Neuropeptide identification method and system based on comparative learning and cross attention mechanism

Abstract

The invention relates to a neuropeptide identification method and system based on comparative learning and a cross attention mechanism. The method comprises the steps that a data set is acquired and preprocessed to obtain a training data set; performing feature extraction on the training data set by using a feature coding strategy, a BLOSUM62 matrix and a protein language model ESM-2; performing feature representation through a bidirectional long-short-term memory network and a multi-scale convolutional neural network; obtaining two target vectors based on a cross attention mechanism, a multi-head attention mechanism and a feedforward neural network; and determining a contrast loss function, performing contrast learning on the two target vectors according to the contrast loss function, performing dimension reduction processing, constructing a neuropeptide identification model, and completing neuropeptide identification. Local details and global context information can be captured through various strategies and network models; by constructing the neuropeptide identification model, the model can more effectively capture the complex relationship between sequences and between samples, and the identification precision of neuropeptide is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of neuropeptide identification, and in particular to a neuropeptide identification method and system based on contrastive learning and cross-attention mechanisms. Background Art

[0002] Neuropeptides (NPs) are a class of signaling molecules that play a key role in regulating physiology and behavior within the nervous system. They are essential for neurotransmission, endocrine regulation, immune responses, and control of mood and appetite, among other aspects. The identification of neuropeptides provides an important scientific basis for early diagnosis, targeted therapy, and personalized medicine for related diseases. Currently, the techniques for identifying neuropeptides (NPs) fall into two main categories: experimental and computational. Experimental methods primarily utilize biochemical and molecular biology techniques to identify neuropeptides. Common experimental methods include bioassays, receptor binding assays, and liquid chromatography-tandem mass spectrometry (LC-MS / MS), which identify and quantify neuropeptides through bioactivity assays, neuropeptide-receptor binding characteristics, and mass spectrometry analysis, respectively. With the advancement of bioinformatics, data-driven computational methods have gradually become an important tool for neuropeptide identification. These methods mainly rely on machine learning and deep learning technologies. Common computational methods include: sequence homology-based methods, machine learning-based models, and deep learning-based models. Among them, the sequence homology-based method identifies new neuropeptides by comparing known neuropeptide sequence databases; the machine learning-based model uses traditional machine learning algorithms such as support vector machines (SVM) and random forests, combined with manually designed features (such as amino acid composition, physical and chemical properties, etc.) to predict neuropeptides; the deep learning-based model uses deep learning models such as convolutional neural networks (CNN) and recurrent neural networks (RNN) to automatically extract features from amino acid sequences and perform classification.

[0003] However, traditional neuropeptide identification techniques often require significant time and resources, have high demands on sample quality and purity, require complex sample preparation, and incur high costs for experimental equipment and reagents. Traditional machine learning methods rely on handcrafted features and are limited to classical machine learning and the simple stacking of multiple features, presenting certain limitations. Consequently, traditional neuropeptide identification techniques often suffer from high costs and low accuracy. Summary of the Invention

[0004] Based on this, in order to solve the above technical problems, a neuropeptide identification method and system based on contrastive learning and cross-attention mechanism are provided, which can improve the identification accuracy and efficiency of neuropeptides and reduce costs.

[0005] A neuropeptide identification method based on contrastive learning and cross-attention mechanism, comprising:

[0006] Acquiring a data set containing neuropeptides and non-neuropeptides, and preprocessing the data set to obtain a training data set;

[0007] Using a feature encoding strategy to encode the neuropeptide sequence corresponding to the training data set into a first feature vector; using a BLOSUM62 matrix to encode the neuropeptide sequence corresponding to the training data set into a second feature vector; using a protein language model ESM-2 to extract features from the training data set to obtain a large model feature;

[0008] Inputting the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; inputting the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and merging the first output feature and the second output feature into a manually encoded output feature;

[0009] The hand-coded output features and the large model output features are processed based on the cross-attention mechanism and input into the multi-head attention mechanism and the feedforward neural network to obtain two target vectors;

[0010] A contrast loss function is determined, and the two target vectors are subjected to contrast learning according to the contrast loss function, and dimensionality reduction processing is performed to construct a neuropeptide identification model to complete neuropeptide identification.

