Protein-nucleic acid binding residue recognition method and related equipment

Abstract

The embodiment of the invention belongs to the technical field of artificial intelligence, and relates to a protein-nucleic acid binding residue recognition method and related equipment, and the method comprises the following steps: obtaining an amino acid sequence and three-dimensional structure data of a to-be-recognized protein; inputting the amino acid sequence into a trained protein language model for sequence feature extraction operation to obtain sequence feature data; inputting the three-dimensional structure data into a sparse convolutional network in a point cloud form for structural feature extraction operation to obtain structural feature data; performing weighted fusion operation on the sequence feature data and the structural feature data to obtain weighted fusion features; and inputting the weighted fusion features into a trained multi-layer perceptron classifier for identification operation to obtain probability information that each residue is a binding site. According to the invention, the binding site of the protein can be accurately and reliably identified.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a protein-nucleic acid binding residue identification method and related equipment. Background Art

[0002] Sequence-based protein-nucleic acid binding residue identification methods rely on the primary structure of the protein (i.e., the amino acid sequence). This method usually predicts residues by extracting local or global features in the protein sequence. In recent years, with the construction of large-scale protein sequence databases and the progress of deep learning technology, sequence-based prediction methods have developed rapidly.

[0003] Early sequence analysis methods are usually based on traditional machine learning models, such as support vector machines (SVM), random forests (RF), and extreme gradient boosting (XGBoost). These methods mainly rely on the physicochemical properties and amino acid features of proteins to construct feature vectors and classify these features using supervised learning. For example, Yan et al. proposed an SVM model based on the physicochemical properties of amino acids that can identify key binding residues in proteins. The advantage of this type of method is that it is relatively simple and does not require three-dimensional structural information of proteins, so the computational cost is low. In addition, sequence data is easier to obtain than three-dimensional structural data because it can be obtained directly through sequencing technology. However, the main limitation of this type of method is that they cannot capture the three-dimensional spatial information of proteins, ignore the spatial position of residues and their interaction with nucleic acids, so their prediction performance is often insufficient in the scenario of complex binding sites.

[0004] With the improvement of large-scale data and computing power, deep learning technology has begun to be widely used in sequence analysis tasks. For example, large-scale pre-training models proposed in recent years, such as ESM (Evolutionary Scale Modeling) and ProtBERT[8], have significantly improved the effect of sequence analysis. Such models can better identify important features in sequences by learning deep contextual information from a large number of protein sequences. The ESM model proposed by Rao et al. uses the Transformer structure to model protein sequences and learns rich semantic information from millions of protein sequences, so that it can effectively capture the positional characteristics of binding residues. Similarly, ProtBERT enables the model to understand the interactions between different amino acids by encoding the context of amino acid sequences. However, these sequence-based deep learning methods still face some limitations. First, they cannot fully utilize the three-dimensional structural information of proteins, which is crucial for the spatial relationship of residues. Second, in some cases, it is difficult to distinguish subtle differences in binding sites based on sequence information alone, especially in highly conserved sequence regions. Therefore, although sequence-based deep learning methods perform well in specific scenarios, they still have bottlenecks in predicting complex binding sites.

[0005] In order to improve the prediction effect, many researchers have also tried to enhance the sequence-based model by introducing the evolutionary information of proteins (such as PSSM and HMM profiles). Evolutionary information reflects the conservation of proteins in different species and can help identify important binding sites. Jones et al. used evolutionary information to improve the sequence-based residue recognition model and significantly improved the prediction accuracy by combining sequence features and evolutionary features. However, the acquisition of evolutionary information usually relies on multiple sequence alignments (MSA), which is not only computationally expensive, but also fails to obtain sufficient information in some cases (such as when there are fewer protein families). In addition, evolutionary information is mainly based on sequence similarity and has a weak ability to capture spatial geometric features.

