Prediction method and system of antibacterial peptide based on deep learning model

Abstract

The invention discloses an antibacterial peptide prediction method and system based on a deep learning model, and the method comprises the steps: building a corresponding sample sequence table according to a positive sample set and a negative sample set in an international antibacterial peptide database and a protein database; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, and multi-dimensional control is performed based on the combination of the sample set sequence table and the frequency feature table, so that the accuracy of the combination of the sample set sequence table and the frequency feature table is ensured; furthermore, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the prediction probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, so that learning of the feature matrix table of the amino acid sequence and the deep learning model is facilitated, and accurate prediction of the antibacterial peptide is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of antibacterial peptides, and in particular to a prediction method and system for antibacterial peptides based on a deep learning model. Background Art

[0002] With the development of science and technology, in the process of developing new antibacterial agents, the broad antibacterial activity, stability and low host cytotoxicity of antibacterial peptides provide a potential solution to the antibiotic resistance crisis; antibacterial peptides are short peptides with diverse structures, which are a universal feature of the defense system of almost all life forms, and are represented in organisms ranging from bacteria to plants, invertebrates and vertebrates. In the existing technology, antibacterial peptides are artificially predicted, and it is impossible to achieve accurate prediction of antibacterial peptides. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides a prediction method and system for antibacterial peptides based on a deep learning model, locates a long-lived population over 100 years old, and collects amino acid sequences in the long-lived population; in an international antimicrobial peptide database and a protein database, a corresponding sample sequence table is constructed according to a positive sample set and a negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, so as to introduce an overall consideration of the combination of the sample set sequence table and the frequency feature table, and multi-dimensional control is performed based on the combination of the sample set sequence table and the frequency feature table, thereby ensuring the accuracy of the combination of the sample set sequence table and the frequency feature table; further, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, so as to facilitate the learning of the feature matrix table of the amino acid sequence and the deep learning model, thereby ensuring the accuracy of the predicted probability value of the antibacterial peptide, and realizing accurate prediction of the antibacterial peptide.

[0004] In order to solve the above technical problems, an embodiment of the present invention provides a prediction method for antibacterial peptides based on a deep learning model, which is applied to the prediction scenario of antibacterial peptides;

[0005] The prediction method of antibacterial peptides based on the deep learning model comprises:

[0006] Locate the long-lived population over 100 years old and collect amino acid sequences from this long-lived population;

[0007] In the international antimicrobial peptide database and protein database, the corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; the characteristic matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency characteristic table;

[0008] A feature matrix table and a deep learning model are matched with the amino acid sequence, and a predicted probability value of the antibacterial peptide is output based on the feature matrix table and the deep learning model of the amino acid sequence.

[0009] Optionally, the method of locating a long-lived population over 100 years old and collecting amino acid sequences from the long-lived population includes:

[0010] Targeting the long-lived population above 100 years old;

[0011] Collect biological samples from people who are over 100 years old;

[0012] Extracting DNA based on the biological sample and performing metagenomic sequencing;

[0013] Obtain amino acid sequences from long-lived populations.

[0014] Optionally, in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; and a feature matrix table of amino acid sequences is formed according to the combination of the sample set sequence table and the frequency feature table, including:

[0015] Collect international antimicrobial peptide database and protein database;

[0016] In the international antimicrobial peptide database and protein database, positive sample sets and negative sample sets of antibacterial peptides were constructed based on the antimicrobial peptide sequence set and the non-antimicrobial peptide sequence set;

[0017] Construct a corresponding sample sequence table according to the positive sample set and the negative sample set;

[0018] A feature matrix table of amino acid sequences is formed by combining the sample set sequence table with the frequency feature table.

[0019] Optionally, in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; and a feature matrix table of amino acid sequences is formed according to the combination of the sample set sequence table and the frequency feature table, further comprising:

[0020] For the frequency feature table:

[0021] Calculate each amino acid sequence in the positive sample set or negative sample, and output the frequency feature of each amino acid. A frequency feature table is formed based on the frequency feature of each amino acid. The calculation formula is as follows:

[0022]

[0023] Among them, AAP ij is the frequency of occurrence of the ith amino acid in the jth sequence; A ij represents the number of times the i-th amino acid appears in the j-th sequence;

[0024] The sample set sequence table and the frequency feature table are combined to form the amino acid sequence and its corresponding feature matrix table G, as shown below:

[0025]

[0026] Each row represents an amino acid sequence, and each element represents an amino acid and its frequency in the sequence.

[0027] Optionally, the feature matrix table and deep learning model for matching amino acid sequences, outputting a predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model for matching amino acid sequences, include:

[0028] The characteristic matrix table of the fixed amino acid sequence;

[0029] Feature matrix table and deep learning model for matching amino acid sequences;

[0030] Collect the feature matrix table of amino acid sequences and the matching coefficient between the deep learning model;

[0031] If the matching coefficient is greater than the preset matching coefficient threshold, the feature matrix table of the amino acid sequence and the deep learning model are in an adaptation state, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, wherein the positive sample set and the negative sample set are divided into a training set and a test set according to a preset ratio, respectively, and the amino acid sequence in the training set is input into the deep learning model for training.