[0011] In one embodiment, preprocessing the data set to obtain a training data set includes:

[0012] Redundant sequences of neuropeptide sequences in the data set are removed, and the data set after removing redundancy is divided into a training data set and a validation data set.

[0013] In one embodiment, encoding the neuropeptide sequence corresponding to the training dataset into a first feature vector using a feature encoding strategy includes:

[0014] The neuropeptide sequence corresponding to the training data set is feature extracted using coding strategies of ordinal coding, amino acid composition, dipeptide composition, position-specific amino acid composition, adaptive jumping dipeptide composition, and amino acid entropy to obtain a first feature vector.

[0015] In one embodiment, the first feature vector and the large model feature are input into a bidirectional long short-term memory network to obtain the first output feature and the large model output feature, including:

[0016] Normalizing the first feature vector and the large model feature, and inputting the processed features into a bidirectional long short-term memory network respectively;

[0017] The long sequence is processed by the bidirectional long short-term memory network, and the captured context information is transferred for feature extraction to obtain the first output feature and the large model output feature.

[0018] In one embodiment, the second feature vector is input into a multi-scale convolutional neural network to obtain a second output feature, including:

[0019] Inputting the second feature vector into a multi-scale convolutional neural network, and capturing information through convolution kernels of different sizes in the multi-scale convolutional neural network;

[0020] extracting features of different scales from the second feature vector based on the captured information;

[0021] The features of different scales are fused to obtain the second output feature.

[0022] In one embodiment, the hand-coded output features and the large model output features are processed based on a cross-attention mechanism and input into a multi-head attention mechanism and a feedforward neural network to obtain two target vectors, including:

[0023] Input the hand-coded output features and the large model output features into the cross-attention mechanism to obtain the interactive features;

[0024] Inputting the interaction features into a multi-head attention mechanism, dividing the interaction features into subspaces through the multi-head attention mechanism, and calculating the attention weights in each subspace to obtain attention weighted features;

[0025] The attention weighted features are input into a feedforward neural network for processing to obtain two target vectors.

[0026] In one embodiment, determining a contrastive loss function, and performing contrastive learning on the two target vectors according to the contrastive loss function, includes:

[0027] Determining a contrast loss function and calculating a similarity score between the two target vectors;

[0028] The gradient is calculated based on the contrast loss function and the similarity score, and the parameters are updated based on the gradient to complete contrast learning.

[0029] In one embodiment, the method further comprises:

[0030] Obtaining an evaluation index, and using the evaluation index to evaluate the neuropeptide identification model to obtain an evaluation result;

[0031] The evaluation indicators include accuracy, sensitivity, specificity, Matthews correlation coefficient, and area under the ROC curve.

[0032] In one embodiment, the method further comprises:

[0033] Collecting a data set containing neuropeptides and non-neuropeptides to be identified and inputting the data set into the neuropeptide identification model;

[0034] The neuropeptide identification model is used to output the neuropeptide identification results and display them on a page.

[0035] A neuropeptide identification system based on contrastive learning and cross-attention mechanism, comprising:

[0036] A data set acquisition module is used to acquire a data set containing neuropeptides and non-neuropeptides, and preprocess the data set to obtain a training data set;

[0037] A feature encoding module is configured to encode the neuropeptide sequence corresponding to the training data set into a first feature vector using a feature encoding strategy; encode the neuropeptide sequence corresponding to the training data set into a second feature vector using a BLOSUM62 matrix; and extract features from the training data set using a protein language model ESM-2 to obtain large model features;

[0038] a feature extraction module, configured to input the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; input the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and merge the first output feature and the second output feature into a manually encoded output feature;

[0039] A feature processing module is used to process the manually encoded output features and the large model output features based on a cross-attention mechanism, and input them into a multi-head attention mechanism and a feedforward neural network to obtain two target vectors;

[0040] The neuropeptide identification module is used to determine a contrast loss function, perform contrast learning on the two target vectors according to the contrast loss function, perform dimensionality reduction processing, construct a neuropeptide identification model, and complete neuropeptide identification.