[0006] It can be seen that the traditional sequence-based protein-nucleic acid binding residue identification method has the problem of low recognition accuracy. Summary of the invention

[0007] The purpose of the embodiments of the present application is to provide a protein-nucleic acid binding residue identification method and related equipment to solve the problem of low recognition accuracy in traditional sequence-based protein-nucleic acid binding residue identification methods.

[0008] In order to solve the above technical problems, the present application embodiment provides a method for identifying protein-nucleic acid binding residues, which adopts the following technical solution:

[0009] Obtain the amino acid sequence and three-dimensional structure data of the protein to be identified;

[0010] Inputting the amino acid sequence into a trained protein language model to perform sequence feature extraction to obtain sequence feature data;

[0011] Inputting the three-dimensional structure data in the form of point cloud into a sparse convolutional network to perform a structure feature extraction operation to obtain structure feature data;

[0012] Performing a weighted fusion operation on the sequence feature data and the structural feature data to obtain a weighted fusion feature;

[0013] The weighted fusion features are input into a trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

[0014] Furthermore, after the step of inputting the three-dimensional structure data in the form of point cloud into a sparse convolutional network to perform a structure feature extraction operation to obtain the structure feature data, the following steps are also included:

[0015] The structural feature data is adaptively adjusted according to the attention mechanism, wherein the attention mechanism is expressed as:

[0016]

[0017] Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

[0018] Furthermore, before the step of inputting the weighted fusion features into the trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site, the following steps are also included:

[0019] An initial multi-layer perceptron classifier is called, and a model training operation is performed on the initial multi-layer perceptron classifier based on a self-supervised learning method to obtain the trained multi-layer perceptron classifier.

[0020] Furthermore, the loss function of the model training operation is expressed as:

[0021]

[0022] Among them, z i and z j represents the protein feature vector after different data enhancements, and τ is the temperature parameter.

[0023] Furthermore, the weighted fusion feature is expressed as:

[0024] F fusion =α·F seq +β·F struct

[0025] Among them, α and β are learnable weight parameters, which control the contribution ratio of sequence features and structural features to the final prediction respectively. seq represents the sequence feature data, F struct Represents the structural feature data.

[0026] Furthermore, the probability information is expressed as:

[0027] p i =σ(W·F fusion +b)

[0028] Where W is the weight matrix, b is the bias term, and σ is the Sigmoid activation function, which is used to normalize the output value to the interval [0,1][0,1][0,1].

[0029] In order to solve the above technical problems, the present application also provides a protein-nucleic acid binding residue recognition device, which adopts the following technical solution:

[0030] A data acquisition module is used to obtain the amino acid sequence and three-dimensional structure data of the protein to be identified;

[0031] A sequence feature extraction module, used for inputting the amino acid sequence into a trained protein language model to perform a sequence feature extraction operation to obtain sequence feature data;

[0032] A structural feature extraction module, used for inputting the three-dimensional structural data in the form of point cloud into a sparse convolutional network to perform a structural feature extraction operation to obtain structural feature data;

[0033] A weighted fusion module, used for performing a weighted fusion operation on the sequence feature data and the structure feature data to obtain a weighted fusion feature;

[0034] The recognition module is used to input the weighted fusion features into a trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

[0035] Furthermore, the device also includes:

[0036] The adaptive adjustment module is used to perform an adaptive adjustment operation on the structural feature data according to an attention mechanism, wherein the attention mechanism is expressed as:

[0037]

[0038] Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

[0039] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0040] The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the above-mentioned protein-nucleic acid binding residue identification method when executing the computer-readable instructions.

[0041] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0042] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the protein-nucleic acid binding residue identification method described above are implemented.

