Nucleic acid binding protein recognition method based on protein map and protein language model

Abstract

The invention discloses a nucleic acid binding protein recognition method based on a protein map and a protein language model, and belongs to the technical field of bioinformatics. According to the method, high-dimensional sequence features extracted by a ProtT5 protein language model are combined with three-dimensional structure information predicted by AlphaFold, so that the model can capture rich feature information of the sequence and the structure at the same time, and the accuracy and reliability of NABPs prediction are remarkably improved; the expression capability of sequence features is enhanced through the GAT network and the BiLSTM network, so that the constructed prediction model can more comprehensively understand important information in the protein sequence, and the prediction performance of NABPs is improved; the features output by the GAT network and the BiLSTM network are fused, and NABPs classification is performed through the MLP, so that the prediction model can make full use of feature information from different sources, and the NABPs prediction performance and robustness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and particularly to a method for identifying nucleic acid binding proteins based on protein graphs and protein language models. Background Art

[0002] Nucleic acid binding proteins (NABPs) play a variety of important roles within cells. They regulate gene expression, maintain genomic stability, and participate in biological processes such as signal transduction by interacting with nucleic acids such as DNA and RNA. These proteins are crucial for the normal function and homeostasis of cells. Through the precise identification and in-depth study of these proteins, we can decipher their cellular functions and the complex regulatory networks in which they are involved. This exploration not only reveals the core principles of cell biology and molecular biology but also enhances our understanding of the complex mechanisms of gene regulation. Therefore, the efficient and accurate identification of NABPs is of great significance for understanding the complex biological responses in organisms and developing new diagnostic and therapeutic methods.

[0003] In the early stage, structural alignment and statistical potential were used to identify NABPs. However, these methods were affected by the limited number of protein three-dimensional (3D) structures. With the rise of artificial intelligence, more and more machine learning and deep learning algorithms have been used to construct NABPs prediction models based on sequence features. Some researchers developed the TriPepSVM model based on sequence features and support vector machines. Other researchers developed StackDPPred based on sequence features using stacking algorithms. Still other researchers used the random forest algorithm to establish a two-layer model, iDRBP-PPCT, based on the Position-Specific Scoring Matrix (PSSM) and the Position-Specific Frequency Matrix (PSFM). The first-layer model is used to distinguish NABPs from non-NABPs, and the second-layer model is used to distinguish RNA-binding proteins (RBPs), DNA-binding proteins (DBPs), and DNA / RNA-binding proteins (DRBPs). In addition, deep learning algorithms have also been widely used in the identification of NABPs. For example, iDRBP_MMC introduced a multi-label learning model based on convolutional neural networks to solve the cross-prediction problem, thereby improving the prediction performance of DNA-binding proteins and RNA-binding proteins. The iDRBP-ECHF model proposed in the paper "iDRBP-ECHF: Identifying DNA-and RNA-binding proteins based on extensible cubic hybrid framework" was constructed based on five sequence features and stacking ensemble learning algorithms, using two deep learning algorithms and two machine learning algorithms to establish the basic model. The iDRPro-SC proposed in the paper "iDRPro-SC: identifying DNA-binding proteins and RNA-binding proteins based on subfunction classifiers" was also constructed using a stacking ensemble learning algorithm based on PSSM features, in which BiLSTM was used to construct the basic model to distinguish the subfunctions of NABPs.

[0004] The above-mentioned existing technologies have the following problems:

[0005] Insufficient utilization of structural information: Due to the scarcity of 3D structural information of NABPs, existing models rarely utilize this structural information. However, structural information plays an important role in understanding the interaction between proteins and nucleic acids. Current sequence-based models, such as TriPepSVM, StackDPPred, and iDRBP_MMC, mainly rely on sequence features for prediction and fail to fully utilize 3D structural information, which limits the prediction ability and accuracy of the models.

[0006] Low data coverage: As described in iDRPro-SC, proteins with available structural information account for only 0.36% of all proteins in UniProtKB. This low coverage limits the wide application of the models. Many existing models (such as iDRBP-PPCT and iDRPro-SC) mainly rely on a small amount of 3D structural data, making it difficult for these models to generalize to a wider dataset.