[0041] The above-mentioned neuropeptide identification method and system based on contrastive learning and cross-attention mechanism, by using multiple feature encoding strategies in combination with the protein language model ESM-2 for feature extraction, and then inputting it into the long short-term memory network and the multi-scale convolutional neural network, can capture local details and global context information and improve data utilization; by constructing a neuropeptide identification model through the cross-attention mechanism and contrastive learning, the model can more effectively capture the complex relationships between sequences and samples, improve the accuracy and efficiency of neuropeptide identification, and reduce costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 FIG2 is an application environment diagram of a neuropeptide identification method based on contrastive learning and cross-attention mechanism in one embodiment;

[0043] Figure 2 1 is a flow chart of a neuropeptide identification method based on contrastive learning and cross-attention mechanism in one embodiment;

[0044] Figure 3 FIG1 is a schematic diagram of the structure of a deep learning model named NeuroCL as a neuropeptide identification model in one embodiment;

[0045] Figure 4 A bar chart comparing the indicators of the neuropeptide identification model and existing models in one embodiment;

[0046] Figure 5 FIG1 is a structural block diagram of a neuropeptide identification system based on contrastive learning and cross-attention mechanism in one embodiment;

[0047] Figure 6 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] It will be understood that the terms "first," "second," etc., used herein may be used herein to describe feature vectors and output features, but these feature vectors and output features are not limited by these terms. These terms are only used to distinguish a first feature vector or output feature from another feature vector or output feature. For example, a first feature vector may be referred to as a second feature vector, and similarly, a second feature vector may be referred to as a first feature vector, without departing from the scope of this application. Both the first feature vector and the second feature vector are feature vectors, but they are not the same feature vector.

[0050] The neuropeptide identification method based on contrastive learning and cross-attention mechanism provided in the embodiment of the present application 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 obtain a data set containing neuropeptides and non-neuropeptides, and preprocess the data set to obtain a training data set; the computer device 110 can use a feature encoding strategy to encode the neuropeptide sequence corresponding to the training data set into a first feature vector; use the BLOSUM62 matrix to encode the neuropeptide sequence corresponding to the training data set into a second feature vector; use the protein language model ESM-2 to extract features of the training data set to obtain a large model feature; the computer device 110 can input the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; input the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and merge the first output feature and the second output feature into a hand-coded output feature; the computer device 110 can process the hand-coded output feature and the large model output feature based on a cross-attention mechanism, and input them into a multi-head attention mechanism and a feedforward neural network to obtain two target vectors; the computer device 110 can determine a contrast loss function, and the two target vectors are contrastively learned according to the contrast loss function, and dimensionality reduction processing is performed to construct a neuropeptide identification model to complete neuropeptide identification. The computer device 110 may be, but is not limited to, various personal computers, laptops, smart phones, robots, tablet computers, and other devices.

[0051] In one embodiment, Figure 2 As shown, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, comprising the following steps:

[0052] Step 202: Acquire a data set containing neuropeptides and non-neuropeptides, and preprocess the data set to obtain a training data set.

[0053] The computer device can obtain a dataset containing neuropeptides and non-neuropeptides from a database. Specifically, the computer device can use datasets from the SwissProt and NeuroPep databases, which contain 2,425 experimentally verified neuropeptide (NP) sequences and the same number of non-neuropeptide (non-NP) sequences. The neuropeptide sequences are obtained from the NeuroPep database, a specialized neuropeptide resource, while the non-neuropeptide sequences are obtained from the SwissProt database.

[0054] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include a data set preprocessing process, specifically including: removing redundant sequences of neuropeptide sequences in the data set, and dividing the data set after removing redundancy into a training data set and a validation data set.

[0055] In order to avoid inaccurate experimental results due to excessive similarity between samples, the computer equipment can use the CD-HIT tool to remove redundant sequences to ensure that the sequence similarity in the data set does not exceed 90%. The data set is divided into a training data set and a test data set. The training data set accounts for 80% of the total data set (including 3,880 samples) and the test data set accounts for 20% (970 samples). The test data set is used to evaluate and compare the performance of various models.

[0056] In this example, 10% of the training set was used as a validation set for hyperparameter tuning, model selection, and to prevent overfitting. The distribution of sequence lengths in the training set was consistent with that in the test set, encompassing sequences ranging from 0 to 100. This consistency made the training set more effective for training a neuropeptide prediction model capable of processing sequences in the 0-100 length range. Furthermore, the test set was comprehensive in predicting whether peptides between 0 and 100 were neuropeptides.