[0043] The present application provides a method for identifying protein-nucleic acid binding residues, including: obtaining the amino acid sequence and three-dimensional structure data of the protein to be identified; inputting the amino acid sequence into a trained protein language model to perform sequence feature extraction operations to obtain sequence feature data; inputting the three-dimensional structure data into a sparse convolutional network in the form of a point cloud to perform structure feature extraction operations to obtain structure feature data; performing weighted fusion operations on the sequence feature data and the structure feature data to obtain weighted fusion features; inputting the weighted fusion features into a trained multi-layer perceptron classifier for recognition operations to obtain probability information of each residue as a binding site. Compared with the prior art, the present application significantly improves the prediction accuracy by introducing a cross-modal learning framework by combining the sequence and three-dimensional structure information of the protein; in addition, the present application introduces an atomic-level attention mechanism, which enables the model to dynamically capture the geometric changes of the protein, thereby improving the adaptability and robustness in complex scenarios; the present application also enhances the generalization ability of the model on small-scale data sets through a self-supervised learning strategy, and effectively solves the challenges brought by data imbalance. In summary, this application achieves an overall improvement in the performance of protein-nucleic acid binding residue identification through multimodal fusion, self-supervised learning, and dynamic structure adjustment technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0046] Figure 2 is a flowchart of the method for identifying protein-nucleic acid binding residues provided in the embodiments of the present application;

[0047] Figure 3 It is a schematic diagram of the overall process of the model provided in the embodiment of the present application;

[0048] Figure 4 It is a schematic diagram of an embodiment of the quantitative performance comparison of the model provided in the embodiments of the present application;

[0049] Figure 5 It is a schematic diagram of an embodiment of the comparison of performance improvement of various parts of the model provided in the embodiment of the present application;

[0050] Figure 6 is a schematic diagram of an embodiment of the protein language model performance comparison provided in the embodiments of the present application;

[0051] Figure 7 It is a schematic diagram of an embodiment of the atomic-level attention mechanism improvement comparison provided by the embodiments of the present application;

[0052] Figure 8 The embodiment of the present application provides a schematic diagram of an embodiment of the present application;

[0053] Fig. 9 is a schematic diagram of the structure of a protein-nucleic acid binding residue recognition device provided in an embodiment of the present application;

[0054] Fig.10 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0056] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0057] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0058] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0059] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0060] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, the tablet computer 1012 or the mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0061] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0062] It should be noted that the protein-nucleic acid binding residue identification method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the protein-nucleic acid binding residue identification device is generally set in the server / terminal device.

[0063] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0064] Continue to refer Figure 2 , shows a flow chart of an embodiment of a protein-nucleic acid binding residue identification method according to the present application. The protein-nucleic acid binding residue identification method comprises: step S201, step S202, step S203, step S204 and step S205.

[0065] In step S201, the amino acid sequence and three-dimensional structure data of the protein to be identified are obtained.

[0066] In the present application example, in the prediction task of protein-nucleic acid binding residues, the input data mainly includes the amino acid sequence of the protein and its three-dimensional structure information. The amino acid sequence of the protein is described by its primary structure, usually represented as a string sequence, in which each character represents a specific amino acid. Specifically, a protein sequence S = {a1, a2, ..., a L} is used to describe the order of amino acids in a protein, where L represents the length of the sequence, and a i represents the i-th amino acid.

[0067] In the embodiments of the present application, the three-dimensional structure of the protein is described in the form of a point cloud, and the three-dimensional coordinates of each atom are defined as (xi ,y i ,z i ), where i represents the number of the atom. The three-dimensional structural information of proteins is crucial for understanding protein functions and identifying their binding to other molecules. This representation can effectively capture the spatial relationship between protein residues.

[0068] In step S202, the amino acid sequence is input into a trained protein language model to perform a sequence feature extraction operation to obtain sequence feature data.

[0069] In the embodiment of the present application, in the process of protein sequence feature extraction, the present application uses a pre-trained protein language model ESM2 (Evolutionary Scale Modeling). The ESM2 model is based on the Transformer architecture and captures the context-dependent features in the protein sequence through its multi-layer self-attention mechanism. This mechanism enables the model to effectively extract deep information from the sequence, especially long-distance dependencies. Given a protein sequence S = {a1, a2, ..., a L}, the feature vector generated by the model is expressed as:

[0070] F seq ={f1,f2,...,f L}

[0071] in, represents the feature vector of the ith amino acid, and d is the dimension of the feature vector. Through this feature extraction method, the model can capture rich information in the sequence and provide rich information for downstream binding site identification. In this way, the model can generate feature representations with high generalization capabilities for protein functional region identification, which is particularly suitable for detecting nucleic acid binding sites.

