model classifies the antimicrobial peptide sequence based on the binary probability value output by the Sigmoid activation function: when the probability value is >0.5, the sequence is judged as an antimicrobial peptide sequence; when the probability value is ≤0.5, it is judged as a non-antimicrobial peptide sequence.

[0038] In a preferred embodiment, the multi-layer combination model is evaluated by a five-fold cross-validation method. The model weight matrix and model structure are adjusted by monitoring the model accuracy and combining the loss value. At the same time, the differences between different models are evaluated to determine the optimal model parameter combination, and the model generalization ability is tested on the test set.

[0039] The evaluation of the multi-layer combination model uses a five-fold cross-validation method to ensure that each sample is used as a validation set at least once. Make full use of every part of the data set to provide more stable and reliable evaluation results. The model evaluation indicators use accuracy, precision, recall, Matthews correlation coefficient (MCC) and area under the precision-recall curve (AUPRC). The specific formula is as follows: True positives (TP) indicate the number of positive samples correctly predicted as positive by the model; true negatives (TN) indicate the number of negative samples correctly predicted as negative by the model; false positives (FP) indicate the number of negative samples incorrectly predicted as positive by the model; and false negatives (FN) indicate the number of positive samples incorrectly predicted as negative by the model.

[0040] Accuracy indicates the ratio of all correct predictions to the total number of samples; Precision indicates the ratio of correct predictions among all positive example predictions, which is the most important indicator in this model; Recall indicates the ratio of the number of predicted positive examples to the total number of actual positive examples; MCC comprehensively considers TP, TN, FP and FN, and has better robustness for unbalanced data sets; AUPRC evaluates the performance of binary classification models under unbalanced data sets by calculating the precision and recall rates under different thresholds.

[0041] For model hyperparameter setting and optimization ( Figure 1 In step ④), the default parameters of the ADAM optimizer are used based on experience (the learning rate is set to 1e-4), the number of iterations is set to 100, and the early stopping strategy (patience value is 5) is used to end the training when the model performance reaches the optimal level. After setting different range combinations for parameters such as the model regularization parameter (0.1-0.3), batch size (8-256), and the number of neurons in each layer, the global optimal solution is found with fewer computing resources through the random search method (search times 200, 5-fold cross validation), and the optimal hyperparameter combination is automatically searched. The model is repeatedly trained using the training set and validation set data divided by the above method to adjust the weight matrix until a stable minimum loss value and the highest accuracy are obtained, and the optimal model parameters with the best performance structure are determined ( Figure 1 Step ⑤).

[0042] Finally, the performance of the multi-layer combination model on unseen data was verified by evaluation on the test set. The results showed that the model performed well in key indicators such as accuracy and MCC. This shows that the constructed model can accurately capture the potential characteristics of antimicrobial peptide sequences, while having good generalization ability and maintaining excellent prediction performance on new data sets ( Figure 1 This model is determined as the final model constructed ( Figure 1 Step ⑦ in the above example). At the same time, the same test set is used to test the performance of other known models for comparison. Compared with other models, the model we built has a greater advantage in parameters such as accuracy.

[0043] For the candidate sequence to be predicted ( Figure 1 In step ⑧ of S2, the model is converted into a digital vector input format according to the method in S2. The peptide sequence with antimicrobial peptide activity is predicted using the model optimized by the above method ( Figure 1 For the peptide sequences predicted to be antimicrobial peptides, the 60 predicted peptide sequences were synthesized and verified by antibacterial experiments, and 54 of them were confirmed to have real antibacterial activity.

[0044] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0045] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting antimicrobial peptides in bacteria based on deep learning, characterized in that: The steps include: S1. Collect antimicrobial peptide sequences from multiple antimicrobial peptide databases as positive data sets, and collect protein sequences from protein databases as negative data sets for training; S2, converting all amino acid sequences in the positive and negative data sets into vector sequences as model input according to the number-amino acid conversion table; S3. Construct a multi-layer combination model, input the vector sequence, and output a predicted probability value, so as to determine the antimicrobial peptide sequence with antimicrobial activity.

2. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 1, characterized in that: In step S1, antimicrobial peptide sequences are collected from multiple antimicrobial peptide databases such as APD3, DRAMP, DBAASP, and dbAMP as positive data sets, and amino acid sequences without antimicrobial, antibiotic, or bactericide annotations are downloaded from the Uniprot protein database as negative data sets.

3. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 1, characterized in that: In step S3, the relative mapping relationship adopts the following formula: in, are multiple embedding vectors of amino acid sequence A A set of, where i represents the position of the amino acid in the mapping, and n is the number of amino acid sequences; The antimicrobial peptide sequence feature extraction formula is as follows: in, represents the output value at position i after the convolution operation, represents the output value at position i after the pooling operation, x is the input vector, w is the convolution kernel, b is the bias term, j is the position index in the convolution kernel, k is the size of the convolution kernel, p is the maximum pooling window size, and s is the step size.

4. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 3, characterized in that: The forward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the backward activation vectors of the input gate, forget gate, and output gate, respectively. is the backward candidate memory unit, is the state of the backward memory unit at the current time step, is the backward hidden state of the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the backward input vector of the input sequence at time step t, yes To the backward weight matrix of each gate, is the backward hidden state at time step t-1 To the backward weight matrix of each gate, is the back bias term of each gate.

5. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 4, characterized in that: The backward long short-term memory calculation formula is as follows: Where → and ← represent the update and transfer directions of LSTM in the time series, respectively. are the forward activation vectors of the input gate, forget gate, and output gate, respectively. is the forward candidate memory unit, is the state of the forward memory unit at the current time step, is the forward hidden state at the current time step, is the sigmoid function, tanh is the hyperbolic tangent function, is the forward input vector of the input sequence at time step t, yes To the forward weight matrix of each gate, is the forward hidden state at time step t+1 To the forward weight matrix of each gate, is the forward bias term for each gate.

6. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 5, characterized in that: Concatenate the forward and backward hidden states into the output of the bidirectional LSTM network: 。 7. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 6, characterized in that: The binary cross entropy loss function calculates the loss expression as: in Represents the loss value of a single sample, represents the average loss value of the entire data set, y and They represent the true label and predicted value of the sequence respectively, and m represents the number of samples.

8. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 7, characterized in that: Based on the binary probability value output by the Sigmoid activation function, the classification of antimicrobial peptide sequences is realized: when the probability value is > 0.5, the sequence is judged as an antimicrobial peptide sequence; when the probability value is ≤ 0.5, it is judged as a non-antimicrobial peptide sequence.

9. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 1, characterized in that: In step S2, sequences 10 times the number of the positive data set are extracted from the collected negative data set and merged with the positive data set into a sample set, and distributed sampling is used to cut the sample set into a training set, a validation set, and a test set in a ratio of 6:2:

2.

10. The method for predicting antimicrobial peptides in bacteria based on deep learning according to claim 1, characterized in that: The five-fold cross-validation method is used to evaluate the multi-layer combination model and automatically search for the optimal hyperparameter combination.

Images