Protein language model-based anticancer peptide prediction method and system

Abstract

The invention relates to an anticancer peptide prediction method and system based on a protein language model. The method comprises the following steps: collecting and screening out anti-cancer peptide samples and non-anti-cancer peptide samples to construct a data set; inputting the data set into a protein language model for feature processing, and extracting high-dimensional biological features in the sample sequence; performing convolution operation on the high-dimensional biological characteristics by using a convolutional neural network through a protein language model, dynamically adjusting the weight of the high-dimensional biological characteristics through a self-attention mechanism, and capturing global characteristics in the peptide sequence; and performing anticancer peptide prediction through a full connection layer in the protein language model to obtain a prediction result. According to the method, feature processing is performed through the protein language model, global features are captured based on the convolutional neural network and the self-attention mechanism, complex feature engineering steps are not needed, and key information in a peptide sequence can be fully excavated, so that the anticancer peptide prediction performance of the model is improved, and the anticancer peptide can be rapidly predicted with low cost and high accuracy.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method and system for predicting anticancer peptides based on a protein language model. Background Art

[0002] Anticancer peptides (ACPs) are a class of short peptides with anticancer properties that can target and kill cancer cells through multiple mechanisms while causing less damage to normal cells. ACPs are usually composed of 5 to 50 amino acids, and their main mechanism of action is to interact with cancer cell membranes, leading to cell membrane lysis or penetration. In addition, ACPs can also exert anticancer effects by inhibiting angiogenesis, inducing tumor cell apoptosis, and targeting key proteins. In recent years, ACPs have attracted widespread attention due to their high efficiency and low toxicity, and have become a hot topic in cancer treatment research. Compared with traditional chemotherapy and radiotherapy, the selective killing effect of ACPs significantly reduces the side effects during treatment and improves the quality of life of patients. In the study of ACPs, it is not only important to identify their existence, but also to predict the strength and activity of ACPs, which is of great significance for optimizing anticancer treatment. Accurately predicting the anticancer activity of ACPs can accelerate drug development, help researchers better understand their mechanism of action, and ultimately promote the implementation of personalized treatment plans. Therefore, the development of accurate and effective ACP prediction tools is of great significance for the future development of cancer treatment.

[0003] At present, the application of anticancer peptides (ACP) is still in the research stage. Traditional cancer treatment methods mainly include chemotherapy, radiotherapy and surgery. Chemotherapy inhibits the proliferation of cancer cells through chemical drugs, while radiotherapy kills cancer cells through high-energy radiation.

[0004] Therefore, traditional cancer treatments are often accompanied by serious side effects, such as damage to normal cells, suppression of the immune system, and a decline in the patient's quality of life. With the rapid development of artificial intelligence technology, anticancer peptide prediction tools based on machine learning and deep learning have gradually emerged. These tools can quickly predict potential anticancer peptides by analyzing protein sequences and physicochemical properties, reducing the time and cost of experimental screening. However, traditional anticancer peptide prediction methods have the problems of high cost and low accuracy. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, a method and system for predicting anticancer peptides based on a protein language model are provided, which can predict anticancer peptides quickly, at low cost and with high accuracy.

[0006] A method for predicting anticancer peptides based on a protein language model, the method comprising:

[0007] Collect and screen anticancer peptide samples and non-anticancer peptide samples from a database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples;

[0008] Inputting the data set into a protein language model, performing feature processing on sample sequences in the data set through the protein language model, and extracting high-dimensional biological features in the sample sequences;

[0009] The protein language model uses a convolutional neural network to perform a convolution operation on the high-dimensional biological features, and dynamically adjusts the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the peptide sequence;

[0010] The high-dimensional biological features and global features are input into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results.

[0011] In one embodiment, anticancer peptide samples and non-anticancer peptide samples are collected and screened from a database, and a data set is constructed based on the anticancer peptide samples and non-anticancer peptide samples, including:

[0012] Collecting anticancer peptides from a database, and eliminating the anticancer peptides by homology to obtain the eliminated anticancer peptides;

[0013] Performing PSSM matrix extraction on the eliminated anticancer peptides to obtain a final anticancer peptide sample;

[0014] Collecting non-anti-cancer peptides from a database, and eliminating the non-anti-cancer peptides by homology to obtain eliminated non-anti-cancer peptides;

[0015] Performing PSSM matrix extraction on the eliminated non-anticancer peptides to obtain a final non-anticancer peptide sample;

[0016] Anti-cancer peptide sequences are selected from the final anti-cancer peptide sample, and non-anti-cancer peptide sequences are selected from the final non-anti-cancer peptide sample to form a data set.