[0072] In step S203, the three-dimensional structure data is input into the sparse convolutional network in the form of point cloud to perform a structure feature extraction operation to obtain structure feature data.

[0073] In the embodiment of the present application, in order to extract important features from the three-dimensional structure of the protein, the present application uses a sparse convolutional network (Sparse ConvNet) to process the point cloud data of the protein. Each atom in the point cloud is represented by a three-dimensional coordinate (x i ,y i ,z i ) representation, which is used to describe the three-dimensional configuration of proteins. Through this representation, the model can capture fine geometric information in the local area of ​​the protein.

[0074] The processing of point cloud data can generate the following structural feature representations:

[0075] F struct ={g1,g2,...,g N}

[0076] in, Represents the characteristics of the i-th atom, and N is the number of atoms in the protein. Sparse convolutional networks can efficiently capture the geometric characteristics of protein molecules by performing local convolution operations on point cloud data, especially playing an important role in understanding the binding process of proteins with other molecules.

[0077] In step S204, a weighted fusion operation is performed on the sequence feature data and the structure feature data to obtain a weighted fusion feature.

[0078] In the embodiment of the present application, after extracting the sequence feature F seq and structural features F struct After that, this application proposes a cross-modal feature fusion strategy. By weighted fusion of the features of these two different modalities, the model can effectively combine the information in the sequence and three-dimensional structure to generate a comprehensive feature representation:

[0079] F fusion =α·F seq +β·F struct

[0080] Among them, α and β are learnable weight parameters, which respectively control the contribution ratio of sequence features and structural features to the final prediction. Through this weighted fusion method, the model can dynamically adjust the balance between sequence and structural features according to the needs of different tasks, making the prediction results more accurate.

[0081] In the embodiments of the present application, cross-modal feature fusion not only improves the model's efficiency in utilizing sequence and structural information, but also captures the interactions between these features, making the model more robust when dealing with complex protein-nucleic acid interactions.

[0082] In step S205, the weighted fusion features are input into the trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

[0083] In the embodiment of the present application, the fused feature F fusion Input into the multi-layer perceptron (MLP) for the prediction of binding residues. Through multiple nonlinear transformations, MLP can make a probability prediction of whether each residue is a nucleic acid binding site. The specific prediction formula is:

[0084] p i =σ(W·F fusion +b)

[0085] Where W is the weight matrix, b is the bias term, and σ is the Sigmoid activation function, which is used to normalize the output value to the interval [0,1][0,1][0,1]. The output result p i Represents the probability that residue i is a binding site. This nonlinear transformation enables the model to effectively utilize the information obtained from feature fusion, thereby achieving better prediction results in complex biological data.

[0086] In the embodiments of this application, see Figure 3 The technical flowchart shown first extracts sequence features and structural features by inputting the amino acid sequence and three-dimensional structure data of the protein. Specifically, the amino acid sequence is processed by a pre-trained protein language model (such as ESM2) to generate context-dependent features for each amino acid; at the same time, the three-dimensional structure of the protein is input into a sparse convolutional network in the form of a point cloud to extract three-dimensional geometric features. The extracted sequence features and structural features are weighted fused through a cross-modal feature fusion module to form a unified feature representation. The fused feature vector is input into a multi-layer perceptron (MLP) classifier to predict the probability of each residue being a binding site.