[0007] Limited prediction performance: Due to the failure to fully utilize 3D structural information, the prediction performance of existing models is limited in some cases, especially when dealing with proteins with complex structures. For example, models such as StackDPPred based on sequence features and iDRBP-ECHF and iDRPro-SC based on stacked ensemble learning cannot fully capture the spatial information of proteins when dealing with proteins with complex structures, thus affecting the accuracy and performance of the prediction.

[0008] The above problems need to be solved urgently. Therefore, the present invention proposes a method for identifying nucleic acid-binding proteins based on protein graphs and protein language models. Summary of the Invention

[0009] The technical problem to be solved by the present invention is: how to solve the problems existing in the prior art, such as insufficient utilization of structural information, low data coverage, and limited prediction performance, and provide a method for identifying nucleic acid-binding proteins based on protein graphs and protein language models.

[0010] The present invention solves the above technical problems through the following technical solutions. The present invention includes the following steps:

[0011] S1: Dataset processing

[0012] Retrieve protein sequence data from the UniProt database, obtain the predicted 3D structural information of the corresponding proteins from the AlphaFold database, remove proteins with sequence similarity exceeding the set value and proteins for which structural information cannot be obtained, obtain the training dataset, and then select the test dataset.

[0013] S2: Feature extraction

[0014] Use the ProtT5 model to generate the embedding features of protein sequences, and represent the protein as a graph according to the predicted 3D structure information to obtain a protein graph, capturing its structural and sequence features;

[0015] S3: Model establishment

[0016] Use the GAT network, BiLSTM network and MLP to establish a prediction model. In the prediction model, enhance the feature representation through the GAT network and BiLSTM network, then connect the output features of the GAT network and BiLSTM network, and perform NABPs classification through the MLP to achieve the recognition of NABPs;

[0017] S4: Model training

[0018] Use the training dataset to train the prediction model to obtain the trained prediction model;

[0019] S5: NABPs recognition

[0020] Use the trained prediction model to perform NABPs recognition on the samples in the test dataset to obtain the recognition results.

[0021] Furthermore, in the step S1, the specific processing process is as follows:

[0022] S11: Search for protein sequences with DNA-binding and RNA-binding functions from the UniProt database and retrieve their corresponding annotation information;

[0023] S12: Retrieve the predicted 3D structure of the corresponding protein sequence from the AlphaFold database;

[0024] S13: Use the BLASTClust tool to remove protein sequences with sequence similarity exceeding the set value and remove protein sequences lacking predicted 3D structure information to obtain a training dataset;

[0025] S14: Select the datasets PDB255 and TEST474 as the test datasets.

[0026] Furthermore, in the step S2, the specific processing process is as follows:

[0027] S21: Use the ProtT5 model for protein language model embedding, extract the embedding features of each residue from the protein sequence, and generate a feature matrix of size L×1024, where L is the sequence length;

[0028] S22: Retrieve through the AlphaFold database to obtain the predicted 3D structure of the protein, and represent each protein as a graph, with each residue as a node, and the edges between the nodes determined according to their Euclidean distance, thus completing the construction of the protein graph.

[0029] Further, in the step S22, each protein is represented as an undirected graph G=(V, H, E, A), where V is the set of nodes in the graph G, V i ∈V, V i represents the i-th residual node, V is the set of all residual nodes in the protein, H is the set of all node features, h∈H |V|×d , |V| is the total number of nodes, d is the dimension of the node features, E is the set of all edges in the graph G, e ij ∈E represents the edge between node i and node j, A is the adjacency matrix of the graph G, if the Euclidean distance between any two nodes is less than or equal to then these two nodes are considered connected, otherwise these two nodes are considered unconnected.

[0030] Further, in the step S3, the GAT network uses the attention mechanism to obtain different weights of adjacent nodes, and then aggregates the information of the weighted adjacent nodes to update the node embeddings; in the prediction model, multiple stacked GAT layers are used to achieve the high-level representation of the nodes, and each GAT layer dynamically adjusts the node features by learning the relationship between the nodes and their adjacent nodes. In the last layer of the entire GAT network, average pooling is performed on the features of all nodes to obtain the feature representation of the entire protein, that is, the graph embedding representation.

[0031] Further, in the step S3, the BiLSTM network can capture the relationships inside the sequence from two directions to maintain the integrity of the sequence semantics. The embedding features generated by the ProtT5 model are processed within the LSTM. The forget gate determines which information to discard from the cell state, while the new information is merged through the input gate and cell state update. The new information includes the embedding features and the data from the previous time step.