[0017] In one embodiment, the method further comprises:

[0018] The final anti-cancer peptide sample is taken as a positive sample, and the final non-anti-cancer peptide sample is taken as a negative sample;

[0019] Randomly selecting anticancer peptide samples from the positive samples as anticancer peptide sequences, and randomly selecting non-anticancer peptide samples from the negative samples as non-anticancer peptide sequences as a data set;

[0020] The remaining anticancer peptide sequences in the positive samples and the remaining non-anticancer peptide sequences in the negative samples are used as test sets.

[0021] In one embodiment, the data set is input into a protein language model, and the protein language model is used to perform feature processing on the sample sequences in the data set to extract high-dimensional biological features in the sample sequences, including:

[0022] Inputting the data set into a protein language model, and performing feature processing on the anticancer peptide sequences in the data set through the protein language model to capture implicit patterns, long-term dependencies, and contextual information in the anticancer peptide sequences;

[0023] Based on the implicit patterns, long-term dependencies, and contextual information, high-dimensional biological features in the anticancer peptide sequence are generated.

[0024] In one embodiment, the protein language model uses a convolutional neural network to perform a convolution operation on the high-dimensional biological features, and dynamically adjusts the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the anticancer peptide sequence, including:

[0025] Using the protein language model, a convolutional neural network is used to perform a convolution operation on the context information to obtain convolved information;

[0026] The weight of the high-dimensional biological features is adjusted through a self-attention mechanism to identify the long-term dependency relationship;

[0027] Based on the convolved information, the weights of the high-dimensional biological features, and the long-term dependencies, the global context information is integrated to obtain the global features.

[0028] In one embodiment, the context information is convolved using a convolutional neural network through the protein language model to obtain the convolved information, including:

[0029] Using a convolutional neural network to perform a nonlinear transformation on each of the context information to obtain nonlinearly transformed information;

[0030] A one-dimensional convolution is used to perform a convolution operation on the information after the nonlinear transformation, and a batch normalization process is added after each convolution layer, and the convolved information is obtained through the output of the pooling layer.

[0031] In one embodiment, adjusting the weight of the high-dimensional biological feature through a self-attention mechanism includes:

[0032] Performing a linear transformation on the high-dimensional biological features through a self-attention mechanism to obtain a transformation matrix;

[0033] Get the self-attention mechanism weight matrix and get the key vector dimension of each attention head;

[0034] The weight of the high-dimensional biological feature is adjusted based on the transformation matrix, the self-attention mechanism weight matrix, and the key vector dimension.

[0035] In one embodiment, the method further comprises:

[0036] Determining a model evaluation index, and performing a model evaluation on the protein language model according to the model evaluation index to obtain an evaluation result;

[0037] The protein language model is optimized based on the evaluation results.

[0038] A protein language model-based anticancer peptide prediction system, comprising:

[0039] A data set construction module, used to collect and screen anticancer peptide samples and non-anticancer peptide samples from a database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples;

[0040] A high-dimensional biological feature extraction module, used to input the data set into a protein language model, perform feature processing on the sample sequence in the data set through the protein language model, and extract high-dimensional biological features in the sample sequence;

[0041] A global feature extraction module, used to perform a convolution operation on the high-dimensional biological features using a convolutional neural network through the protein language model, and dynamically adjust the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the peptide sequence;

[0042] The anticancer peptide prediction module is used to input the high-dimensional biological features and global features into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results.