[0087] In an embodiment of the present application, a method for identifying protein-nucleic acid binding residues is provided, including: obtaining the amino acid sequence and three-dimensional structure data of the protein to be identified; inputting the amino acid sequence into a trained protein language model to perform sequence feature extraction operations to obtain sequence feature data; inputting the three-dimensional structure data into a sparse convolutional network in the form of a point cloud to perform structure feature extraction operations to obtain structure feature data; performing weighted fusion operations on the sequence feature data and the structure feature data to obtain weighted fusion features; inputting the weighted fusion features into a trained multi-layer perceptron classifier for recognition operations to obtain probability information of each residue as a binding site. Compared with the prior art, the present application significantly improves the prediction accuracy by introducing a cross-modal learning framework by combining the sequence and three-dimensional structure information of the protein; in addition, the present application introduces an atomic-level attention mechanism, which enables the model to dynamically capture the geometric changes of the protein, thereby improving the adaptability and robustness in complex scenarios; the present application also enhances the generalization ability of the model on small-scale data sets through a self-supervised learning strategy, and effectively solves the challenges brought by data imbalance. In summary, this application achieves an overall improvement in the performance of protein-nucleic acid binding residue identification through multimodal fusion, self-supervised learning, and dynamic structure adjustment technology.

[0088] In some optional implementations of the embodiments of the present application, after the step of inputting the three-dimensional structure data in the form of a point cloud into a sparse convolutional network to perform a structure feature extraction operation to obtain the structure feature data, the following steps are also included:

[0089] The structural feature data is adaptively adjusted according to the attention mechanism, where the attention mechanism is expressed as:

[0090]

[0091] Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

[0092] In the embodiments of the present application, in order to further improve the prediction accuracy of the model, the present application introduces an atomic-level attention mechanism (AWA). The AWA mechanism improves the recognition ability of binding sites by dynamically focusing on the interactions between different atoms in protein molecules. This mechanism can adaptively adjust the weight of each atom in the model according to its position and characteristics.

[0093] The AWA attention calculation formula is as follows:

[0094]

[0095] Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor. Through this attention mechanism, the model can focus on the local structural changes at the binding site, thereby improving the accuracy and stability of residue prediction.

[0096] In some optional implementations of the embodiments of the present application, before the step of inputting the weighted fusion features into the trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site, the following steps are also included:

[0097] The initial multi-layer perceptron classifier is called, and a model training operation is performed on the initial multi-layer perceptron classifier based on a self-supervised learning method to obtain a trained multi-layer perceptron classifier.

[0098] In some optional implementations of the embodiments of the present application, the loss function of the model training operation is expressed as:

[0099]

[0100] Among them, z i and z j represents the protein feature vector after different data enhancements, and τ is the temperature parameter.

[0101] In the embodiments of the present application, in order to enhance the generalization ability of the model, the present application introduces a self-supervised learning (SSL) strategy. Self-supervised learning can improve the adaptability of the model to small samples and data imbalance by training on unlabeled data. Through random data enhancement (such as rotation, translation, scaling, etc.), the model generates protein samples from different perspectives and uses contrastive learning to maintain feature consistency.

[0102] The loss function of contrastive learning is defined as:

[0103]

[0104] Among them, z i and z j represents the protein feature vector after different data enhancements, and τ is the temperature parameter. By maximizing the similarity between similar samples, the model can effectively learn important protein structural features in a small sample environment and improve its adaptability to new samples.

[0105] In practical applications, through the cross-modal learning framework, combining the sequence characteristics and three-dimensional structural characteristics of proteins, significant breakthroughs have been made in the field of protein-nucleic acid binding residue recognition. Figure 4 As shown, the present application has significant advantages on the DNA 129 dataset. The F1-score of CrossBind on this dataset reached 0.602, while the F1-score of the existing best method GraphBind was only 0.512, an improvement of 17.6%. In addition, CrossBind's performance on the Matthews correlation coefficient (MCC) was also significantly improved, from 0.484 of GraphBind to 0.581, an increase of 20%. This series of significant performance improvements is mainly due to the multimodal feature fusion strategy adopted by this application. By integrating protein sequence information and three-dimensional structural features, the model can comprehensively capture the key areas of protein-nucleic acid interactions. AUPR (area under the precision-recall curve) has also been significantly improved, further reflecting the model's powerful processing capabilities in complex residue prediction tasks.