[0057] In step 204, the neuropeptide sequence corresponding to the training dataset is encoded into a first feature vector using a feature encoding strategy; the neuropeptide sequence corresponding to the training dataset is encoded into a second feature vector using a BLOSUM62 matrix; and the protein language model ESM-2 is used to extract features from the training dataset to obtain large model features.

[0058] The computer device can use a variety of feature encoding strategies and protein language model ESM-2 for feature extraction.

[0059] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may include a process of extracting features using multiple feature encoding strategies. The specific process includes: using ordinal coding, amino acid composition, dipeptide composition, position-specific amino acid composition, adaptive jumping dipeptide composition, and amino acid entropy encoding strategies to extract features of the neuropeptide sequence corresponding to the training data set to obtain a first feature vector.

[0060] That is, in this embodiment, multiple feature encoding strategies may include ordinal encoding, amino acid composition (AAC), dipeptide composition (DPC), position-specific amino acid composition (PSAAC), adaptive jumping dipeptide composition (ASDC), and amino acid entropy (AAE), which can generate the first feature vectors of L*100, L*21, L*231, L*40, L*400, and L*60, respectively.

[0061] The multiple feature encoding strategies may also include a BLOSUM62 matrix, and the computer device may use the BLOSUM62 matrix to encode the neuropeptide sequence corresponding to the training dataset into a second feature vector of L*100*23.

[0062] At the same time, the computer equipment can use the protein language model ESM-2 to extract features from the training data set and obtain a large model feature of L*1280.

[0063] In step 206, the first eigenvector and the large model feature are input into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; the second eigenvector is input into a multi-scale convolutional neural network to obtain a second output feature; and the first output feature and the second output feature are combined into a manually encoded output feature.

[0064] The computer device can construct a neuropeptide identification model, which can include a multi-scale convolutional neural network, a bidirectional long short-term memory network, a cross attention mechanism, a multi-head attention mechanism, a feedforward neural network, and a contrast loss function.

[0065] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include a process of extracting features through a bidirectional long short-term memory network. The specific process includes: normalizing the first feature vector and the large model feature, and inputting the processed features into the bidirectional long short-term memory network respectively; processing the long sequence through the bidirectional long short-term memory network, and passing the captured context information for feature extraction to obtain the first output feature and the large model output feature.

[0066] The computer device can input the first feature vector extracted by the pre-trained large model (ESM-2) into the bidirectional long short-term memory network, and the position composition related manual feature encoding (OE, AAC, DPC, PSAAC, ASDC, AAE), that is, the large model feature, into the bidirectional long short-term memory network, thereby obtaining the first output feature and the large model output feature.

[0067] Among them, the BiLSTM captures the contextual information in the sequence through forward and backward transmission, processes long sequences and maintains long-term information propagation. In the BiLSTM, the sequence is processed forward from time step t = 1 to t = T, and the hidden state h is calculated. t The process may include:

[0068] Forget gate calculation: f t =σ(W f ·[h t-1 ,x t +b f ]);

[0069] Input gate calculation: i t =σ(W i ·[h t-1 ,x t +b i ]);

[0070] Temporary cell state:

[0071] Memory Update:

[0072] Output gate calculation: o t =σ(W o ·[h t-1 ,x t ]+b o );

[0073] Hidden layer state: h t =o t tanh(C t ). Among them, W f 、W i 、W c 、W o are the parameters of the bidirectional long short-term memory network; h t-1 represents the hidden state of the previous moment, x t Indicates the input at the current moment; b f 、b i 、b c 、b o Both represent constants; the output results in the forward and reverse directions are recorded as and Finally, the forward and backward hidden states are concatenated.

[0074] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanisms may further include a feature extraction process using a multi-scale convolutional neural network. The specific process includes: inputting the second feature vector into the multi-scale convolutional neural network, capturing information using convolution kernels of different sizes in the multi-scale convolutional neural network; extracting features of different scales from the second feature vector based on the captured information; fusing the features of different scales to obtain a second output feature; and merging the first and second output features into a manually encoded output feature.

[0075] The computer device can input the manual feature encoding of physicochemical properties (BLOSUM62) into a multi-scale convolutional neural network to obtain the second output feature. Among them, the multi-scale convolutional neural network (Multi-CNN) uses convolution kernels of different sizes to process the encoded amino acid sequence to capture local details and global context information. In this embodiment, the calculation formula of the multi-scale convolutional neural network can be expressed as follows: For input X, the convolution layer calculates:

[0076] in, represents the weight matrix of the convolution kernel of the kth layer with a size of P×Q, b k is the bias. The pooling layer is calculated as: pooling(X) i,k =max(X iM,k ,X iM+1,k ,…X iM+M-1,k ).