[0043] The above-mentioned anticancer peptide prediction method and system based on protein language model perform feature processing through protein language model, extract high-dimensional biological features, capture global features based on convolutional neural network and self-attention mechanism, and do not require complex feature engineering steps. It can fully mine the key information in the peptide sequence, thereby improving the anticancer peptide prediction performance of the model, and can predict anticancer peptides quickly, at low cost and with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 FIG. 1 is an application environment diagram of an anticancer peptide prediction method based on a protein language model in one embodiment;

[0045] Figure 2 FIG. 1 is a flow chart of a method for predicting anticancer peptides based on a protein language model in one embodiment;

[0046] Figure 3A schematic diagram of a model framework of ACP-ESM2 in one embodiment;

[0047] Figure 4 This is a schematic diagram for visualizing features after training;

[0048] Figure 5 This is a schematic diagram of feature visualization before training;

[0049] Figure 6 is a structural block diagram of an anticancer peptide prediction system based on a protein language model in one embodiment;

[0050] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] The anticancer peptide prediction method based on the protein language model provided in the present application example can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can collect and screen anticancer peptide samples and non-anticancer peptide samples from a database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples; the computer device 110 can input the data set into a protein language model, perform feature processing on the sample sequence in the data set through the protein language model, and extract high-dimensional biological features in the sample sequence; the computer device 110 can use a convolutional neural network to perform convolution operations on high-dimensional biological features through the protein language model, and dynamically adjust the weights of high-dimensional biological features through a self-attention mechanism to capture global features in the peptide sequence; the computer device 110 can input high-dimensional biological features and global features into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results. Among them, the computer device 110 can be, but is not limited to, various personal computers, laptops, smart phones, robots, unmanned aerial vehicles and other devices.

[0053] In one embodiment, Figure 2 As shown, a method for predicting anticancer peptides based on a protein language model is provided, comprising the following steps:

[0054] Step 202 , collect and screen anticancer peptide samples and non-anticancer peptide samples from the database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples.

[0055] In order to train the protein language model, a data set can be constructed first. Specifically, experimentally verified anticancer peptides and non-anticancer peptide samples can be collected and screened from public databases, and after homology filtering and PSSM matrix extraction, a training set can be constructed for model training.

[0056] In one embodiment, a method for predicting anticancer peptides based on a protein language model is provided, which may also include a process for constructing a data set, and the specific process includes: collecting anticancer peptides from a database, and performing homology elimination on the anticancer peptides to obtain the eliminated anticancer peptides; performing PSSM matrix extraction on the eliminated anticancer peptides to obtain a final anticancer peptide sample; collecting non-anticancer peptides from a database, and performing homology elimination on the non-anticancer peptides to obtain the eliminated non-anticancer peptides; performing PSSM matrix extraction on the eliminated non-anticancer peptides to obtain a final non-anticancer peptide sample; selecting anticancer peptide sequences from the final anticancer peptide sample, and selecting non-anticancer peptide sequences from the final non-anticancer peptide sample to form a data set.

[0057] In this embodiment, a total of 1350 experimentally verified anticancer peptides can be collected from the public databases CancerPPD, APD3 and SATPd to eliminate anticancer peptide samples with a homology of more than 90%, aiming to eliminate the excessive recognition accuracy caused by homology bias. Next, the Seqkit tool is used to extract sequences with lengths from 5 to 50, and only sequences that can access the PSSM matrix through the PSI-BLAST tool are retained. At the same time, the non-anticancer peptides can be subjected to homology bias elimination and PSSM matrix extraction in the same manner, and finally 1839 non-anticancer peptide samples are obtained.

[0058] In one embodiment, a provided anticancer peptide prediction method based on a protein language model may also include a process of dividing a data set, the specific process including: taking the final anticancer peptide sample as a positive sample, and taking the final non-anticancer peptide sample as a negative sample; randomly selecting anticancer peptide samples from the positive samples as anticancer peptide sequences, and randomly selecting non-anticancer peptide samples from the negative samples as non-anticancer peptide sequences as data sets; and taking the remaining anticancer peptide sequences in the positive samples and the remaining non-anticancer peptide sequences in the negative samples as test sets.

[0059] For example, in this embodiment, 487 sequences can be randomly selected from the anticancer peptide positive samples, and 1479 sequences can be randomly selected from the anticancer peptide negative samples to construct a training data set. In addition, two test sets Test1 and Test2 are constructed to verify the generalization ability of the protein language model to predict anticancer peptide ACP, and the two constructed test sets are also subjected to homology filtering and PSSM matrix extraction. Among them, the number of positive samples of Test1 is 135, and the number of negative samples is 360, which is the remaining part of the constructed anticancer peptide data set, and Test2 can be composed of 99 positive samples and 157 negative samples.

