A method, apparatus and device for identifying a protein-metal ion binding site

Abstract

The application provides a protein-metal ion binding site recognition method, device and equipment, and belongs to the field of protein detection. The method comprises the following steps: extracting features of known metal ion binding proteins to obtain a plurality of sample evolutionary information features; for an unknown protein sequence, candidate distant homologous metal ion binding proteins are determined through multiple sequence comparison and cosine similarity screening, and a training set is constructed; a composite framework of a bidirectional long short-term memory network and a full connection neural network is adopted, input features include evolutionary information features and physicochemical property features, and output is probability values of different metal ion binding sites; the composite framework is trained through the training set to obtain a site prediction model; and the prediction model is used to predict the binding sites of the unknown protein sequence. Stable and efficient prediction of metal ion binding sites is realized.

Description

Technical Field

[0001] This invention belongs to the field of protein detection, specifically relating to a method, apparatus, and device for identifying the binding sites of proteins and metal ions. Background Technology

[0002] Proteins are the primary executors of normal biological activities, and metal ions play a crucial role in the formation of stable protein structures and the expression of their functions. For example, zinc fingers are a common structural domain; proteins bind to zinc ions to form stable folded conformations, maintaining the stability of their spatial structure. Some enzymes, such as carbonic anhydrase and superoxide dismutase, require copper and zinc ions as cofactors to complete their catalytic reactions. These metal ions bind to specific amino acids in proteins, forming active sites that participate in substrate binding and conversion. Calcium ions play a vital role in signal transduction in living organisms. Calcium ions bind to proteins, such as calmodulin, which can regulate various cellular processes, such as muscle contraction, nerve transmission, and gene expression. Iron ions are important components of hemoglobin and cytochromes, participating in oxygen transport and electron transport.

[0003] Studying the binding interactions between proteins and metal ions, particularly the accurate identification of protein-metal ion binding sites, can reveal the structure, function, binding forms, and evolutionary pathways of these protein-metal ion complexes. Theoretically, this research helps to elucidate the functional mechanisms of proteins, the laws governing biomolecular interactions, and the mysteries of the origin and evolution of life. Practically speaking, this research provides crucial scientific evidence and technological support for disease treatment and drug development.

[0004] Existing biochemical experimental methods for identifying protein-metal ion binding sites include X-ray crystallography, nuclear magnetic resonance spectroscopy, mass spectrometry, and isothermal titration calorimetry. These methods suffer from drawbacks such as cumbersome procedures, complex techniques, and poor stability of identification results, leading to difficulties in ensuring reproducibility. Summary of the Invention

[0005] To address the problems of complexity and poor stability in existing detection methods, this invention provides a method, apparatus, and device for identifying protein-metal ion binding sites.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for identifying protein-metal ion binding sites, the method comprising:

[0008] Feature extraction was performed on metal ion-binding proteins with known structures and functions to obtain evolutionary information features of multiple samples.

[0009] For unknown protein sequences, a set of homologous proteins is screened through multiple sequence alignment, and the evolutionary information features of the homologous proteins are extracted as target feature vectors. The similarity value between the target feature vector and the evolutionary information features of multiple samples is determined by cosine similarity, and a predetermined number of candidate distant homologous metal ion-binding proteins are screened based on the similarity value. A training set is constructed based on the candidate distant homologous metal ion-binding proteins.

[0010] A composite framework combining a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN) is employed. The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using the training set to obtain a site prediction model.

[0011] The prediction model is used to predict the metal ion binding sites of unknown protein sequences.

[0012] Optionally, the evolutionary information features include the mean and standard deviation vectors of the position-specific scoring matrix and the weighted observation matrix, and the feature extraction formula is:

[0013]

[0014] Among them, M PSSM and M WOP These are the location-specific scoring matrix and the weighted observation matrix, respectively, P i,X and W i,X The scores for replacing amino acid i with amino acid X in the position-specific scoring matrix and the weighted observation matrix, respectively, are given by i∈{1,2,3,…,l}, where l represents the protein sequence length and X∈{A,R,…,V} represents the set of 20 amino acids.

[0015] Analyze the 20 amino acids of each metal ion-bound protein at M PSSM and M WOP The average value and standard deviation of the corresponding positions are used to obtain evolutionary information features.

[0016] Optionally, the formula for calculating the similarity value is:

[0017]

[0018]

[0019] Where sim represents the similarity value, a i and b i Let V(A) and V(B) represent the values ​​at the i-th position. This represents the average vector of the position-specific scoring matrices for 20 amino acids. This represents the standard deviation vector of the position-specific scoring matrix for 20 amino acids. This represents the average vector of the weighted observation matrix of 20 amino acids. Let A represent the standard deviation vector of the weighted observation matrix of 20 amino acids, where A represents any protein in the homologous protein set, B represents any protein in the metal ion-binding protein set, and α, β, γ, and δ are weighting coefficients.

[0020] Optionally, constructing a training set based on candidate distantly homologous metal ion-binding proteins includes:

[0021] Using candidate distantly homologous metal ion-binding proteins as real samples, virtual samples are generated by data augmentation of the real samples through a generative adversarial network (GAN).