[0032] Optionally, the feature matrix table and deep learning model for matching amino acid sequences, outputting the predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model of the amino acid sequence, further comprises:

[0033] When the deep learning model is an ATT model, the amino acid sequence is taken as input, and each amino acid is represented by an index or a one-hot encoding;

[0034] For the embedding layer, the one-hot encoding of each amino acid is converted into a dense vector representation using the embedding layer; the output of the embedding layer is a low-dimensional vector, and the embedding layer contains a learnable embedding matrix, where each row of the matrix corresponds to the embedding vector of an amino acid;

[0035] Introduce an attention layer to calculate the weight of each position in the sequence, emphasizing the parts that are important for the classification task; a multi-head attention mechanism can be used to improve the model's ability to capture information from different subspaces; each head focuses on a different part of the input sequence and generates a weighted sum output; calculate the attention weight of each amino acid in each head, usually using scaled dot product attention or a more complex function; the attention weight can be calculated using the following formula:

[0036]

[0037] e ij =a(W h h i ,W s s j )

[0038] Among them, e ij is the energy score calculated by the attention mechanism, h i is the embedding vector of the ith amino acid in the input sequence, s j is the context vector of the jth position, a is a differentiable function, W h and W s is a learnable weight matrix;

[0039] The outputs of all heads are combined according to the attention weights, usually by concatenation or summation, to obtain a fixed-length vector that contains the weighted information of the entire sequence;

[0040] Pass the weighted summed vector through one or more fully connected layers to extract higher-level features;

[0041] The softmax activation function is used to output the predicted probability value of the antibacterial peptide.

[0042] Optionally, the feature matrix table and deep learning model for matching amino acid sequences, outputting the predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model of the amino acid sequence, further comprises:

[0043] Use the test set to evaluate the ATT model. Usually, the accuracy, precision, recall, and F1 score indicators are used to evaluate the model performance.

[0044] Among them, accuracy refers to the ratio of the total number of samples correctly predicted by the model to the total number of samples:

[0045]

[0046] Among them, TP stands for True Positive, TN stands for True Negative, FP stands for False Positive, and FN stands for False Negative.

[0047] Precision refers to the proportion of samples that the model predicts to be positive that are actually positive:

[0048]

[0049] Recall refers to the proportion of samples that are actually positive examples that the model correctly predicts as positive examples:

[0050]

[0051] The F1 score is the harmonic mean of precision and recall, which is used to comprehensively measure the accuracy and robustness of the model:

[0052]

[0053] After that, the model is tuned according to the evaluation results, mainly by adjusting hyperparameters such as learning rate, batch size, and parameters in the attention mechanism to optimize the model. Finally, the best model is selected based on the evaluation results.

[0054] Optionally, the antibacterial peptide prediction method based on the deep learning model also includes:

[0055] Collect stool samples from long-lived people over 100 years old, extract DNA from the samples, and perform metagenomic sequencing to obtain the original sequences;

[0056] Trimmomatic was used to remove the adapters from the original sequences, and bowtie2 was used to remove the host sequences. After that, the data without the host sequences were assembled, and the assembly effect was evaluated using QUAST, and sequences with a length of less than 500 were filtered out;

[0057] Prodigal software was used to predict ORFs from metagenomic data to obtain an ORFs dataset;

[0058] The deep learning model predicts the ORFs dataset to obtain a set of candidate antibacterial peptides;

[0059] PALADIN software and SAMtools software were used to calculate all candidate AMPs in the candidate antibacterial peptide set and define the abundance of all candidate AMPs in the candidate antibacterial peptide set in the sample;

[0060] Spearman correlation analysis was used to calculate the correlation between potential antibacterial peptides and bacteria, and antibacterial peptide sequences with higher correlation were selected and compared with the international antibacterial peptide database to obtain key antibacterial peptide sequences with higher matching degree.

[0061] Optionally, the use of Prodigal software to perform ORFs prediction on the metagenomic data to obtain an ORFs data set includes:

[0062] Prodigal builds and trains a hidden Markov model based on statistical features in the genome, which can identify coding and non-coding regions of sequences in the genome;

[0063] For the input longevity source metagenomic data, the Markov model will scan the genome sequence and predict ORFs; thereafter, the start and end positions of each ORF are determined based on the HMM score and certain heuristic rules;

[0064] During the prediction process, you can adjust multiple parameters to increase the reliability of the prediction, such as setting the type of translation table, controlling the size of the training set, setting the format of the output file, etc.

[0065] After the prediction is completed, each ORF contains a score, which reflects the possibility that the ORF is a true coding region. Thresholds can be set based on these scores to filter out ORFs with low scores, thereby improving the accuracy of the prediction.

[0066] This process will obtain a dataset of predicted ORFs from metagenomic data;

[0067] The deep learning model predicts the ORFs dataset to obtain a candidate antibacterial peptide set, including:

[0068] The amino acid sequences in the ORFs dataset are input into the ATT model, and the model will output a probability value for each ORF that is an antibacterial peptide; when the value is greater than 0.5, the ORF is predicted to be an antibacterial peptide, otherwise it is a non-antibacterial peptide;

[0069] All ORFs in the ORFs data set with an output probability greater than 0.5 from the ATT model are formed into a set of predicted amino acid sequences, which is a set of candidate antibacterial peptides.

[0070] In addition, the embodiment of the present invention also provides a prediction method and system for antibacterial peptides based on a deep learning model, wherein the precise bidirectional matching system for people and positions includes:

[0071] A collection model is used to locate a long-lived population over 100 years old and collect amino acid sequences from this long-lived population;

[0072] The feature matrix table model is used to construct the corresponding sample sequence table according to the positive sample set and the negative sample set in the international antimicrobial peptide database and protein database; the feature matrix table of the amino acid sequence is formed by combining the sample set sequence table with the frequency feature table;

[0073] The prediction module is used to match the feature matrix table of the amino acid sequence and the deep learning model, and output the predicted probability value of the antibacterial peptide based on the feature matrix table of the amino acid sequence and the deep learning model.