[0060] Step 204 , input the data set into a protein language model, perform feature processing on the sample sequences in the data set through the protein language model, and extract high-dimensional biological features in the sample sequences.

[0061] In one embodiment, a provided anticancer peptide prediction method based on a protein language model may also include a process of extracting high-dimensional biological features, the specific process including: inputting a data set into a protein language model, performing feature processing on the anticancer peptide sequences in the data set through the protein language model, capturing implicit patterns, long-term dependencies, and contextual information in the anticancer peptide sequences; based on the implicit patterns, long-term dependencies, and contextual information, generating high-dimensional biological features in the anticancer peptide sequences.

[0062] Among them, the protein language model ESM2 performs feature processing on anticancer peptides through a deep learning pre-training model, which can deeply understand the complex structure and functional information of protein sequences; and the pre-training process based on large-scale data enables the model to capture implicit patterns and long-term dependencies in anticancer peptides and generate high-dimensional sequence representations.

[0063] Step 206, convolution operation is performed on the high-dimensional biological features using a convolutional neural network through a protein language model, and the weights of the high-dimensional biological features are dynamically adjusted through a self-attention mechanism to capture the global features in the peptide sequence.

[0064] In one embodiment, a method for predicting anticancer peptides based on a protein language model is provided that may further include a convolution operation process, the specific process including: performing a convolution operation on context information using a convolutional neural network through a protein language model to obtain convolved information; adjusting the weights of high-dimensional biological features through a self-attention mechanism to identify long-term dependencies; integrating global context information based on the convolved information, the weights of high-dimensional biological features, and long-term dependencies to obtain global features.

[0065] Computer equipment can extract high-dimensional biological features through ESM2. The convolutional neural network CNN no longer processes the original sequence, but instead performs convolution operations on the contextual information extracted by ESM2, making the features extracted by CNN more representative and helping the model capture important local patterns more efficiently.

[0066] In one embodiment, a provided anti-cancer peptide prediction method based on a protein language model may also include a process of adjusting weights using a self-attention mechanism, the specific process including: performing a linear transformation on high-dimensional biological features through a self-attention mechanism to obtain a transformation matrix; obtaining a self-attention mechanism weight matrix and obtaining a key vector dimension of each attention head; adjusting the weights of high-dimensional biological features based on the transformation matrix, the self-attention mechanism weight matrix, and the key vector dimension.

[0067] Computer equipment can dynamically adjust the weights of each feature through the self-attention mechanism to identify key long-distance dependencies in the sequence. The combination of this local feature extraction capability and the self-attention mechanism allows the model to more accurately integrate global context information and capture important global features in anticancer peptides, thereby improving the classification and prediction performance of anticancer peptides.

[0068] Among them, the self-attention mechanism of ESM2 can be described as:

[0069]

[0070] Among them, Q, K and V represent the query, key and value matrices obtained by performing different linear transformations on the input sequence, respectively. and represents the weight matrix, d k Represents the dimension of the key vector for each attention head in the key matrix.

[0071] The feedforward neural network sublayer of ESM2 performs a nonlinear transformation on the context representation of each position. This sublayer usually includes an activation function and two linear transformations, and its calculation process is as follows:

[0072] output(x)=max(0,xW1+b1)W2+b2

[0073] Among them, x represents the input, W1 represents the convolution kernel of the first layer, W2 represents the convolution kernel of the second layer, and b1 and b2 represent biases.

[0074] In one embodiment, the computer device can use a convolutional neural network to perform a nonlinear transformation on each context information to obtain the information after the nonlinear transformation; use a one-dimensional convolution to perform a convolution operation on the information after the nonlinear transformation, and add a batch normalization process after each convolution layer, and obtain the convolved information through the pooling layer output.

[0075] Specifically, three one-dimensional convolutional layers can be applied to process the extracted features through convolution operations. Batch normalization is also added after each convolutional layer to normalize the features of each small batch of data to speed up the network training process and improve the robustness of the model. Let the input be x and the convolution kernel of the i-th layer be W i , with a bias of b i , then the convolution operation is: i =x*W i +b i .