[0032] Further, in the step S3, the processing formula of the MLP is as follows:

[0033]

[0034] where, G ′ represents the output of the GAT network, h birepresents the output of the BiLSTM network, concat represents the Concat function, which is used to concatenate the output features of the GAT network and the BiLSTM network, and softmax represents the softmax function, which is used to map the result to the range between 0 and 1. represents the prediction result.

[0035] The present invention has the following advantages compared with the prior art:

[0036] 1. By combining the high-dimensional sequence features extracted by the ProtT5 protein language model with the three-dimensional structure information predicted by AlphaFold, the model can capture rich feature information of both the sequence and the structure simultaneously, significantly improving the accuracy and reliability of NABPs prediction.

[0037] 2. By representing the protein as a graph structure and introducing a graph attention neural network to process these graphs, the GAT network assigns different weights to each node through the attention mechanism to integrate the structural information of the nodes. At the same time, the BiLSTM network is used to process the sequence embedding features extracted from the ProtT5 model to capture the long-range dependencies in the sequence. Both the GAT network and the BiLSTM network enhance the representation ability of the sequence features, enabling the constructed prediction model to more comprehensively understand the important information in the protein sequence, thereby improving the prediction performance of NABPs.

[0038] 3. The features output by the GAT network and the BiLSTM network are fused, and the classification of NABPs is performed through an MLP, enabling the prediction model to make full use of the feature information from different sources and improving the performance and robustness of NABPs prediction. Brief Description of the Drawings

[0039] Figure 1 is a schematic flowchart of the method for identifying nucleic acid-binding proteins based on a protein graph and a protein language model according to the present invention;

[0040] Figure 2 is a comparison graph of the AUROC values of the GraphNABP model and other methods on PDB255 in an embodiment of the present invention, where A: DNA-binding protein, B: RNA-binding protein;

[0041] Figure 3 is the ROC curve of DNA-binding proteins in the PDB255 dataset in an embodiment of the present invention;

[0042] Figure 4 is the ROC curve of RNA-binding proteins in the PDB255 dataset in an embodiment of the present invention;

[0043] Figure 5It is the ROC curve of DNA-binding proteins in the TEST474 dataset in the embodiments of the present invention;

[0044] Figure 6 It is the ROC curve of RNA-binding proteins in the TEST474 dataset in the embodiments of the present invention;

[0045] Figure 7 It is the performance graph of predicting known nucleic acid-binding proteins in the tomato genome by different methods in the embodiments of the present invention;

[0046] Figure 8 It is the result graph of the case study of Q7AKE4 in the embodiments of the present invention. Among them, A: the heat map of the attention weights of the Q7AKE4 sequence, B: the adjacency matrix graph of the active cavity of Q7AKE4 predicted by CavityPlus, C: the structural visualization graph of Q7AKE4 with residues E64, A94, G137, L142, L150. Detailed implementation manners

[0047] The embodiments of the present invention will be described in detail below. These embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given. However, the protection scope of the present invention is not limited to the following embodiments.

[0048] Embodiment 1

[0049] As Figure 1 shown, this embodiment provides a technical solution: a method for identifying nucleic acid-binding proteins based on protein graphs and protein language models, mainly including the following steps:

[0050] The following steps:

[0051] 1. Dataset processing

[0052] Retrieve protein sequence data from the UniProt database, and obtain the predicted 3D structure information of the corresponding proteins from the AlphaFold database. Remove proteins with sequence similarity exceeding the set value and proteins for which structure information cannot be obtained to obtain a training dataset, and then select a test dataset;

[0053] 2. Feature extraction

[0054] Use the ProtT5 model to generate the embedding features of the protein sequence, and represent the protein as a graph according to the predicted 3D structure information to obtain a protein graph, and capture its structural and sequence features;

[0055] 3. Model establishment

[0056] A prediction model is established using the GAT network, BiLSTM network, and MLP. In the prediction model, the feature representation is enhanced through the GAT network and BiLSTM network, and then the output features of the GAT network and BiLSTM network are concatenated and classified into NABPs through MLP to achieve the recognition of NABPs.

[0057] 4. Model Evaluation and Prediction Result Analysis

[0058] 5-fold cross-validation is used for model selection. The evaluation metrics include precision, recall, F1-score, MCC, and AUROC, and the contribution of different modules to the model performance is evaluated through ablation experiments. The model is validated on the PDB255 and TEST474 test sets, and NABPs prediction is carried out on the tomato genome.