[0074] In an embodiment of the present invention, by the method in the embodiment of the present invention, a long-lived population over 100 years old is located, and amino acid sequences in the long-lived population are collected; in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, so as to introduce an overall consideration of the combination of the sample set sequence table and the frequency feature table, and multi-dimensional control is performed based on the combination of the sample set sequence table and the frequency feature table to ensure the accuracy of the combination of the sample set sequence table and the frequency feature table; further, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, so as to facilitate the learning of the feature matrix table of the amino acid sequence and the deep learning model, thereby ensuring the accuracy of the predicted probability value of the antibacterial peptide and achieving accurate prediction of the antibacterial peptide. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0076] Figure 1 is a schematic diagram of a process for predicting antibacterial peptides based on a deep learning model in an embodiment of the present invention;

[0077] Figure 2 is a schematic flow chart of S11 in the method for predicting antibacterial peptides based on a deep learning model in an embodiment of the present invention;

[0078] Figure 3 is a schematic diagram of the process of S12 in the method for predicting antibacterial peptides based on a deep learning model in an embodiment of the present invention;

[0079] Figure 4 is a schematic diagram of the process of S13 in the method for predicting antibacterial peptides based on a deep learning model in an embodiment of the present invention;

[0080] Figure 5 Schematic diagram of the structural composition of the prediction system of antibacterial peptides based on the deep learning model in an embodiment of the present invention;

[0081] Figure 6 The figure is a hardware diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] See also Figures 1 to 6 , a prediction method for antibacterial peptides based on a deep learning model, applied to the prediction scenario of antibacterial peptides; the prediction method for antibacterial peptides based on a deep learning model includes:

[0084] Step S11: locating a long-lived population over 100 years old, and collecting amino acid sequences from the long-lived population;

[0085] Step S12: in the international antimicrobial peptide database and protein database, construct a corresponding sample sequence table according to the positive sample set and the negative sample set; form a feature matrix table of the amino acid sequence according to the combination of the sample set sequence table and the frequency feature table;

[0086] Step S13: Match the feature matrix table of the amino acid sequence and the deep learning model, and output the predicted probability value of the antibacterial peptide based on the feature matrix table of the amino acid sequence and the deep learning model.

[0087] In an embodiment of the present invention, by the method in the embodiment of the present invention, a long-lived population over 100 years old is located, and amino acid sequences in the long-lived population are collected; in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, so as to introduce an overall consideration of the combination of the sample set sequence table and the frequency feature table, and multi-dimensional control is performed based on the combination of the sample set sequence table and the frequency feature table to ensure the accuracy of the combination of the sample set sequence table and the frequency feature table; further, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, so as to facilitate the learning of the feature matrix table of the amino acid sequence and the deep learning model, thereby ensuring the accuracy of the predicted probability value of the antibacterial peptide and achieving accurate prediction of the antibacterial peptide.

[0088] refer to Figure 2 , in step S11, locating a long-lived population over 100 years old, and collecting amino acid sequences from the long-lived population;

[0089] In the specific implementation process of the present invention, the specific steps may be:

[0090] S111: Targeting the long-lived population above 100 years old;

[0091] S112: Collect biological samples from people over 100 years old;

[0092] S113: extracting DNA based on the biological sample and performing metagenomic sequencing;

[0093] S114: Obtain amino acid sequences in long-lived people.

[0094] In the embodiments of the present application, the long-lived population over 100 years old is located and controlled, so as to facilitate the control of amino acid sequences in the long-lived population over 100 years old.

[0095] Therefore, biological samples from long-lived people over 100 years old are collected; DNA is extracted based on the biological samples, and metagenomic sequencing is performed; the amino acid sequences in the long-lived people are obtained, so as to further control the amino acid sequences in the long-lived people and ensure the subsequent processing of the amino acid sequences in the long-lived people.

[0096] refer to Figure 3 In step S12, in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table;

[0097] In the specific implementation process of the present invention, the specific steps may be:

[0098] S121: Collection of international antimicrobial peptide database and protein database;

[0099] S122: In the international antimicrobial peptide database and protein database, the corresponding amino acid sequence data are matched based on the amino acid sequences in the long-lived population;

[0100] S123: performing data preprocessing based on the amino acid sequence data;

[0101] S124: outputting a positive sample set and a negative sample set of antibacterial peptides based on data preprocessing of the amino acid sequence data;

[0102] S125: constructing a corresponding sample sequence table according to the positive sample set and the negative sample set;

[0103] S126: Forming a feature matrix table of amino acid sequences according to the combination of the sample set sequence table and the frequency feature table.

[0104] In the embodiments of the present application, the international antimicrobial peptide database and protein database are collected to facilitate data collection for the international antimicrobial peptide database and protein database. At this time, in the international antimicrobial peptide database and protein database, the corresponding amino acid sequence data are matched based on the amino acid sequence in the long-lived population, so as to introduce the amino acid sequence data, thereby controlling the amino acid sequence data.

[0105] At this time, data preprocessing is performed based on the amino acid sequence data; positive sample sets and negative sample sets of antibacterial peptides are output based on the data preprocessing of the amino acid sequence data; optionally, sequence tag identification is performed on the antibacterial peptide sequence set, and sequences that only contain antibacterial tags and are 6 to 299 AAs in length are retained, and the sequences do not contain non-essential amino acids, do not contain signal peptides, and cannot be homologous to ribosomal amino acid sequences, thereby establishing a positive sample set.

[0106] The selected sequence features are: sequence length is 6-299AAs, does not contain non-essential amino acids, does not contain signal peptides, cannot be homologous to ribosomal amino acid sequences, does not contain antibacterial, antibiotic, antiviral, antifungal, effector, excretion tags, and contains cytoplasmic or nuclear tags in the subcellular location. The selected sequences are used to establish the initial negative sample set (IN-negative sample set). The positive sample set is grouped according to the sequence length, and the length groups include 6-50AAs, 51-100AAs, 101-150AAs, 151-200AAs, 201-250AAs, and 251-299AAs. The number of each length segment in the positive sample set is calculated, X6, X51, X101, X151, X201, and X251. The IN-negative sample set is grouped according to the above sequence lengths, and the same number of sequences as in the corresponding group of the positive sample set are randomly selected from the corresponding group, and the selected sequences are used to construct the negative sample set.