[0076] In order to obtain more accurate results, it is very important to use pooling layers to reduce the dimensionality of the final convolutional layer output. Pooling layers usually contain average pooling layers and maximum pooling layers. After three convolutional layers, average pooling can be applied to the obtained features to obtain a lower dimensional representation, which can be expressed as: Where Y = {y1, y2, ..., y m} is the sample in the pooling area, m is the number of samples in the pooling area. The final output is the average of all values ​​in the pooling area.

[0077] Step 208 , input the high-dimensional biological features and global features into the fully connected layer in the protein language model to perform anticancer peptide prediction to obtain prediction results.

[0078] After being processed by the self-attention mechanism, the generated feature vector will be passed to the fully connected classification module for subsequent processing. In this classification module, the feature vector will be transformed through two fully connected layers, and finally the result will be mapped to the range of 0 to 1 through the sigmoid function to perform the final binary classification task. For samples with an output probability lower than 0.5, they are classified as non-anticancer peptides (non-ACP), while for samples with an output probability equal to or greater than 0.5, they are classified as anticancer peptides (ACP). The specific formula can be expressed as: The definition of the fully connected layer is: i =q i-1 W i +b i ; Among them, q i-1 is the input data of the i-th layer, q i is the output data of this layer, q i is the weight matrix of this layer, b i is the bias term of this layer. The function of this function is to map the input real value so that it falls within the interval of (0,1). It is usually used in the output layer of binary classification problems.

[0079] In one embodiment, a method for predicting anticancer peptides based on a protein language model is provided, which may also include a model evaluation process, wherein the specific process includes: determining a model evaluation index, and performing a model evaluation on the protein language model according to the model evaluation index to obtain an evaluation result; and optimizing the protein language model based on the evaluation result.

[0080] In this embodiment, the following four commonly used indicators can be used for evaluation: sensitivity (SN), specificity (SP), accuracy (ACC), and Matthews correlation coefficient (MCC). The formulas of these indicators are described as follows:

[0081]

[0082] Among them, TP, FN, TN, and FP represent the number of true positive, false negative, true negative, and false positive samples, respectively. SN represents the proportion of true ACP, and SP represents the proportion of non-ACP samples correctly identified by the model. MCC and ACC are used to measure the overall prediction performance of the model. These indicators are used to quantitatively evaluate the overall effectiveness of the prediction model. For each of the aforementioned indicators, the higher the value, the better the performance of the model.

[0083] In one embodiment, a method for predicting anticancer peptides based on a protein language model can be applied to Figure 3 The model framework of ACP-ESM2 is shown.

[0084] like Figure 3 As shown in the figure, the model framework of ACP-ESM2 can include ESM2 protein language model, convolutional neural network-self-attention mechanism, fully connected layer, and Web server. Among them, ESM2 protein language model can perform feature extraction to extract high-dimensional biological features in sample sequences; convolutional neural network-self-attention mechanism can adjust weights after convolution operation on high-dimensional biological features to capture global features; fully connected layer can perform classification prediction based on high-dimensional biological features and global features to obtain anti-cancer peptide prediction results; Web server can visualize the prediction results for easy viewing.

[0085] In one embodiment, an experiment was conducted on a model framework of a method for predicting anticancer peptides based on a protein language model, wherein 487 sequences can be randomly selected from anticancer peptide positive samples, and 1479 sequences can be randomly selected from anticancer peptide negative samples to construct a training data set. In addition, two test sets Test1 and Test2 are constructed to verify the generalization ability of the protein language model for predicting anticancer peptide ACP, and the two constructed test sets are also subjected to homology filtering and PSSM matrix extraction. Among them, the number of positive samples of Test1 is 135, and the number of negative samples is 360, which is the remainder of the constructed anticancer peptide data set, and Test2 can be composed of 99 positive samples and 157 negative samples.