[0077] X is the input feature matrix, i is the row index in the feature matrix, k is the column index corresponding to different feature channels, and M is the window size of the pooling operation.

[0078] In step 208, the manually encoded output features and the large model output features are processed based on the cross-attention mechanism, and are input into the multi-head attention mechanism and the feedforward neural network to obtain two target vectors.

[0079] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include a feature processing process, and the specific process includes: inputting the manually encoded output features and the large model output features into the cross-attention mechanism to obtain interactive features; inputting the interactive features into the multi-head attention mechanism, dividing the interactive features into various subspaces through the multi-head attention mechanism, and calculating the attention weights in each subspace to obtain attention-weighted features; inputting the attention-weighted features into a feedforward neural network for processing to obtain two target vectors.

[0080] The computer device can encode the manual feature as Q and the large model feature as K and V. At the same time, the large model feature is encoded as Q and the manual feature is encoded as K and V, respectively input into the cross attention module, and then input into the self-attention mechanism and the feedforward neural network respectively to obtain two target vectors.

[0081] Among them, the cross-attention mechanism is used for information interaction between different input sequences, helping the model better capture the correlation between sequences. The calculation formula can be expressed as: V=A·X.

[0082] Q, K, and V represent the query matrix (Query), key matrix (Key), and value matrix (Value), respectively. Softmax is a soft function used to normalize the scores. Used for scaling to prevent excessively large values from being generated when calculating dot products. The multi-head attention mechanism is used to divide the input information into multiple subspaces, independently calculate the attention weights, and enhance the expressive power of the model. The calculation formula can be expressed as: Q, K, and V represent the query matrix (Query), key matrix (Key), and value matrix (Value) respectively; softmax is a soft function used to normalize the scores. Used for scaling to prevent overly large values from being generated when computing dot products.

[0083] In step 210 , a contrast loss function is determined, and the two target vectors are subjected to contrast learning according to the contrast loss function, and dimensionality reduction processing is performed to construct a neuropeptide identification model to complete neuropeptide identification.

[0084] The computer device can perform comparative learning on the two target vectors obtained based on the contrast loss, and finally enter the fully connected layer to reduce the dimension of the feature vector and enter the final prediction.

[0085] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include a contrastive learning process, the specific process including: determining a contrastive loss function and calculating a similarity score between two target vectors; calculating a gradient based on the contrastive loss function and the similarity score, and updating parameters based on the gradient to complete contrastive learning.

[0086] Among them, the contrast loss function is used to maximize the similarity of similar samples and minimize the similarity of dissimilar samples through contrast learning, thereby improving the model's discriminative ability. The calculation formula is as follows:

[0087] For two independent samples x1 and x2 in the data set, the feature space distance is calculated using the Euclidean distance:

[0088] D w(x1,x2)=||x1-x2||2;

[0089] For positive sample pairs (i.e. similar samples), the loss is defined as:

[0090] L positive =(1-y)×D w (x1-x2) 2 ;

[0091] For negative sample pairs (i.e., dissimilar samples), the loss is defined as:

[0092] L negative =y×max(0,margin-D w (x1,x2)) 2 ;

[0093] Set the binary classification label y. When y=0, it means that the sample units x1 and x2 have similar features; when y=1, it means that the sample units x1 and x2 have dissimilar features. Margin is the preset minimum distance between samples. The total contrast loss is defined as:

[0094] L contrastive =L positive +L negative .

[0095] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include a process of evaluating the neuropeptide identification model. The specific process includes: obtaining evaluation indicators, using the evaluation indicators to evaluate the neuropeptide identification model, and obtaining evaluation results; wherein the evaluation indicators include accuracy, sensitivity, specificity, Matthews correlation coefficient, and area under the ROC curve.

[0096] Specifically, the model performance can be evaluated using indicators such as accuracy (ACC), sensitivity (Sn), specificity (Sp), Matthews correlation coefficient (MCC) and area under the ROC curve (AUROC).