[0106] like Figure 5As shown in the ablation experiment on the DNA 129 dataset, the importance of each module of the present application in the protein-nucleic acid binding residue identification task is verified. The effect of using the protein language model alone on this task is not ideal, and using only the atomic point cloud segmentation encoder leads to worse results. However, when combining sequence and structural features, the AUPRC results are improved by 5.7% and the MCC is improved by 7.8%, which is significantly better than the existing state-of-the-art methods. These results fully demonstrate the effectiveness of the cross-modal module proposed in this application in feature fusion. The AWA module plays a key role in combining the local geometric knowledge of atoms to amino acids, significantly improving the performance of the model. When the AWA module is removed, the prediction results drop significantly, further verifying its importance in accurately locating the binding site. In addition, the introduction of the SSL module improves the robustness of the segmentation encoder when processing unbalanced data. Removing this module leads to a 2.2% reduction in AUPRC, indicating the necessity of self-supervised learning in strengthening data processing capabilities. The RPF module optimizes cross-modal prediction results using biological prior knowledge. Introducing traditional co-evolution features did not improve the performance, but rather decreased it, because the pre-trained language model already covers most of the important feature information. Overall, the ablation study results are as follows Figure 6 As shown, the irreplaceability and technical effectiveness of each module in the CrossBind method of this application are further confirmed.

[0107] The study also showed that using a smaller LLM (language model) structure will significantly affect task performance. For example, reducing the 33-layer Transformer to 12 layers will cause the AUC to drop by 2.4%. In contrast, the fine-tuned 33-layer LLM model improved AUPRC by 3.1%, proving that fine-tuning a larger LLM model helps better capture the language information in the protein sequence and is beneficial to the accuracy of residue prediction.

[0108] In the design of the atomic attention mechanism, this application adopts an attention mechanism based on MLP (multi-layer perceptron), which integrates the fine-grained local geometric knowledge between atoms and amino acids. Figure 7 As shown in Figure 2, simply averaging or stacking the atomic features leads to a significant performance degradation because valuable local geometric knowledge is lost. Instead, replacing the atomic attention with a single self-attention layer on the cross-modal module, although the performance is degraded, it is still better than simply stacking all atomic features. This suggests that a single layer of self-attention may over-smooth the local geometric relationship between atoms, resulting in a poorer feature capture effect than a fine MLP mechanism.

[0109] In addition, the present application also shows the qualitative results of the present application in the prediction of binding residues, such as Figure 8As shown. This application visualizes three cases predicted by GraphSite, GraphBind and the model proposed in this application on DNA 129. This application selects the example protein 6YMW_B. This protein contains 668 residues, 13 of which are binding residues. Figure 8 As shown, the model CrossBind proposed in this application predicts 10 true binding residues and 4 false positive residues, with a Rec of 0.64, a Pre of 0.53, and an F1 score of 0.58. In contrast, GraphSite predicts 8 true binding residues and 13 false positive residues, with a Rec of 0.60, a Pre of 0.47, and an F1 score of 0.52. On the other hand, GraphBind only predicts 6 true binding residues and 5 false positive residues, with a Rec of 0.40, a Pre of 0.26, and an F1 score of 0.32. Although GraphSite also predicts enough true binding residues, it has a higher false positive rate, while CrossBind shows a higher accuracy.

[0110] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0111] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0112] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0113] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0114] Further references Fig. 9 , as a response to the above Figure 2 In order to realize the method shown in the figure, the present application provides an embodiment of a protein-nucleic acid binding residue recognition device, and the device embodiment is Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0115] like Fig. 9 As shown, the protein-nucleic acid binding residue identification device 200 of the embodiment of the present application includes:

[0116] A data acquisition module 210 is used to acquire the amino acid sequence and three-dimensional structure data of the protein to be identified;

[0117] A sequence feature extraction module 220 is used to input the amino acid sequence into a trained protein language model to perform a sequence feature extraction operation to obtain sequence feature data;

[0118] The structural feature extraction module 230 is used to input the three-dimensional structural data in the form of point cloud into the sparse convolutional network to perform structural feature extraction operations to obtain structural feature data;