[0086] The experimental results of the model framework ACP-ESM2 of the anticancer peptide prediction method based on the protein language model provided in this embodiment are compared with the experimental results of the traditional predictor on the benchmark data set. From the results of test set one and test set two, ACP-ESM2 performs well in various indicators, especially in test set two, showing stronger generalization ability. The ACC in test set two reached 0.976 and the MCC reached 0.950, indicating that the model not only has excellent classification ability, but also performs particularly well in the balance of processing positive and negative samples. In comparison, the performance of ACP-ML is also relatively stable, but it is slightly insufficient in the MCC and SP indicators. Although MACPpred and ACPred have been improved to a certain extent on both test sets, their performance is still significantly lower than OurWork. The performance of ACP-MHCNN and ACPred-BMF is the most unsatisfactory, especially on SP and MCC, reflecting their significant deficiencies in negative sample prediction. Overall, OurWork not only achieved the best prediction performance on both test sets, but also, especially in test set two, its excellent generalization ability further highlights the advantages of the model. The comparison results of test set 1 and the traditional predictor and the comparison results of test set 2 and the traditional predictor are shown in the following tables:

[0087] Table 1: Comparison with existing methods on test set 1

[0088]

[0089] Table 2: Comparison with existing methods on test set 2

[0090]

[0091] In this embodiment, an ablation experiment is also performed on the model framework of the provided anticancer peptide prediction method based on a protein language model. The ablation experiment results are shown in the following table:

[0092] Table 3: Model ablation experiment on the training set

[0093]

[0094] From the ablation experiment results, it can be seen that the model performance gradually improves with the increase of components, especially the combination of CNN and self-attention mechanism shows a significant synergistic effect. When ESM2 is used alone, the ACC is 0.916 and the MCC is 0.756. Although it has certain capabilities in basic feature extraction, it performs weakly in SN (0.760), indicating that the recognition ability of positive samples is limited. After adding CNN, the ACC and MCC of the ESM2+CNN model are increased to 0.938 and 0.820 respectively, and the SN increases to 0.796, which proves the effectiveness of CNN in local feature extraction. After further adding the self-attention mechanism, ESM2+CNN+Attention has improved in all indicators, with ACC reaching 0.942 and MCC increasing to 0.832, especially in SP (0.982) and SN (0.810). It shows that the self-attention mechanism effectively enhances the model's ability to capture important information in the sequence, thereby improving the overall performance. This combination enables the models to complement each other in feature extraction and information integration, further improving the accuracy of anticancer peptide identification.

[0095] Next, you can visualize the data set before model training and the data set after model training. Figure 4 As shown in Figure 2, feature visualization shows that the data after model training is clustered into two distinct clusters. Figure 5 As shown, the two types of untrained datasets are scattered together.

[0096] It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0097] In one embodiment, Figure 6 As shown, a system for predicting anticancer peptides based on a protein language model is provided, comprising: a data set construction module 610, a high-dimensional biological feature extraction module 620, a global feature extraction module 630 and an anticancer peptide prediction module 640, wherein:

[0098] A data set construction module 610 is used to collect and screen anticancer peptide samples and non-anticancer peptide samples from the database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples;

[0099] A high-dimensional biological feature extraction module 620 is used to input the data set into the protein language model, perform feature processing on the sample sequence in the data set through the protein language model, and extract high-dimensional biological features in the sample sequence;

[0100] A global feature extraction module 630 is used to perform convolution operations on high-dimensional biological features using a convolutional neural network through a protein language model, and dynamically adjust the weights of the high-dimensional biological features through a self-attention mechanism to capture global features in the peptide sequence;

[0101] The anticancer peptide prediction module 640 is used to input high-dimensional biological features and global features into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results.

[0102] In one embodiment, the data set construction module 610 is also used to collect anticancer peptides from the database, and perform homology elimination on the anticancer peptides to obtain the eliminated anticancer peptides; perform PSSM matrix extraction on the eliminated anticancer peptides to obtain the final anticancer peptide sample; collect non-anticancer peptides from the database, and perform homology elimination on the non-anticancer peptides to obtain the eliminated non-anticancer peptides; perform PSSM matrix extraction on the eliminated non-anticancer peptides to obtain the final non-anticancer peptide sample; select anticancer peptide sequences from the final anticancer peptide sample, and select non-anticancer peptide sequences from the final non-anticancer peptide sample to form a data set.