[0059] Step 1: Dataset Processing

[0060] The benchmark dataset is obtained from the UniProt database and screened according to the functions of proteins. Specifically, proteins with DNA-binding and RNA-binding functions are searched, and their corresponding annotation information (UniProt ID, functional description, protein name, sequence information, etc.) is retrieved. These annotation information is used to ensure the accuracy of the data and provide detailed background information for subsequent processing. To ensure data quality and reliability, the BLASTClust tool is used to exclude proteins with sequence similarity exceeding 35%. According to this processing flow, the final benchmark dataset includes 10,188 non-NABPs and 11,875 NABPs. These NABPs are further classified into 7,594 DNA-binding proteins (DBPs), 3,948 RNA-binding proteins (RBPs), and 333 proteins that bind both DNA and RNA (DRBPs). When constructing the GraphNABP model, we utilized the protein structure information predicted by AlphaFold. By querying the UniProt ID, we can retrieve the predicted 3D structure of the corresponding protein from the AlphaFold database. It should be noted that due to the lack of predicted 3D structures of several proteins, we removed these proteins from the dataset. After the above processing, the training dataset has 7,070 DBPs, 3,591 RBPs, and 9,732 non-NABPs.

[0061] To evaluate and fairly compare our model with other methods, we used two widely used test datasets for NABPs prediction tasks: PDB255 and TEST474. PDB255 contains 93 DBPs, 70 RBPs, and 92 non-NABPs. TEST474 includes 183 DBPs, 76 RBPs, and 223 non-NABPs. The predicted structures of proteins in the test datasets were also obtained from AlphaFold.

[0062] Step 2: Feature Extraction

[0063] The feature extraction method of the present invention comprehensively utilizes protein language model embeddings and three-dimensional (3D) structure information. First, the ProtT5 model is used for protein language model embeddings to extract the embedding features of each residue from the protein sequence, generating a feature matrix of size L×1024, where L is the sequence length. These embedding features capture the semantic and contextual information of the residues. Then, the protein structure is retrieved through the AlphaFold database to obtain the predicted 3D structure of the protein, and each protein is represented as a graph with each residue as a node, and the edges between the nodes are determined according to their Euclidean distance (less than or equal to 14 angstroms), thus completing the construction of the protein graph. In this way, GraphNABP can comprehensively capture and integrate the sequence and structure information of proteins, providing rich feature representations for subsequent NABPs prediction.

[0064] 2.1 Protein Language Model Embeddings

[0065] Protein language models have been widely used in current bioinformatics research. These models utilize the Transformer architecture in the field of natural language processing and are adjusted and optimized for protein sequence analysis to better understand the semantic information of protein sequences. Among them, the widely used ProtT5 model is pre-trained on a large number of unlabeled protein sequences. Existing studies have shown that protein language models have excellent performance in predicting hot-spot residues at the protein-DNA binding interface and may have great potential in nucleic acid binding protein prediction. In the present invention, we use ProtT5 to extract the embedding features of each residue and then use them for model construction. Each protein sequence is encoded as a feature matrix of size L×1024, where L is the sequence length.

[0066] 2.2 Protein Structure Retrieval and Protein Graph Construction

[0067] The protein structures in our dataset are from the AlphaFold database, and all atomic coordinates of the proteins are from the PDB files provided by the AlphaFold database. Based on the obtained predicted 3D structures, each protein is represented as a graph, with each residue as a node. The distance matrix is calculated according to the coordinates of the alpha-carbon atoms of each residue. Formally, each protein is represented as an undirected graph G=(V, H, E, A), where V is the set of nodes in graph G, V i ∈V (V i represents the i-th residue node, and V is the set of all residue nodes in the protein), H is the set of all node features, h∈H |V|×d , where |V| is the total number of nodes and d is the dimension of the node features. E is the set of all edges in the protein graph, where e ij ∈E represents the edge between node i and node j. A is the adjacency matrix of the graph. If the Euclidean distance between any two nodes is less than or equal to 14 angstroms then these two nodes are considered connected, and the judgment method is as follows:

[0068]

[0069] where 1 means the two nodes are connected and 0 means the two nodes are not connected.