[0107] Furthermore, the positive sample set and the negative sample set are associated, and a corresponding sample sequence table is constructed based on the positive sample set and the negative sample set. The sample set sequence table is further managed and controlled, so as to form a feature matrix table of the amino acid sequence based on the combination of the sample set sequence table and the frequency feature table, so as to facilitate multi-dimensional control through the combination of the sample set sequence table and the frequency feature table, thereby ensuring the accuracy of the feature matrix table of the amino acid sequence.

[0108] In addition, since the lengths of the amino acid sequences are not uniform, a special filler (such as 0 or a specific padding token) is needed to uniformly fill shorter amino acid sequences for identifiability, such as filling sequences with zero values ​​for lengths less than 299, and unifying all lengths to 299. The filled amino acid sequence table is recorded as the original sequence table M.

[0109] For the frequency feature table:

[0110] Calculate each amino acid sequence in the positive sample set or negative sample, and output the frequency feature of each amino acid. A frequency feature table is formed based on the frequency feature of each amino acid. The calculation formula is as follows:

[0111]

[0112] Among them, AAP ij is the frequency of occurrence of the ith amino acid in the jth sequence; Aij represents the number of times the ith amino acid appears in the jth sequence;

[0113] The sample set sequence table and the frequency feature table are combined to form the amino acid sequence and its corresponding feature matrix table G, as shown below:

[0114]

[0115] Each row represents an amino acid sequence, and each element represents an amino acid in the sequence and its frequency. For example, G(0.2) means that the first amino acid in the sequence is G, and its frequency is 0.2. An element of 0(0) means that there is no amino acid at that position in the sequence, and its corresponding frequency is uniformly set to 0.

[0116] refer to Figure 4 In step S13, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model. ;

[0117] In the specific implementation process of the present invention, the specific steps may be:

[0118] S131: Feature matrix table of fixed amino acid sequence;

[0119] S132: Feature matrix table and deep learning model for matching amino acid sequences;

[0120] S133: Collecting the feature matrix table of the amino acid sequence and the matching coefficient between the deep learning model;

[0121] S134: If the matching coefficient is greater than the preset matching coefficient threshold, the feature matrix table of the amino acid sequence and the deep learning model are in an adaptation state, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, wherein the positive sample set and the negative sample set are divided into a training set and a test set according to a preset ratio, respectively, and the amino acid sequence in the training set is input into the deep learning model for training.

[0122] In an embodiment of the present application, a feature matrix table of an amino acid sequence is frozen to facilitate the selection of a deep learning model for the feature matrix table of the amino acid sequence. In the selection of the deep learning model, the feature matrix table of the amino acid sequence and the deep learning model are matched; the matching coefficient between the feature matrix table of the amino acid sequence and the deep learning model is collected, and the degree of matching between the feature matrix table of the amino acid sequence and the deep learning model is presented based on the matching coefficient.

[0123] Therefore, if the matching coefficient is greater than the preset matching coefficient threshold, the feature matrix table of the amino acid sequence and the deep learning model are in an adaptation state, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model to facilitate the learning of the feature matrix table of the amino acid sequence and the deep learning model, thereby ensuring the accuracy of the predicted probability value of the antibacterial peptide and achieving accurate prediction of the antibacterial peptide.

[0124] Among them, the positive sample set and the negative sample set are divided into training set and test set according to preset ratios, and the amino acid sequence in the training set is input into the deep learning model for training.

[0125] In addition, when the deep learning model is an ATT model, the amino acid sequence is taken as input, and each amino acid is represented by an index or one-hot encoding;

[0126] For the embedding layer, the one-hot encoding of each amino acid is converted into a dense vector representation using the embedding layer; the output of the embedding layer is a low-dimensional vector, and the embedding layer contains a learnable embedding matrix, where each row of the matrix corresponds to the embedding vector of an amino acid;

[0127] Introduce an attention layer to calculate the weight of each position in the sequence, emphasizing the parts that are important for the classification task; a multi-head attention mechanism can be used to improve the model's ability to capture information from different subspaces; each head focuses on a different part of the input sequence and generates a weighted sum output; calculate the attention weight of each amino acid in each head, usually using scaled dot product attention or a more complex function; the attention weight can be calculated using the following formula:

[0128]

[0129] e ij=a(W h h i ,W s s j )

[0130] Among them, e ij is the energy score calculated by the attention mechanism, h i is the embedding vector of the ith amino acid in the input sequence, s j is the context vector of the jth position, a is a differentiable function, W h and W s is a learnable weight matrix;

[0131] The outputs of all heads are combined according to the attention weights, usually by concatenation or summation, to obtain a fixed-length vector that contains the weighted information of the entire sequence;

[0132] Pass the weighted summed vector through one or more fully connected layers to extract higher-level features;

[0133] The softmax activation function is used to output the predicted probability value of the antibacterial peptide.

[0134] In addition, the feature matrix table and deep learning model for matching amino acid sequences, outputting predicted probability values ​​of antibacterial peptides based on the feature matrix table and deep learning model for matching amino acid sequences, further include:

[0135] Use the test set to evaluate the ATT model. Usually, the accuracy, precision, recall, and F1 score indicators are used to evaluate the model performance.

[0136] Among them, accuracy refers to the ratio of the total number of samples correctly predicted by the model to the total number of samples:

[0137]

[0138] Among them, TP stands for True Positive, TN stands for True Negative, FP stands for False Positive, and FN stands for False Negative.