[0103] In one embodiment, the data set construction module 610 is also used to take the final anti-cancer peptide sample as a positive sample and the final non-anti-cancer peptide sample as a negative sample; randomly select anti-cancer peptide samples from the positive samples as anti-cancer peptide sequences, and randomly select non-anti-cancer peptide samples from the negative samples as non-anti-cancer peptide sequences as a data set; and take the remaining anti-cancer peptide sequences in the positive samples and the remaining non-anti-cancer peptide sequences in the negative samples as a test set.

[0104] In one embodiment, the high-dimensional biological feature extraction module 620 is also used to input the data set into a protein language model, perform feature processing on the anticancer peptide sequences in the data set through the protein language model, capture implicit patterns, long-term dependencies, and contextual information in the anticancer peptide sequences; based on the implicit patterns, long-term dependencies, and contextual information, generate high-dimensional biological features in the anticancer peptide sequences.

[0105] In one embodiment, the global feature extraction module 630 is also used to perform convolution operations on context information using a convolutional neural network through a protein language model to obtain convolved information; adjust the weights of high-dimensional biological features through a self-attention mechanism to identify long-term dependencies; integrate global context information based on the convolved information, the weights of high-dimensional biological features, and long-term dependencies to obtain global features.

[0106] In one embodiment, the global feature extraction module 630 is also used to use a convolutional neural network to perform a nonlinear transformation on each context information to obtain the information after the nonlinear transformation; use a one-dimensional convolution to perform a convolution operation on the information after the nonlinear transformation, and add batch normalization processing after each convolution layer to obtain the convolved information through the pooling layer output.

[0107] In one embodiment, the global feature extraction module 630 is also used to perform linear transformation on high-dimensional biological features through a self-attention mechanism to obtain a transformation matrix; obtain a self-attention mechanism weight matrix and obtain a key vector dimension of each attention head; and adjust the weight of the high-dimensional biological features based on the transformation matrix, the self-attention mechanism weight matrix, and the key vector dimension.

[0108] In one embodiment, a protein language model-based anticancer peptide prediction system is provided, which also includes a model evaluation module for determining a model evaluation index, and performing a model evaluation on the protein language model according to the model evaluation index to obtain an evaluation result; and optimizing the protein language model based on the evaluation result.

[0109] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting anticancer peptides based on a protein language model is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0110] Those skilled in the art will understand that Figure 7The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0111] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the anti-cancer peptide prediction method based on the protein language model are implemented.

[0112] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for predicting anticancer peptides based on a protein language model are implemented.

[0113] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0114] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for predicting anticancer peptides based on a protein language model, characterized in that: The method comprises: Collect and screen anticancer peptide samples and non-anticancer peptide samples from a database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples; Inputting the data set into a protein language model, performing feature processing on sample sequences in the data set through the protein language model, and extracting high-dimensional biological features in the sample sequences; The protein language model uses a convolutional neural network to perform a convolution operation on the high-dimensional biological features, and dynamically adjusts the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the peptide sequence; The high-dimensional biological features and global features are input into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results.

2. The method for predicting anticancer peptides based on a protein language model according to claim 1, characterized in that: Anticancer peptide samples and non-anticancer peptide samples are collected and screened from the database, and a data set is constructed based on the anticancer peptide samples and non-anticancer peptide samples, including: Collecting anticancer peptides from a database, and eliminating the anticancer peptides by homology to obtain the eliminated anticancer peptides; Performing PSSM matrix extraction on the eliminated anticancer peptides to obtain a final anticancer peptide sample; Collecting non-anti-cancer peptides from a database, and eliminating the non-anti-cancer peptides by homology to obtain eliminated non-anti-cancer peptides; Performing PSSM matrix extraction on the eliminated non-anticancer peptides to obtain a final non-anticancer peptide sample; Anti-cancer peptide sequences are selected from the final anti-cancer peptide sample, and non-anti-cancer peptide sequences are selected from the final non-anti-cancer peptide sample to form a data set.

3. The method for predicting anticancer peptides based on a protein language model according to claim 2, characterized in that: The method further comprises: The final anti-cancer peptide sample is taken as a positive sample, and the final non-anti-cancer peptide sample is taken as a negative sample; Randomly selecting anticancer peptide samples from the positive samples as anticancer peptide sequences, and randomly selecting non-anticancer peptide samples from the negative samples as non-anticancer peptide sequences as a data set; The remaining anticancer peptide sequences in the positive samples and the remaining non-anticancer peptide sequences in the negative samples are used as test sets.