[0070] Step 3: Model establishment

[0071] In the model establishment part of the present invention, the graph attention (GAT) network and the bidirectional long short-term memory (BiLSTM) network are combined to enhance the feature representation. First, the GAT network is used to process the protein graph, and the attention mechanism is used to assign different weights to each node, so as to integrate the structural information of the nodes. Then, the BiLSTM network is used to process the sequence embedding features extracted from the ProtT5 model to capture the long-distance dependencies in the sequence. Finally, the output features of the GAT network and the BiLSTM network are connected, and the NABPs classification is performed through a multi-layer perceptron (MLP) to achieve efficient prediction. In this way, the GraphNABP model (prediction model) can make full use of the sequence and structural information of the protein to improve the accuracy and reliability of the prediction.

[0072] 3.1 Graph attention (GAT) network

[0073] The GAT network utilizes the attention mechanism to obtain different weights for adjacent nodes, and then aggregates the information of these weighted adjacent nodes to update the node embeddings. In our prediction model (GraphNABP model), multiple stacked GAT layers are used to achieve the high-level representation of nodes. Each GAT layer dynamically adjusts the node features by learning the relationships between nodes and their adjacent nodes. At the last layer of the entire GAT network, average pooling is performed on the features of all nodes to obtain the feature representation of the entire protein. The principle of the GAT network can be described by the following formulas:

[0074]

[0075] In the first formula above, h i and h j represent the embeddings of node i and its adjacent node j respectively, where ( is the set of adjacent nodes of node i). W represents the learnable parameter matrix of different layers, a is the activation function, and e ij is the attention coefficient. The third formula is an extension of the second formula, and the softmax operation is performed on the attention coefficients of each adjacent node to obtain the attention coefficients between node i and its adjacent nodes. Where ∥ represents the concatenation operation of feature vectors. In the fourth formula, the obtained attention weights are used to update the feature vector of node i. In the fifth formula, average pooling is performed on all updated nodes to obtain the updated feature vector of the entire graph, where represents the total number of nodes in the graph.

[0076] 3.2 Bidirectional Long Short-Term Memory (BiLSTM) Network

[0077] The Bidirectional Long Short-Term Memory (BiLSTM) network is used to enhance the representation of the sequence information generated by the protein language model. The BiLSTM network can capture the relationships within the sequence from two directions to maintain the integrity of the sequence semantics. Specifically, the embedding representation generated by the protein language model is processed within the LSTM. The "forget gate" determines which information to discard from the cell state, and the new information, including the embedding representation and the data from the previous time step, is merged through the "input gate" and "cell state update".

[0078] For the forward LSTM, given the input X = (x 1 , x 2 , …, x T ), where T is the sequence length. The calculation formulas are as follows:

[0079] i t = σ(W ix x t + W ih h t-1+b i )

[0080] f t = σ(W fx x t + w fh h t-1 +b f )

[0081] o t = σ(W ox x t + W oh h t-1 +b o )

[0082] C t = f t C t-1 + i t tanh(W cx x t + W ch h t-1 +b c )

[0083] h t = o t tanh(C t )

[0084] where σ represents the sigmoid activation function, and i t , f t , C t and h t represent the input gate, forget gate, cell state, and output gate at time step t, respectively. W and b are the weight matrix and bias vector, respectively.

[0085] For the backward LSTM, the input is represented as X rev = (x T , x T-1 , …, x 1 ). In the LSTM module, the final backward hidden state is represented as The final output is obtained by concatenating the forward and backward hidden states

[0086] 3.3 Multilayer Perceptron (MLP)

[0087] The outputs of GAT and BiLSTM are concatenated and then a Multilayer Perceptron (MLP) is used to classify the NABPs. The specific formula is as follows:

[0088]

[0089] The Concat function concatenates the graph embedding representation with the hidden state of the BiLSTM and passes it to a multi-layer perceptron for classification. The softmax function maps the result to a range between 0 and 1, representing the prediction result. (The hyperparameters used for predicting model training are shown in Table 1)

[0090] Table 1 Hyperparameters used for GraphNABP model training

[0091]

[0092]

[0093] Step 4: Model evaluation and prediction result analysis

[0094] 4.1 Cross-validation and evaluation metrics

[0095] During the model training process, we used 5-fold cross-validation for model selection. To measure the performance of the model, we used the following metrics for evaluation:

[0096]

[0097]

[0098] where TP is true positive, FP is false positive, TN is true negative, and FN is false negative.