[0119] The weighted fusion module 240 is used to perform a weighted fusion operation on the sequence feature data and the structure feature data to obtain a weighted fusion feature;

[0120] The recognition module 250 is used to input the weighted fusion features into the trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

[0121] In an embodiment of the present application, a protein-nucleic acid binding residue identification device 200 is provided, including: a data acquisition module 210, used to obtain the amino acid sequence and three-dimensional structure data of the protein to be identified; a sequence feature extraction module 220, used to input the amino acid sequence into a trained protein language model to perform a sequence feature extraction operation to obtain sequence feature data; a structural feature extraction module 230, used to input the three-dimensional structure data in the form of a point cloud into a sparse convolutional network to perform a structural feature extraction operation to obtain structural feature data; a weighted fusion module 240, used to perform a weighted fusion operation on the sequence feature data and the structural feature data to obtain a weighted fusion feature; and an identification module 250, used to input the weighted fusion feature into a trained multi-layer perceptron classifier to perform a recognition operation to obtain probability information of each residue being a binding site. Compared with the existing technology, this application significantly improves the prediction accuracy by introducing a cross-modal learning framework by combining the sequence and three-dimensional structure information of proteins. In addition, this application introduces an atomic-level attention mechanism, which enables the model to dynamically capture the geometric changes of proteins, thereby improving adaptability and robustness in complex scenarios. This application also enhances the generalization ability of the model on small-scale data sets through a self-supervised learning strategy, and effectively solves the challenges brought by data imbalance. In short, this application achieves an overall performance improvement in protein-nucleic acid binding residue recognition through multimodal fusion, self-supervised learning, and dynamic structure adjustment technology.

[0122] In some optional implementations of the embodiments of the present application, the above-mentioned protein-nucleic acid binding residue identification device 200 further includes:

[0123] The adaptive adjustment module is used to perform adaptive adjustment operations on the structural feature data according to the attention mechanism, where the attention mechanism is expressed as:

[0124]

[0125] Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

[0126] To solve the above technical problems, the present application also provides a computer device. Fig.10 , Fig.10 This is a basic structural block diagram of a computer device according to an embodiment of the present application.

[0127] The computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 300 having components 310-330, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0128] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0129] The memory 310 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 310 can be an internal storage unit of the computer device 300, such as a hard disk or memory of the computer device 300. In other embodiments, the memory 310 can also be an external storage device of the computer device 300, such as a plug-in hard disk equipped on the computer device 300, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 310 can also include both the internal storage unit of the computer device 300 and its external storage device. In the embodiment of the present application, the memory 310 is generally used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions of the protein-nucleic acid binding residue identification method, etc. In addition, the memory 310 can also be used to temporarily store various data that have been output or are to be output.

[0130] The processor 320 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 320 is generally used to control the overall operation of the computer device 300. In the embodiment of the present application, the processor 320 is used to run the computer-readable instructions or process data stored in the memory 310, such as computer-readable instructions for running the protein-nucleic acid binding residue identification method.

[0131] The network interface 330 may include a wireless network interface or a wired network interface. The network interface 330 is generally used to establish a communication connection between the computer device 300 and other electronic devices.

[0132] The computer device provided in this application significantly improves the prediction accuracy by introducing a cross-modal learning framework by combining the sequence and three-dimensional structure information of proteins. In addition, an atomic-level attention mechanism is introduced to enable the model to dynamically capture the geometric changes of proteins, thereby improving adaptability and robustness in complex scenarios. The self-supervised learning strategy also enhances the generalization ability of the model on small-scale data sets and effectively solves the challenges brought by data imbalance. In short, through multimodal fusion, self-supervised learning and dynamic structure adjustment technology, the overall performance of protein-nucleic acid binding residue recognition is improved.

[0133] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the protein-nucleic acid binding residue identification method as described above.