[0139] Precision refers to the proportion of samples that the model predicts to be positive that are actually positive:

[0140]

[0141] Recall refers to the proportion of samples that are actually positive examples that the model correctly predicts as positive examples:

[0142]

[0143] The F1 score is the harmonic mean of precision and recall, which is used to comprehensively measure the accuracy and robustness of the model:

[0144]

[0145] After that, the model is tuned according to the evaluation results, mainly by adjusting hyperparameters such as learning rate, batch size, and parameters in the attention mechanism to optimize the model. Finally, the best model is selected based on the evaluation results.

[0146] The prediction method of antibacterial peptides based on the deep learning model also includes:

[0147] Collect stool samples from long-lived people over 100 years old, extract DNA from the samples, and perform metagenomic sequencing to obtain the original sequences;

[0148] Trimmomatic was used to remove the adapters from the original sequences, and bowtie2 was used to remove the host sequences. After that, the data without the host sequences were assembled, and the assembly effect was evaluated using QUAST, and sequences with a length of less than 500 were filtered out;

[0149] Prodigal software was used to predict ORFs from metagenomic data to obtain an ORFs dataset;

[0150] The deep learning model predicts the ORFs dataset to obtain a set of candidate antibacterial peptides;

[0151] PALADIN software and SAMtools software were used to calculate all candidate AMPs in the candidate antibacterial peptide set and define the abundance of all candidate AMPs in the candidate antibacterial peptide set in the sample;

[0152] Spearman correlation analysis was used to calculate the correlation between potential antibacterial peptides and bacteria, and antibacterial peptide sequences with higher correlation were selected and compared with the international antibacterial peptide database to obtain key antibacterial peptide sequences with higher matching degree.

[0153] Wherein, the use of Prodigal software to predict ORFs on the metagenomic data to obtain an ORFs data set includes:

[0154] Prodigal builds and trains a hidden Markov model based on statistical features in the genome, which can identify coding and non-coding regions of sequences in the genome;

[0155] For the input longevity source metagenomic data, the Markov model will scan the genome sequence and predict ORFs; thereafter, the start and end positions of each ORF are determined based on the HMM score and certain heuristic rules;

[0156] During the prediction process, you can adjust multiple parameters to increase the reliability of the prediction, such as setting the type of translation table, controlling the size of the training set, setting the format of the output file, etc.

[0157] After the prediction is completed, each ORF contains a score, which reflects the possibility that the ORF is a true coding region. Thresholds can be set based on these scores to filter out ORFs with low scores, thereby improving the accuracy of the prediction.

[0158] This process will obtain a dataset of predicted ORFs from metagenomic data.

[0159] The deep learning model predicts the ORFs dataset to obtain a candidate antibacterial peptide set, including:

[0160] The amino acid sequences in the ORFs dataset are input into the ATT model, and the model will output a probability value for each ORF that is an antibacterial peptide; when the value is greater than 0.5, the ORF is predicted to be an antibacterial peptide, otherwise it is a non-antibacterial peptide;

[0161] All ORFs in the ORFs data set with an output probability greater than 0.5 from the ATT model are formed into a set of predicted amino acid sequences, which is a set of candidate antibacterial peptides.

[0162] In the specific implementation of the present invention:

[0163] Example 1 Acquisition of known antibacterial peptide data

[0164] This embodiment is to obtain data of antibacterial peptides and construct positive sample and negative sample sets, specifically including:

[0165] Antimicrobial peptide sequence collections were downloaded from multiple known international antimicrobial peptide databases, including APD3, CAMP, LAMP, DADP, DBAASP, DRAMP, etc. In addition, non-antimicrobial peptide sequence collections were downloaded from international protein databases, such as the Uniprot database, and the data used are shown in Table 1.

[0166] Table 1 Antimicrobial peptide data sources

[0167]

[0168] Antibacterial peptide positive samples and negative samples were established based on the antimicrobial peptide sequence set and the non-antimicrobial peptide sequence set.

[0169] Antibacterial peptide sequence set: Sequence tag identification is performed on the antibacterial peptide sequence set, and sequences containing only antibacterial tags and with a length of 6 to 299 AAs are retained. The sequences do not contain non-essential amino acids, do not contain signal peptides, and cannot be homologous to ribosomal amino acid sequences, so as to establish a positive sample set.

[0170] Non-antimicrobial peptide sequence set: The selected sequence features are: sequence length is 6-299AAs, does not contain non-essential amino acids, does not contain signal peptides, cannot be homologous to ribosomal amino acid sequences, does not contain antibacterial, antibiotic, antiviral, antifungal, effector, or excretion tags, and contains cytoplasmic or nuclear tags in the subcellular location. The selected sequences are used to establish the initial negative sample set (IN-negative sample set).

[0171] Example 2: Positive and negative sample data preprocessing

[0172] Before building a deep learning model, the sample set needs to be processed, including data balancing and filling, as follows:

[0173] Data balance of two sample sets: The positive sample set is grouped according to the sequence length. The length groups include 6-50AAs, 51-100AAs, 101-150AAs, 151-200AAs, 201-250AAs, and 251-299AAs. The number of each length segment in the positive sample set is calculated as X6, X 51 , X 101 , X 151 , X 201 and X 251 The IN-negative sample set is grouped according to the above sequence lengths, and the same number of sequences as those in the corresponding group of the positive sample set are randomly selected from the corresponding group, and the selected sequences are used to construct the negative sample set.