[0097]

[0098] Where TP, FP, TN, and FN represent the number of true positives, false positives, true negatives, and false negatives in the sample classification, respectively; ACC represents the proportion of all samples correctly classified; Sn represents the proportion of positive samples correctly identified; Sp represents the proportion of negative samples correctly identified; MCC measures the correlation between the true value and the predicted value and ranges from [-1 to 1]; AUC values range from 0 to 1, with higher AUC values indicating better predictive performance of the underlying model.

[0099] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, which may also include an online prediction process, the specific process including: collecting a data set containing neuropeptides and non-neuropeptides to be identified, and inputting it into a neuropeptide identification model; outputting the neuropeptide identification results through the neuropeptide identification model, and displaying them on a page.

[0100] In this embodiment, a user-friendly online prediction platform can be installed on a computer device. Users can predict neuropeptides by inputting a protein sequence or uploading a FASTA file. The prediction results are displayed in an intuitive form and can be downloaded. The online prediction tool allows users to perform neuropeptide predictions through simple operations, and the results are intuitive and easy to understand.

[0101] In one embodiment, a neuropeptide identification method based on contrastive learning and cross-attention mechanism is provided, in which a deep learning model called NeuroCL can be introduced as a neuropeptide identification model. The model adopts the ESM-2 large language model and integrates the cross-attention mechanism and contrastive learning to more effectively capture the complex relationships between sequences and focus on key biological features, thereby efficiently identifying neuropeptides.

[0102] The deep learning model named NeuroCL is used as the framework of the neuropeptide identification model. Figure 3As shown, it can include A. data acquisition and preprocessing; B. feature encoding and extraction; C. contrast loss learning; D. online prediction; E. self-attention mechanism processing; F. ESM-2 feature extraction. The computer device obtains a data set containing neuropeptides and non-neuropeptides and preprocesses the data set to obtain a training data set; uses a feature encoding strategy to encode the neuropeptide sequence corresponding to the training data set into a first eigenvector; uses the BLOSUM62 matrix to encode the neuropeptide sequence corresponding to the training data set into a second eigenvector; uses the protein language model ESM-2 to extract features from the training data set to obtain a large model feature; inputs the first eigenvector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; inputs the second eigenvector into a multi-scale convolutional neural network to obtain a second output feature; and merges the first output feature and the second output feature into a hand-coded output feature; processes the hand-coded output feature and the large model output feature based on a cross-attention mechanism, and inputs them into a multi-head attention mechanism and a feedforward neural network to obtain two target vectors; determines a contrast loss function, and the two target vectors are contrastively learned according to the contrast loss function, and dimensionality reduction processing is performed to construct a neuropeptide identification model to complete neuropeptide identification.

[0103] In one embodiment, the NeuroCL neuropeptide identification model introduced in this application is compared with the model performance of other existing models, such as Figure 4 As shown, the NeuroCL neuropeptide identification model demonstrated excellent performance across multiple evaluation metrics. Using independent test data, it achieved ACC of 93.8%, MCC of 87.8%, AUROC of 97.2%, Sp of 96.5%, and Sn of 91.1%, all exceeding existing models. Furthermore, through contrastive learning and cross-attention mechanisms, this model was able to more effectively capture complex relationships between sequences and samples, improving recognition efficiency.

[0104] This application provides a fast, low-cost, and highly accurate neuropeptide identification method based on contrastive learning and cross-attention mechanisms, which explores the cross-fusion between multiple feature dimensions, fully absorbs the contextual information between features, and improves the accuracy of the model; and uses a contrastive learning strategy between the features extracted by the pre-trained protein language model and the manually encoded features to improve the prediction accuracy of the model.

[0105] It should be understood that, although the various steps in the above flow chart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above flow chart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0106] In one embodiment, Figure 5 As shown, a neuropeptide identification system based on contrastive learning and cross-attention mechanism is provided, comprising: a data set acquisition module 510, a feature encoding module 520, a feature extraction module 530, a feature processing module 540 and a neuropeptide identification module 550, wherein:

[0107] The data set acquisition module 510 is used to acquire a data set containing neuropeptides and non-neuropeptides, and pre-process the data set to obtain a training data set;

[0108] A feature encoding module 520 is configured to encode the neuropeptide sequence corresponding to the training dataset into a first feature vector using a feature encoding strategy; encode the neuropeptide sequence corresponding to the training dataset into a second feature vector using a BLOSUM62 matrix; and extract features from the training dataset using a protein language model ESM-2 to obtain large model features.