[0134] The computer-readable storage medium provided by this application significantly improves the prediction accuracy by introducing a cross-modal learning framework by combining the sequence and three-dimensional structure information of the protein; in addition, an atomic-level attention mechanism is introduced to enable the model to dynamically capture the geometric changes of the protein, thereby improving the adaptability and robustness in complex scenarios; and through a self-supervised learning strategy, the generalization ability of the model on small-scale data sets is enhanced, and the challenges brought by data imbalance are effectively solved. In short, through multimodal fusion, self-supervised learning and dynamic structure adjustment technology, the overall performance of protein-nucleic acid binding residue recognition is improved.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0136] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for identifying protein-nucleic acid binding residues, characterized in that: The steps include: Obtain the amino acid sequence and three-dimensional structure data of the protein to be identified; Inputting the amino acid sequence into a trained protein language model to perform sequence feature extraction to obtain sequence feature data; Inputting the three-dimensional structure data in the form of point cloud into a sparse convolutional network to perform a structure feature extraction operation to obtain structure feature data; Performing a weighted fusion operation on the sequence feature data and the structural feature data to obtain a weighted fusion feature; The weighted fusion features are input into a trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

2. The method for identifying protein-nucleic acid binding residues according to claim 1, characterized in that: After the step of inputting the three-dimensional structure data in the form of point cloud into a sparse convolutional network to perform a structure feature extraction operation to obtain the structure feature data, the following steps are also included: The structural feature data is adaptively adjusted according to the attention mechanism, wherein the attention mechanism is expressed as: Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

3. The method for identifying protein-nucleic acid binding residues according to claim 1, characterized in that: Before the step of inputting the weighted fusion features into the trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site, the following steps are also included: An initial multi-layer perceptron classifier is called, and a model training operation is performed on the initial multi-layer perceptron classifier based on a self-supervised learning method to obtain the trained multi-layer perceptron classifier.

4. The method for identifying protein-nucleic acid binding residues according to claim 3, characterized in that: The loss function of the model training operation is expressed as: Among them, z i and z j represents the protein feature vector after different data enhancements, and τ is the temperature parameter.

5. The method for identifying protein-nucleic acid binding residues according to claim 1, characterized in that: The weighted fusion feature is expressed as: F fusion =α·F seq +β·F struct Among them, α and β are learnable weight parameters, which control the contribution ratio of sequence features and structural features to the final prediction respectively. seq represents the sequence feature data, F struct Represents the structural feature data.

6. The method for identifying protein-nucleic acid binding residues according to claim 1, characterized in that: The probability information is expressed as: p i =σ(W·F fusion +b) Among them, W is the weight matrix, b is the bias term, and σ is the Sigmoid activation function, which is used to normalize the output value to the interval [0, 1][0, 1][0, 1].

7. A protein-nucleic acid binding residue recognition device, characterized in that: include: A data acquisition module is used to obtain the amino acid sequence and three-dimensional structure data of the protein to be identified; A sequence feature extraction module, used for inputting the amino acid sequence into a trained protein language model to perform a sequence feature extraction operation to obtain sequence feature data; A structural feature extraction module, used for inputting the three-dimensional structural data in the form of point cloud into a sparse convolutional network to perform a structural feature extraction operation to obtain structural feature data; A weighted fusion module, used for performing a weighted fusion operation on the sequence feature data and the structure feature data to obtain a weighted fusion feature; The recognition module is used to input the weighted fusion features into a trained multi-layer perceptron classifier for recognition operation to obtain the probability information of each residue being a binding site.

8. The protein-nucleic acid binding residue identification device according to claim 7, characterized in that: The device also includes: The adaptive adjustment module is used to perform an adaptive adjustment operation on the structural feature data according to an attention mechanism, wherein the attention mechanism is expressed as: Among them, q i and k j denote the query vector and bond vector of atom i and neighboring atom j respectively, d k is the scaling factor.

9. A computer device comprising a memory and a processor, characterized in that: The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the protein-nucleic acid binding residue identification method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the protein-nucleic acid binding residue identification method according to any one of claims 1 to 6 are implemented.

Images