[0099] 4.2 Ablation experiments

[0100] To evaluate the contributions of different modules to the GraphNABP model, we conducted an ablation study on the training dataset using 5-fold cross-validation. Considering the significant differences in structure and function between DNA-binding proteins and RNA-binding proteins, we conducted ablation experiments for these two tasks separately. Based on the embeddings obtained from the ProtT5 model, our model uses GAT and BiLSTM networks in parallel to enhance the feature representation. Therefore, each of the two networks was removed for ablation analysis. As shown in Table 2, in terms of the AUROC value, the cross-validation performance of the BiLSTM-only and GAT-only models is worse than that of GraphNABP, which demonstrates the effectiveness of the proposed framework. In addition, we noticed that the prediction performance of RBPs is slightly worse than that of DBPs, which may be due to the fact that there is less training data available for RBPs compared to DBPs.

[0101] Table 2 Ablation results of the GraphNABP model

[0102]

[0103] 4.3 Prediction result analysis

[0104] 4.3.1 Prediction Results for PDB255

[0105] To evaluate the generalization performance of the GraphNABP model, we conducted a fair performance comparison with other state-of-the-art methods on the PDB255 dataset, which is widely used to evaluate NABPs prediction models. It should be noted that most nucleic acid-binding protein prediction models lack open-source code, so we collected the prediction performance reports of other models from the literature. As Figures 2 - 4 shown, on the PDB255 dataset, GraphNABP achieved an AUROC value of 0.8889 in the DNA-binding protein (DBP) prediction task, outperforming the second-best method, iDRBP-ECHF, by 7.29 percentage points and significantly exceeding the other three methods. In the RNA-binding protein (RBP) prediction task, the GraphNABP model achieved an AUROC value of 0.8297, slightly higher than that of iDRBP-ECHF (0.828). These results indicate that the GraphNABP model performs excellently in predicting NABPs.

[0106] 4.3.2 Prediction Performance on TEST474

[0107] Compared with PDB255, TEST474 contains a larger amount of data, and the distribution of positive and negative samples is unbalanced, reflecting the true distribution of NABPs. As shown in Table 3, for DBP prediction, we compared the prediction performance of GraphNABP with 8 state-of-the-art methods (DNAbinder, stackDPPred, DPP-PseAAC, DeepDRBP-2L, iDRBP_MMC, IDRBP-PPCT, iDRBP-ECHF, iDRPro-SC). The AUROC value of the GraphNABP model is 0.9732, which is nearly 2.3 percentage points higher than that of the second-best method, iDRBP-ECHF (AUROC value of 0.950). In addition, in terms of indicators such as ACC, PRE, REC, F1, and MCC, the GraphNABP model obtained values of 0.9367, 0.9408, 0.9344, 0.9194, and 0.8676 respectively, all significantly higher than other prediction methods, indicating a significant improvement in performance and highlighting the effectiveness of our model in model design and feature utilization. In addition, for RBP prediction, we made a detailed comparison with 11 state-of-the-art methods, as shown in Table 3. An AUROC of 0.9201 was obtained on the TEST474 dataset, which is very close to that of iDRBP-ECHF (AUROC: 0.925) and is better than other methods reporting AUROC values. In addition, the MCC value of the GraphNABP model is 0.7409, which is 2.09 percentage points higher than that of the second-best model (0.72). Combining the results of DNA-binding protein prediction for TEST474 and nucleic acid-binding protein prediction for PDB255, the GraphNABP model demonstrated excellent ability in predicting nucleic acid-binding proteins. Figures 5 - 6 Figure 2 shows the ROC curve of the GraphNABP model on the TEST474 dataset.

[0108] Table 3 Comparison of the GraphNABP model with other methods on TEST474

[0109]

[0110]

[0111] 4.3.3 Prediction of NABPs in the tomato genome

[0112] To verify the practical application of this prediction model, we predicted the NABPs of the tomato genome. Specifically, using GO terms for prediction, 577 known DNA-binding proteins and 171 RNA-binding proteins were obtained from the tomato genome dataset (ITAG4.1). The GraphNABP model identified 557 NBPs (96.53%) and 167 RBPs (97.66%) from the known NABPs. To compare with other methods, we used the available web servers of iDRBP-ECHF, iDRPro-SC, and iDRBP_MMC to predict the corresponding proteins. As Figure 7 shown, the GraphNABP model is significantly superior to other prediction methods, demonstrating its generality and superior performance in predicting NABPs in real-world scenarios.