4. The method for predicting anticancer peptides based on a protein language model according to claim 1, characterized in that: The data set is input into a protein language model, and the sample sequences in the data set are subjected to feature processing by the protein language model to extract high-dimensional biological features in the sample sequences, including: Inputting the data set into a protein language model, and performing feature processing on the anticancer peptide sequences in the data set through the protein language model to capture implicit patterns, long-term dependencies, and contextual information in the anticancer peptide sequences; Based on the implicit patterns, long-term dependencies, and contextual information, high-dimensional biological features in the anticancer peptide sequence are generated.

5. The method for predicting anticancer peptides based on a protein language model according to claim 4, characterized in that: The protein language model uses a convolutional neural network to perform a convolution operation on the high-dimensional biological features, and dynamically adjusts the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the peptide sequence, including: Using the protein language model, a convolutional neural network is used to perform a convolution operation on the context information to obtain convolved information; The weight of the high-dimensional biological features is adjusted through a self-attention mechanism to identify the long-term dependency relationship; Based on the convolved information, the weights of the high-dimensional biological features, and the long-term dependencies, the global context information is integrated to obtain the global features.

6. The method for predicting anticancer peptides based on a protein language model according to claim 5, characterized in that: The protein language model is used to perform a convolution operation on the context information using a convolutional neural network to obtain convolutional information, including: Using a convolutional neural network to perform a nonlinear transformation on each of the context information to obtain nonlinearly transformed information; A one-dimensional convolution is used to perform a convolution operation on the information after the nonlinear transformation, and a batch normalization process is added after each convolution layer, and the convolved information is obtained through the output of the pooling layer.

7. The method for predicting anticancer peptides based on a protein language model according to claim 5, characterized in that: The weights of the high-dimensional biological features are adjusted through a self-attention mechanism, including: Performing a linear transformation on the high-dimensional biological features through a self-attention mechanism to obtain a transformation matrix; Get the self-attention mechanism weight matrix and get the key vector dimension of each attention head; The weight of the high-dimensional biological feature is adjusted based on the transformation matrix, the self-attention mechanism weight matrix, and the key vector dimension.

8. The method for predicting anticancer peptides based on a protein language model according to claim 1, characterized in that: The method further comprises: Determining a model evaluation index, and performing a model evaluation on the protein language model according to the model evaluation index to obtain an evaluation result; The protein language model is optimized based on the evaluation results.

9. An anticancer peptide prediction system based on a protein language model, characterized in that: The system comprises: A data set construction module, used to collect and screen anticancer peptide samples and non-anticancer peptide samples from a database, and construct a data set based on the anticancer peptide samples and non-anticancer peptide samples; A high-dimensional biological feature extraction module, used to input the data set into a protein language model, perform feature processing on the sample sequence in the data set through the protein language model, and extract high-dimensional biological features in the sample sequence; A global feature extraction module, used to perform a convolution operation on the high-dimensional biological features using a convolutional neural network through the protein language model, and dynamically adjust the weights of the high-dimensional biological features through a self-attention mechanism to capture the global features in the peptide sequence; The anticancer peptide prediction module is used to input the high-dimensional biological features and global features into the fully connected layer in the protein language model to predict anticancer peptides and obtain prediction results.

10. The anticancer peptide prediction system based on protein language model according to claim 9, characterized in that: The data set construction module is also used to: collect anticancer peptides from the database, and perform homology elimination on the anticancer peptides to obtain eliminated anticancer peptides; perform PSSM matrix extraction on the eliminated anticancer peptides to obtain final anticancer peptide samples; collect non-anticancer peptides from the database, and perform homology elimination on the non-anticancer peptides to obtain eliminated non-anticancer peptides; perform PSSM matrix extraction on the eliminated non-anticancer peptides to obtain final non-anticancer peptide samples; select anticancer peptide sequences from the final anticancer peptide samples, and select non-anticancer peptide sequences from the final non-anticancer peptide samples to form a data set.

Images