[0022] The training set is formed by mixing virtual samples with real samples.

[0023] Optionally, the training set includes samples of metal ion-protein binding and binding site labels. The complex framework is trained using this training set to obtain a site prediction model, including:

[0024] The binding sample is input into the composite frame to obtain the predicted binding site;

[0025] The composite framework was trained to obtain a site prediction model with the goal of minimizing the deviation between the predicted binding site and the binding site label.

[0026] A device for recognizing protein-metal ion binding sites, the device comprising:

[0027] The extraction module is used to extract features from metal ion-binding proteins with known structures and functions, and obtain evolutionary information features from multiple samples.

[0028] The screening module is used to screen a set of homologous proteins for unknown protein sequences through multiple sequence alignment and extract the evolutionary information features of homologous proteins as target feature vectors; it determines the similarity value between the target feature vector and the evolutionary information features of multiple samples through cosine similarity, and screens a preset number of candidate distant homologous metal ion-binding proteins based on the similarity value; and it constructs a training set based on the candidate distant homologous metal ion-binding proteins.

[0029] A construction module is used to construct a composite framework that combines a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN). The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using the training set to obtain a site prediction model.

[0030] The prediction module is used to predict the metal ion binding sites of the target protein using the prediction model.

[0031] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for identifying protein-metal ion binding sites.

[0032] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned method for recognizing protein-metal ion binding sites.

[0033] The method for recognizing protein-metal ion binding sites provided by this invention has the following beneficial effects:

[0034] First, by extracting features from known metal-binding proteins, evolutionary information features of the samples are obtained, avoiding the tedious process of manually designing features. For unknown protein sequences, homologous proteins are screened through multiple sequence alignment, and target feature vectors are extracted. Cosine similarity is then used to quickly screen candidate distantly homologous metal-binding proteins, significantly improving the screening efficiency. A training set is constructed based on the similarity values ​​used to screen candidate distantly homologous metal-binding proteins, ensuring data quality consistency. A composite framework of Bi-LSTM and FCNN is adopted, which can simultaneously capture long-term dependencies and local physicochemical features of sequences. The prediction model obtained after training the model on the training set can directly output the probability value of metal-binding sites, realizing an end-to-end prediction process. This method transforms the complex process that relies on expert experience into a standardized data-driven process, which not only reduces the randomness caused by human intervention, but also improves the model's ability to identify low-homology sequences through multi-dimensional feature fusion. While ensuring prediction accuracy, it significantly improves the repeatability and cross-platform applicability of experiments, ultimately achieving stable and efficient metal-binding site prediction. Attached Figure Description

[0035] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is an overall architecture diagram of a method for identifying protein-metal ion binding sites according to an exemplary embodiment of the present invention.

[0037] Figure 2This is a flowchart illustrating a method for identifying protein-metal ion binding sites according to an exemplary embodiment of the present invention.

[0038] Figure 3 This is a diagram illustrating a deep learning framework architecture according to an exemplary embodiment of the present invention.

[0039] Figure 4 This is a comparison chart of the number of homologous proteins searched by different methods according to an exemplary embodiment of the present invention.

[0040] Figure 5 This invention provides a different method for predicting sensitivity values ​​according to an exemplary embodiment.

[0041] Figure 6 This is a block diagram of a protein-metal ion binding site recognition device provided by the present invention according to an exemplary embodiment. Detailed Implementation

[0042] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0043] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0044] This invention relates to a method for recognizing protein-metal ion binding sites based on distantly homologous proteins and generative adversarial networks. For example... Figure 1 The diagram illustrates the overall architecture of the method involved in this invention. For unknown proteins, firstly, sequence alignment is performed based on evolutionary information. This sequence alignment information is then used to search for distantly related homologous metal-binding proteins with known structures and functions in BioLiP2, a database storing proteins with currently known structures and ligand annotations. Secondly, for the obtained distantly related homologous metal-binding proteins, high-similarity data is generated using Generative Adversarial Networks (GANs). This high-similarity data is mixed with real distantly related homologous proteins to form a training set. Finally, a deep learning framework for metal-binding sites is constructed on these training sets and used to predict the probability values ​​of different types of metal-binding sites for unknown proteins.

[0045] This invention provides a method for identifying protein-metal ion binding sites, specifically as follows: Figure 2 As shown, it includes the following steps:

[0046] S201. Feature extraction is performed on metal ion-binding proteins with known structures and functions to obtain evolutionary information features of multiple samples.

[0047] The evolutionary information features include the mean and standard deviation vectors of the position-specific scoring matrix and the weighted observation matrix. In this step, it is first necessary to construct a sample pool of metal ion-binding proteins and a feature space for that pool.