[0174] Data filling: Since the lengths of the amino acid sequences are not uniform, special fillers (such as 0 values) are needed to uniformly fill shorter amino acid sequences to make them identifiable, and unify all lengths to 299. The filled amino acid sequence table is recorded as the original sequence table M;

[0175] Amino acid position frequency feature filling: Calculate the frequency of occurrence of each amino acid in each amino acid sequence in the current sequence to display the amino acid frequency feature of the sequence. The formula is as follows:

[0176]

[0177] AAP ij is the frequency of occurrence of the ith amino acid in the jth sequence; Aij represents the number of times the ith amino acid appears in the jth sequence;

[0178] Use a matrix to store the amino acid frequency features corresponding to all amino acid sequence outputs, recorded as a frequency matrix table N;

[0179] The original sequence table M and the matrix table N are combined to form the amino acid sequence and its corresponding feature matrix table G, which is specifically expressed as follows:

[0180]

[0181] In this matrix, each row represents an amino acid sequence, and each element represents an amino acid in the sequence and its frequency. For example, G(0.2) means that the amino acid at the first position in the sequence is G, and its frequency in the sequence is 0.2. An element of 0(0) means that there is no amino acid at that position in the sequence, and its corresponding frequency is uniformly set to 0.

[0182] Example 3 Construction of a deep learning model for identifying antibacterial peptide characteristics

[0183] This part includes model training, evaluation and tuning, as follows:

[0184] Model training:

[0185] The positive sample set and the negative sample set are divided into training set and test set according to the ratio (8:2), and the FASTA sequence in the training set is input into the ATT deep learning model for training. The model is trained through multiple iterations (epochs). In each iteration, the prediction value is calculated by forward propagation, and then the gradient is calculated and the model parameters are updated by back propagation. The cross entropy loss function is selected for the classification task.

[0186] Model evaluation and tuning:

[0187] The ATT model is evaluated using the test set. The model performance is usually evaluated using indicators such as accuracy, precision, recall, and F1 score. The accuracy calculation formula is:

[0188]

[0189] Among them, TP stands for True Positive, TN stands for True Negative, FP stands for False Positive, and FN stands for False Negative.

[0190] The accuracy calculation formula is:

[0191]

[0192] The recall calculation formula is:

[0193]

[0194] The F1 score calculation formula is:

[0195]

[0196] After that, the model is tuned according to the evaluation results, mainly by adjusting the hyperparameters such as learning rate, batch size, and parameters in the attention mechanism to optimize the model. Finally, the best model is selected based on the evaluation results. The performance parameters of the best model are shown in Table 2.

[0197] Table 2 Prediction performance of the ATT model on the test set

[0198]

[0199] Example 4 Acquisition of longevity source intestinal bacteria metagenetic data

[0200] This embodiment is used to obtain longevity gene data, specifically including:

[0201] (1) Acquisition of intestinal metagenomic data from Changshouyuan: Fecal samples from 20 healthy centenarians were collected, and metagenomic sequencing was performed after processing the samples (such as DNA extraction) to obtain offline data.

[0202] (2) Metagenomic data preprocessing: Trimmomatic was used to remove the adapters from the original sequences, and bowtie2 was used to remove the host sequences. After that, the data without the host sequences were assembled (MegaHit), and the assembly effect was evaluated using QUAST, and sequences with a length of less than 500 were filtered out.

[0203] (3) ORFs prediction of representative genomes: Prodigal software was used to predict ORFs of metagenomic data to obtain an ORFs dataset.

[0204] Example 5 Identification of longevity-derived antibacterial peptides

[0205] (1) Prediction of longevity-derived antibacterial peptides: Use the ATT deep learning model constructed above to predict the ORFs set to obtain a set of candidate antibacterial peptides. Specifically, it includes:

[0206] The amino acid sequences in the ORFs set are input into the ATT model. The model will output a probability value of an antibacterial peptide for each ORF. When the value is greater than 0.5, the ORF is predicted to be an antibacterial peptide, otherwise it is a non-antibacterial peptide.

[0207] All ORFs with output probability greater than 0.5 from the ATT model are used to form a set of predicted amino acid sequences, which is the candidate antibacterial peptide set (c_ABP);

[0208] Table 3 Potential antimicrobial peptide sequences from longevity sources (partial)

[0209]

[0210]

[0211] Calculation of the abundance of potential antibacterial peptides from longevity sources: PALADIN software and SAMtools software were used to calculate the abundance of all candidate AMPs in the data set in the sample, mainly by aligning the AMP sequence with the original metagenome sequence and calculating the coverage of the reads.

[0212] Screening of potential antibacterial peptides for key longevity sources: MetaPhlAn2 was used to calculate the relative abundance of bacteria corresponding to representative genomes. Spearman correlation analysis was used to calculate the correlation between potential antibacterial peptides and bacteria, and antibacterial peptide sequences with higher correlation were selected and compared with the international antibacterial peptide database to obtain key antibacterial peptide sequences with higher matching degree.

[0213] This process will enable the identification of potential antibacterial peptide sequences and their characteristics (such as abundance, sequence length, etc.) from samples of long-lived people.

[0214] In an embodiment of the present invention, by the method in the embodiment of the present invention, a long-lived population over 100 years old is located, and amino acid sequences in the long-lived population are collected; in the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, so as to introduce an overall consideration of the combination of the sample set sequence table and the frequency feature table, and multi-dimensional control is performed based on the combination of the sample set sequence table and the frequency feature table to ensure the accuracy of the combination of the sample set sequence table and the frequency feature table; further, the feature matrix table of the amino acid sequence and the deep learning model are matched, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, so as to facilitate the learning of the feature matrix table of the amino acid sequence and the deep learning model, thereby ensuring the accuracy of the predicted probability value of the antibacterial peptide and achieving accurate prediction of the antibacterial peptide.

[0215] See also Figure 5 , Figure 5 It is a schematic diagram of the structural composition of the prediction system of antibacterial peptides based on the deep learning model in an embodiment of the present invention.