[0113] In this invention, the GraphNABP model, a nucleic acid-binding protein (NABPs) prediction framework that combines graph attention neural network (GAT) and protein language model, is proposed. The GraphNABP model significantly improves the prediction performance of NABPs by integrating protein sequence and predicted three-dimensional structure information. Specifically, sequence embeddings are first obtained from the ProtT5 protein language model, and the 3D structure of the protein is predicted by AlphaFold. Then, graph attention network (GAT) and bidirectional long short-term memory network (BiLSTM) are used to enhance feature representation, and finally, NABPs are predicted through a fully connected layer. To our knowledge, this is the first time that AlphaFold and protein language model are combined for the prediction of NABPs.

[0114] Most of the existing technologies only use the sequence information of proteins for NABPs prediction, ignoring the three-dimensional structure information, which limits the prediction ability and accuracy of the model. However, GraphNABP realizes more efficient feature extraction and accurate prediction by comprehensively using sequence and structure information. The GraphNABP model has the following advantages: First, the GraphNABP model represents proteins as graph structures by combining the high-dimensional sequence features extracted by the ProtT5 model and the three-dimensional protein structure information predicted by AlphaFold. This method enables the model to capture more diverse and rich feature information, thus significantly improving the accuracy and reliability of NABPs prediction. Second, GraphNABP adopts the GAT network to process the protein graph structure, using the attention mechanism to assign different weights to each node, thereby more effectively integrating the structural information of the nodes. The GAT network can better capture the topological structure information of proteins and enhance the accuracy of feature representation. Compared with traditional methods, the GraphNABP model performs better in dealing with complex protein structures and can more accurately predict NABPs. In addition, the bidirectional long short-term memory (BiLSTM) network is used to process the sequence embedding features extracted from the ProtT5 model, which can capture the long-range dependencies in the sequence. The BiLSTM network enhances the representation ability of sequence features, enabling the model to more comprehensively understand the important information in the protein sequence, thereby improving the prediction performance of NABPs. By combining the GAT and BiLSTM neural networks, the GraphNABP model fully utilizes the complementarity between protein structure and sequence information, improving the performance and robustness of NABPs prediction. Experimental results show that the GraphNABP model outperforms the existing state-of-the-art methods on multiple independent test sets and the tomato genome. Moreover, the GraphNABP model enhances the transparency and credibility of the model through interpretability analysis, visualizing the attention weight matrix of the GAT network, enabling us to understand which residues play a key role in NABPs prediction. These advantages and effects make the GraphNABP model significantly innovative and practical in the existing technologies.

[0115] Example 2

[0116] In this example, we further discuss the effects of the embeddings obtained from different protein language models. To this end, we compare the performance of the embeddings obtained from three widely used protein language models (ProtBert, ESM-1b, and ProtT5). As shown in Table 4, in terms of cross-validation performance, ESM-1b and ProtT5 have the highest AUROC for DBP prediction, which is 0.9715, and ProtT5 has the highest AUROC for RBP prediction, which is 0.9630. Considering the performance of the two prediction tasks, we select the ProtT5 model as the model for embedding feature extraction.

[0117] Table 4 5-fold cross-validation performance based on different protein language models

[0118]

[0119] Interpretability analysis: The GraphNABP model uses a graph attention (GAT) network to capture the topological structure information of proteins. In the present invention, we visualize the attention weights learned by the GAT network and analyze the potential factors that contribute to the accurate prediction of NABPs by the model. Specifically, we extract the attention weight matrix generated by the last GAT layer of the GraphNABP model and sum the weights at each sequential position to represent the contribution of each residue to the prediction of NABPs. By quantifying the contribution of residues, the interpretability of the GraphNABP model is significantly enhanced, providing insights into the interaction between nucleic acids and proteins. The following is a case study of NABP Q7AKE4 from TEST474. Q7AKE4 is a transcription factor that is essential for the formation of aerial hyphae on rich medium and binds to the ramC promoter. The visualization of the attention weights of the Q7AKE4 sequence is shown as Figure 8 shown in A below, where the attention weights of some residues are significantly higher than those at other positions, indicating that these high-weight residues play a crucial role in the entire sequence. These residues (E64, A94, G137, L142, and L150) are mapped to the three-dimensional structure of the protein, as shown in Figure 8 C below, and these residues are located in a potential protein pocket region. To confirm this, we used CavityPlus to predict a protein pocket with high confidence (as shown in Figure 8 B below), including residues E64, A94, G137, L142, and L150. These findings indicate that the GraphNABP model has successfully identified a potential nucleic acid binding region, further verifying the strong prediction performance of the model.