[0109] Feature extraction module 530, configured to input the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; input the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and combine the first output feature and the second output feature into a manually encoded output feature;

[0110] Feature processing module 540 is used to process the manually encoded output features and the large model output features based on the cross-attention mechanism, and input them into the multi-head attention mechanism and the feedforward neural network to obtain two target vectors;

[0111] The neuropeptide identification module 550 is used to determine a contrast loss function, perform contrast learning on two target vectors according to the contrast loss function, perform dimensionality reduction processing, construct a neuropeptide identification model, and complete neuropeptide identification.

[0112] In one embodiment, the dataset acquisition module 510 is further configured to remove redundant neuropeptide sequences in the dataset, and divide the dataset after removing redundant sequences into a training dataset and a validation dataset.

[0113] In one embodiment, the feature encoding module 520 is further used to extract features of the neuropeptide sequence corresponding to the training data set using encoding strategies of ordinal coding, amino acid composition, dipeptide composition, position-specific amino acid composition, adaptive jumping dipeptide composition, and amino acid entropy to obtain a first feature vector.

[0114] In one embodiment, the feature extraction module 530 is also used to normalize the first feature vector and the large model feature, and input the processed features into the bidirectional long short-term memory network respectively; process the long sequence through the bidirectional long short-term memory network, and pass the captured context information for feature extraction to obtain the first output feature and the large model output feature.

[0115] In one embodiment, the feature extraction module 530 is also used to input the second feature vector into a multi-scale convolutional neural network, capture information through convolution kernels of different sizes in the multi-scale convolutional neural network; extract features of different scales from the second feature vector based on the captured information; and fuse features of different scales to obtain a second output feature.

[0116] In one embodiment, the feature processing module 540 is also used to input the manually encoded output features and the large model output features into the cross-attention mechanism to obtain interactive features; input the interactive features into the multi-head attention mechanism, divide the interactive features into various subspaces through the multi-head attention mechanism, and calculate the attention weights in each subspace to obtain attention-weighted features; input the attention-weighted features into the feedforward neural network for processing to obtain two target vectors.

[0117] In one embodiment, the neuropeptide identification module 550 is further configured to determine a contrastive loss function and calculate a similarity score between two target vectors; calculate a gradient based on the contrastive loss function and the similarity score, and update parameters based on the gradient to complete contrastive learning.

[0118] In one embodiment, a neuropeptide identification system based on contrastive learning and cross-attention mechanism can also include a model evaluation module for obtaining evaluation indicators, using the evaluation indicators to evaluate the neuropeptide identification model, and obtaining evaluation results; wherein the evaluation indicators include accuracy, sensitivity, specificity, Matthews correlation coefficient, and area under the ROC curve.

[0119] In one embodiment, a neuropeptide identification system based on contrastive learning and cross-attention mechanism can also include an identification display module for collecting a data set containing neuropeptides and non-neuropeptides to be identified and inputting it into a neuropeptide identification model; outputting the neuropeptide identification results through the neuropeptide identification model and displaying them on a page.

[0120] 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 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. 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 via a network connection. When the computer program is executed by the processor, a neuropeptide identification method based on contrastive learning and cross-attention mechanism 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 key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0121] Those skilled in the art will understand that Figure 6 The 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 shown in the figure, or combine certain components, or have a different component arrangement.

[0122] In one embodiment, a computer device is provided, comprising 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 a neuropeptide identification method based on contrastive learning and cross-attention mechanism are implemented.

[0123] 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 a neuropeptide identification method based on contrastive learning and cross-attention mechanism are implemented.

[0124] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. 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. By way of illustration and not limitation, RAM is available in various 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).

[0125] The technical features of the above embodiments can be combined arbitrarily. 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.