[0216] like Figure 5As shown, a prediction system for antibacterial peptides based on a deep learning model, the precise bidirectional matching system for people and positions includes:

[0217] Collection model 21, used to locate a long-lived population over 100 years old and collect amino acid sequences from the long-lived population;

[0218] The feature matrix table model 22 is used to construct a corresponding sample sequence table according to the positive sample set and the negative sample set in the international antimicrobial peptide database and protein database; and to form a feature matrix table of the amino acid sequence according to the combination of the sample set sequence table and the frequency feature table;

[0219] The prediction module 23 is used to match the feature matrix table of the amino acid sequence and the deep learning model, and output the predicted probability value of the antibacterial peptide based on the feature matrix table of the amino acid sequence and the deep learning model.

[0220] See also Figure 6 , refer to the following Figure 6 The electronic device 40 according to this embodiment of the present invention will be described. Figure 6 The electronic device 40 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0221] like Figure 6 As shown, the electronic device 40 is in the form of a general computing device. The components of the electronic device 40 may include but are not limited to: at least one processing unit 41, at least one storage unit 42, and a bus 43 connecting different system components (including the storage unit 42 and the processing unit 41).

[0222] The storage unit stores program codes, which can be executed by the processing unit 41, so that the processing unit 41 executes the steps according to various exemplary embodiments of the present invention described in the above “Embodiment Method” section of this specification.

[0223] The storage unit 42 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 421 and / or a cache memory unit 422 , and may further include a read-only memory unit (ROM) 423 .

[0224] The storage unit 42 may also include a program / utility 424 having a set (at least one) of program modules 425, such program modules 425 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0225] Bus 43 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0226] The electronic device 40 may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device 40, and / or any device that enables the electronic device 40 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 44. Furthermore, the electronic device 40 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 45. Figure 6 As shown, the network adapter 45 communicates with other modules of the electronic device 40 via the bus 43. It should be understood that although Figure 6 Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup planning systems.

[0227] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0228] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc. In addition, it stores computer program instructions, and when the computer program instructions are executed by a computer, the computer executes the above method.

[0229] In addition, the above is a detailed introduction to the prediction system and system of antibacterial peptides based on the deep learning model provided in the embodiments of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A prediction method for antibacterial peptides based on a deep learning model, characterized in that: Application to the prediction scenario of antibacterial peptides; The prediction method of antibacterial peptides based on the deep learning model comprises: Locate the long-lived population over 100 years old and collect amino acid sequences from this long-lived population; In the international antimicrobial peptide database and protein database, the corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; the characteristic matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency characteristic table; A feature matrix table and a deep learning model are matched with the amino acid sequence, and a predicted probability value of the antibacterial peptide is output based on the feature matrix table and the deep learning model of the amino acid sequence.

2. The method for predicting antibacterial peptides based on a deep learning model according to claim 1, characterized in that: The method of locating a long-lived population over 100 years old and collecting amino acid sequences from the long-lived population includes: Targeting the long-lived population above 100 years old; Collect biological samples from people who are over 100 years old; Extracting DNA based on the biological sample and performing metagenomic sequencing; Obtain amino acid sequences from long-lived populations.

3. The method for predicting antibacterial peptides based on a deep learning model according to claim 2, characterized in that: In the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, including: Collect international antimicrobial peptide database and protein database; In the international antimicrobial peptide database and protein database, positive sample sets and negative sample sets of antibacterial peptides were constructed based on the antimicrobial peptide sequence set and the non-antimicrobial peptide sequence set; Construct a corresponding sample sequence table according to the positive sample set and the negative sample set; A feature matrix table of amino acid sequences is formed by combining the sample set sequence table with the frequency feature table.

4. The method for predicting antibacterial peptides based on a deep learning model according to claim 3, characterized in that: In the international antimicrobial peptide database and protein database, a corresponding sample sequence table is constructed according to the positive sample set and the negative sample set; a feature matrix table of the amino acid sequence is formed according to the combination of the sample set sequence table and the frequency feature table, and further includes: For the frequency feature table: Calculate each amino acid sequence in the positive sample set or negative sample, and output the frequency feature of each amino acid. A frequency feature table is formed based on the frequency feature of each amino acid. The calculation formula is as follows: Among them, AAP ij is the frequency of occurrence of the ith amino acid in the jth sequence; A ij represents the number of times the i-th amino acid appears in the j-th sequence; The sample set sequence table and the frequency feature table are combined to form the amino acid sequence and its corresponding feature matrix table G, as shown below: Each row represents an amino acid sequence, and each element represents an amino acid and its frequency in the sequence.

5. The method for predicting antibacterial peptides based on a deep learning model according to claim 1, characterized in that: The feature matrix table and deep learning model for matching amino acid sequences, outputting the predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model of the amino acid sequence, comprises: The characteristic matrix table of the fixed amino acid sequence; Feature matrix table and deep learning model for matching amino acid sequences; Collect the feature matrix table of amino acid sequences and the matching coefficient between the deep learning model; If the matching coefficient is greater than the preset matching coefficient threshold, the feature matrix table of the amino acid sequence and the deep learning model are in an adaptation state, and the predicted probability value of the antibacterial peptide is output based on the feature matrix table of the amino acid sequence and the deep learning model, wherein the positive sample set and the negative sample set are divided into a training set and a test set according to a preset ratio, respectively, and the amino acid sequence in the training set is input into the deep learning model for training.