[0120] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for identifying nucleic acid binding proteins based on protein graphs and protein language models, characterized in that: The following steps are involved: S1: Dataset processing Retrieve protein sequence data from the UniProt database and obtain the predicted 3D structure information of the corresponding proteins from the AlphaFold database. Remove proteins whose sequence similarity exceeds the set value and proteins whose structural information cannot be obtained to obtain a training data set, and then select a test data set; S2: Feature extraction The ProtT5 model is used to generate embedded features of protein sequences, and the proteins are represented as graphs based on the predicted 3D structure information to obtain protein graphs that capture their structural and sequence features; S3: Model building A prediction model is established using the GAT network, BiLSTM network and MLP. In the prediction model, the feature representation is enhanced by the GAT network and the BiLSTM network. Then the output features of the GAT network and the BiLSTM network are connected, and the NABPs are classified by MLP to realize the recognition of NABPs. S4: Model training Using the training data set to train the prediction model, and obtaining a trained prediction model; S5: NABPs identification The trained prediction model is used to perform NABPs recognition on samples in the test data set to obtain the recognition results.

2. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 1, characterized in that: In step S1, the specific processing process is as follows: S11: Search for protein sequences with DNA binding and RNA binding functions from the UniProt database and retrieve their corresponding annotation information; S12: The predicted 3D structure of the corresponding protein sequence was retrieved from the AlphaFold database; S13: Use the BLASTClust tool to remove protein sequences whose sequence similarity exceeds a set value and remove protein sequences that lack predicted 3D structure information to obtain a training data set; S14: Datasets PDB255 and TEST474 are selected as test data sets.

3. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 2, characterized in that: In step S2, the specific processing process is as follows: S21: Use the ProtT5 model to embed the protein language model, extract the embedding features of each residue from the protein sequence, and generate a feature matrix of size L×1024, where L is the sequence length; S22: Search through the AlphaFold database to obtain the predicted 3D structure of the protein, and represent each protein as a graph, with each residue as a node and the edges between nodes determined according to their Euclidean distance, thereby completing the construction of the protein graph.

4. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 3, characterized in that: In step S22, each protein is represented as an undirected graph G=(V,H,E,A), where V is the set of nodes in the graph G, V i ∈V,V i represents the ith residual node, V is the set of all residual nodes in the protein, H is the set of all node features, h∈H |V|×d , |V| is the total number of nodes, d is the dimension of node features, E is the set of all edges in graph G, e ij ∈E represents the edge between node i and node j, A is the adjacency matrix of graph G, if the Euclidean distance between any two nodes is less than or equal to If , the two nodes are considered to be connected, otherwise, the two nodes are considered to be disconnected.

5. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 4, characterized in that: In step S3, the GAT network uses an attention mechanism to obtain different weights of adjacent nodes, and then aggregates the information of weighted adjacent nodes to update the node embedding; in the prediction model, multiple stacked GAT layers are used to achieve high-level representation of nodes, and each GAT layer dynamically adjusts node features by learning the relationship between nodes and adjacent nodes. In the last layer of the entire GAT network, the features of all nodes are averaged and pooled to obtain the feature representation of the entire protein, that is, the graph embedding representation.

6. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 5, characterized in that: In step S3, the BiLSTM network is able to capture the relationship within the sequence from two directions to maintain the integrity of the sequence semantics. The embedded features generated by the ProtT5 model are processed within the LSTM, and the forget gate determines which information to discard from the cell state, while the new information is merged through the input gate and the cell state update. The new information includes the embedded features and the data of the previous time step.

7. The method for identifying nucleic acid binding proteins based on protein graphs and protein language models according to claim 6, characterized in that: In step S3, the processing formula of MLP is as follows: Among them, G ′ represents the output of the GAT network, h bi represents the output of the BiLSTM network, concat represents the Concat function, which is used to connect the output features of the GAT network and the BiLSTM network, and softmax represents the softmax function, which is used to map the result to the range between 0 and 1. Represents the prediction result.

Images