[0126] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A neuropeptide identification method based on contrastive learning and cross-attention mechanism, characterized in that: The method comprises: Acquiring a data set containing neuropeptides and non-neuropeptides, and preprocessing the data set to obtain a training data set; Using a feature encoding strategy to encode the neuropeptide sequence corresponding to the training data set into a first feature vector; using a BLOSUM62 matrix to encode the neuropeptide sequence corresponding to the training data set into a second feature vector; using a protein language model ESM-2 to extract features from the training data set to obtain a large model feature; Inputting the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; inputting the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and merging the first output feature and the second output feature into a manually encoded output feature; The hand-coded output features and the large model output features are processed based on the cross-attention mechanism and input into the multi-head attention mechanism and the feedforward neural network to obtain two target vectors; A contrast loss function is determined, and the two target vectors are subjected to contrast learning according to the contrast loss function, and dimensionality reduction processing is performed to construct a neuropeptide identification model to complete neuropeptide identification.

2. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The data set is preprocessed to obtain a training data set, including: Redundant sequences of neuropeptide sequences in the data set are removed, and the data set after removing redundancy is divided into a training data set and a validation data set.

3. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The neuropeptide sequence corresponding to the training data set is encoded into a first feature vector using a feature encoding strategy, including: The neuropeptide sequence corresponding to the training data set is feature extracted using coding strategies of ordinal coding, amino acid composition, dipeptide composition, position-specific amino acid composition, adaptive jumping dipeptide composition, and amino acid entropy to obtain a first feature vector.

4. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: Inputting the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature includes: Normalizing the first feature vector and the large model feature, and inputting the processed features into a bidirectional long short-term memory network respectively; The long sequence is processed by the bidirectional long short-term memory network, and the captured context information is transferred for feature extraction to obtain the first output feature and the large model output feature.

5. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The second feature vector is input into a multi-scale convolutional neural network to obtain a second output feature, including: Inputting the second feature vector into a multi-scale convolutional neural network, and capturing information through convolution kernels of different sizes in the multi-scale convolutional neural network; extracting features of different scales from the second feature vector based on the captured information; The features of different scales are fused to obtain the second output feature.

6. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The hand-coded output features and the large model output features are processed based on the cross-attention mechanism and input into the multi-head attention mechanism and feedforward neural network to obtain two target vectors, including: Input the hand-coded output features and the large model output features into the cross-attention mechanism to obtain the interactive features; Inputting the interaction features into a multi-head attention mechanism, dividing the interaction features into subspaces through the multi-head attention mechanism, and calculating the attention weights in each subspace to obtain attention weighted features; The attention weighted features are input into a feedforward neural network for processing to obtain two target vectors.

7. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: Determining a contrast loss function, and performing contrast learning on the two target vectors according to the contrast loss function, including: Determining a contrast loss function and calculating a similarity score between the two target vectors; The gradient is calculated based on the contrast loss function and the similarity score, and the parameters are updated based on the gradient to complete contrast learning.

8. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The method further comprises: Obtaining an evaluation index, and using the evaluation index to evaluate the neuropeptide identification model to obtain an evaluation result; The evaluation indicators include accuracy, sensitivity, specificity, Matthews correlation coefficient, and area under the ROC curve.

9. The neuropeptide identification method based on contrastive learning and cross-attention mechanism according to claim 1, characterized in that: The method further comprises: Collecting a data set containing neuropeptides and non-neuropeptides to be identified and inputting the data set into the neuropeptide identification model; The neuropeptide identification model is used to output the neuropeptide identification results and display them on a page.

10. A neuropeptide identification system based on contrastive learning and cross-attention mechanism, characterized in that: The system comprises: A data set acquisition module is used to acquire a data set containing neuropeptides and non-neuropeptides, and preprocess the data set to obtain a training data set; A feature encoding module is configured to encode the neuropeptide sequence corresponding to the training data set into a first feature vector using a feature encoding strategy; encode the neuropeptide sequence corresponding to the training data set into a second feature vector using a BLOSUM62 matrix; and extract features from the training data set using a protein language model ESM-2 to obtain large model features; a feature extraction module, configured to input the first feature vector and the large model feature into a bidirectional long short-term memory network to obtain a first output feature and a large model output feature; input the second feature vector into a multi-scale convolutional neural network to obtain a second output feature; and combine the first output feature and the second output feature into a manually encoded output feature; A feature processing module is used to process the manually encoded output features and the large model output features based on a cross-attention mechanism, and input them into a multi-head attention mechanism and a feedforward neural network to obtain two target vectors; The neuropeptide identification module is used to determine a contrast loss function, perform contrast learning on the two target vectors according to the contrast loss function, perform dimensionality reduction processing, construct a neuropeptide identification model, and complete neuropeptide identification.

Images