6. The method for predicting antibacterial peptides based on a deep learning model according to claim 5, characterized in that: The feature matrix table and deep learning model for matching amino acid sequences, outputting the predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model of the amino acid sequence, also include: When the deep learning model is an ATT model, the amino acid sequence is taken as input, and each amino acid is represented by an index or a one-hot encoding; For the embedding layer, the one-hot encoding of each amino acid is converted into a dense vector representation using the embedding layer; the output of the embedding layer is a low-dimensional vector, and the embedding layer contains a learnable embedding matrix, where each row of the matrix corresponds to the embedding vector of an amino acid; Introduce an attention layer to calculate the weight of each position in the sequence, emphasizing the parts that are important for the classification task; a multi-head attention mechanism can be used to improve the model's ability to capture information from different subspaces; each head focuses on a different part of the input sequence and generates a weighted sum output; calculate the attention weight of each amino acid in each head, usually using scaled dot product attention or a more complex function; the attention weight can be calculated using the following formula: have been ij =a(W h h i ,W s s j ) Among them, e ij is the energy score calculated by the attention mechanism, h i is the embedding vector of the ith amino acid in the input sequence, s j is the context vector of the jth position, a is a differentiable function, W h and W s is a learnable weight matrix; The outputs of all heads are combined according to the attention weights, usually by concatenation or summation, to obtain a fixed-length vector that contains the weighted information of the entire sequence; Pass the weighted summed vector through one or more fully connected layers to extract higher-level features; The softmax activation function is used to output the predicted probability value of the antibacterial peptide.

7. The method for predicting antibacterial peptides based on a deep learning model according to claim 6, characterized in that: The feature matrix table and deep learning model for matching amino acid sequences, outputting the predicted probability value of the antibacterial peptide based on the feature matrix table and deep learning model of the amino acid sequence, also include: Use the test set to evaluate the ATT model. Usually, the accuracy, precision, recall, and F1 score indicators are used to evaluate the model performance. Among them, accuracy refers to the ratio of the total number of samples correctly predicted by the model to the total number of samples: Among them, TP stands for True Positive, TN stands for True Negative, FP stands for False Positive, and FN stands for False Negative. Precision refers to the proportion of samples that the model predicts to be positive that are actually positive: Recall refers to the proportion of samples that are actually positive examples that the model correctly predicts as positive examples: The F1 score is the harmonic mean of precision and recall, which is used to comprehensively measure the accuracy and robustness of the model: After that, the model is tuned according to the evaluation results, mainly by adjusting hyperparameters such as learning rate, batch size, and parameters in the attention mechanism to optimize the model. Finally, the best model is selected based on the evaluation results.

8. The method for predicting antibacterial peptides based on a deep learning model according to any one of claims 1 to 7, characterized in that: The prediction method of antibacterial peptides based on the deep learning model also includes: Collect stool samples from long-lived people over 100 years old, extract DNA from the samples, and perform metagenomic sequencing to obtain the original sequences; Trimmomatic was used to remove the adapters from the original sequences, and bowtie2 was used to remove the host sequences. After that, the data without the host sequences were assembled, and the assembly effect was evaluated using QUAST, and sequences with a length of less than 500 were filtered out; Prodigal software was used to predict ORFs from metagenomic data to obtain an ORFs dataset; The deep learning model predicts the ORFs dataset to obtain a set of candidate antibacterial peptides; PALADIN software and SAMtools software were used to calculate all candidate AMPs in the candidate antibacterial peptide set and define the abundance of all candidate AMPs in the candidate antibacterial peptide set in the sample; Spearman correlation analysis was used to calculate the correlation between potential antibacterial peptides and bacteria, and antibacterial peptide sequences with higher correlation were selected and compared with the international antibacterial peptide database to obtain key antibacterial peptide sequences with higher matching degree.

9. The method for predicting antibacterial peptides based on a deep learning model according to claim 8, characterized in that: The ORFs prediction of the metagenomic data using the Prodigal software is performed to obtain an ORFs data set, including: Prodigal builds and trains a hidden Markov model based on statistical features in the genome, which can identify coding and non-coding regions of sequences in the genome; For the input longevity source metagenomic data, the Markov model will scan the genome sequence and predict ORFs; thereafter, the start and end positions of each ORF are determined based on the HMM score and certain heuristic rules; During the prediction process, you can adjust multiple parameters to increase the reliability of the prediction, such as setting the type of translation table, controlling the size of the training set, setting the format of the output file, etc. After the prediction is completed, each ORF contains a score, which reflects the possibility that the ORF is a true coding region. Thresholds can be set based on these scores to filter out ORFs with low scores, thereby improving the accuracy of the prediction. This process will obtain a dataset of predicted ORFs from metagenomic data; The deep learning model predicts the ORFs dataset to obtain a candidate antibacterial peptide set, including: The amino acid sequences in the ORFs dataset are input into the ATT model, and the model will output a probability value for each ORF that is an antibacterial peptide; when the value is greater than 0.5, the ORF is predicted to be an antibacterial peptide, otherwise it is a non-antibacterial peptide; All ORFs in the ORFs data set with an output probability greater than 0.5 from the ATT model are formed into a set of predicted amino acid sequences, which is a set of candidate antibacterial peptides.

10. A prediction system for antibacterial peptides based on a deep learning model, characterized in that: The prediction system of antibacterial peptides based on the deep learning model is applied to the prediction method of antibacterial peptides based on the deep learning model as claimed in any one of claims 1 to 9, and the prediction system of antibacterial peptides based on the deep learning model comprises: A collection model is used to locate a long-lived population over 100 years old and collect amino acid sequences from this long-lived population; The feature matrix table model is used to construct the corresponding sample sequence table according to the positive sample set and the negative sample set in the international antimicrobial peptide database and protein database; the feature matrix table of the amino acid sequence is formed by combining the sample set sequence table with the frequency feature table; The prediction module is used to match the feature matrix table of the amino acid sequence and the deep learning model, and output the predicted probability value of the antibacterial peptide based on the feature matrix table of the amino acid sequence and the deep learning model.

Images