[0048] In one embodiment, the BioLiP2 database stores proteins with currently known structures and ligand annotations. The BioLiP2 database is synchronized with the PDB database and, as of February 2025, it stored approximately 943,000 accurately annotated protein entries, of which approximately 205,000 are metal ion ligand-binding proteins. Of these metal ion ligand-binding proteins, approximately 32.4% are zinc ion-binding proteins, 25.6% are calcium ion-binding proteins, 23.2% are magnesium ion-binding proteins, 7.8% are manganese ion-binding proteins, and 4.2% are iron and ferrous ion-binding proteins, totaling 93.2% of all metal ion-binding proteins. The remaining approximately 6.8% are other types of metal ion-binding proteins. Therefore, the metal ion-binding proteins involved in this invention include zinc ion, calcium ion, magnesium ion, manganese ion, and iron and ferrous ion-binding proteins.

[0049] For these metal ion-binding proteins, the position-specific scoring matrix M was calculated using the MMseqs2 program. PSSM The weighted observation matrix M is calculated using the PSI-BLAST program. WOP To ensure the generated data contained the richest possible protein evolution information, the Uniclust30 protein database was used during the calculations. PSSM and M WOP Both are matrices of size l×20, and their expressions are as follows:

[0050]

[0051] P i,X and W i,X Let $X$ represent the score of replacing amino acid $i$ with amino acid $X$ in the position-specific scoring matrix and the weighted observation matrix, respectively, where $i$ ∈ {1, 2, 3, ..., l}, $l$ is a positive integer representing the protein sequence length, and $X$ ∈ {A, R, ..., V}. Here, ${A, R, ..., V} represents the set of 20 amino acids.

[0052] To quantify the evolutionary levels of different amino acids in proteins, this invention statistically analyzes the evolution of each of the 20 amino acids (i.e., {A, R, ..., V}) in a metal ion-bound protein at the M... PSSM and MWOP The mean and standard deviation of the corresponding position (i.e., the score when the amino acid is replaced with amino acid X).

[0053] Finally, the evolutionary information feature representation of each protein includes four parts, namely: (The average vector of the position-specific scoring matrix of 20 amino acids, 20×20=400 dimensions) (Standard deviation vector of the position-specific scoring matrix of 20 amino acids, 20×20=400 dimensions) (The average vector of the weighted observation matrix of 20 amino acids, 20×20=400 dimensions) (The standard deviation vector of the weighted observation matrix of 20 amino acids, 20×20=400 dimensions), and the standard deviation vector of the 20 amino acids in each metal ion-bound protein is statistically analyzed at M. PSSM and M WOP The mean and standard deviation of the corresponding positions are used to obtain the evolutionary information features of the sample.

[0054] S202. For unknown protein sequences, candidate distantly related homologous metal ion binding proteins are identified through multiple sequence comparison and cosine similarity screening, and then a training set is constructed.

[0055] Specifically, for unknown protein sequences, a set of homologous proteins is screened through multiple sequence alignment, and the evolutionary information features of the homologous proteins are extracted as target feature vectors. The similarity values ​​between the target feature vectors and the evolutionary information features of multiple samples are determined by cosine similarity, and a preset number of candidate distantly homologous metal ion-binding proteins are screened based on the similarity values. A training set is constructed based on the candidate distantly homologous metal ion-binding proteins.

[0056] In this step, the first step is to search for distantly related homologous metal ion-binding proteins.

[0057] In one embodiment, for unknown proteins, the HH-suite3 program is used to search for multiple sequence alignments in the Uniclust30 database. To achieve better search results, two search iterations are performed. The proteins with the found multiple sequence alignments are referred to as the homologous protein set. Next, this invention uses the same method to quantify the evolutionary levels of different amino acid types within the proteins.

[0058] Specifically, the position-specific scoring matrix M is calculated using the MMseqs2 program on a set of homologous proteins. PSSM The weighted observation score matrix M was calculated using the PSI-BLAST program. WOP The calculation formula is consistent with that in step 1; and based on this, the calculation is performed. (400 dimensions) (400 dimensions) (400 dimensions) (400 dimensions), consistent with step 1.

[0059] Based on the searched set of homologous proteins and their corresponding evolutionary information features, this invention calculates the similarity between each protein and each protein in the metal ion-binding protein sample pool, using cosine similarity for the calculation:

[0060]

[0061] Among them, a i and b i Let V(A) and V(B) represent the values ​​at the i-th position. Let V(A) and V(B) be the inner product, representing the sum of the products of the components of the two vectors in each dimension. A represents any protein in the set of homologous proteins, and B represents any protein in the set of metal-binding proteins. Therefore, the similarity sim between two proteins is represented as follows:

[0062]

[0063] In this study, sim represents the similarity value, and α, β, γ, and δ are weighting coefficients used to quantify the contributions of the four vectors in finding distantly homologous metal-binding proteins. The four weighting coefficients are set to α = 0.527, β = 0.115, γ = 0.303, and δ = 0.055. Finally, the similarity values ​​of each protein in the metal-binding protein sample pool are sorted; for proteins with multiple similarity values, the maximum value is extracted. This study extracts the top 50 metal-binding proteins from the sorted pool as a high-value set of distantly homologous metal-binding proteins.

[0064] Secondly, generative adversarial networks (GANs) are introduced to create highly similar generated data. For example, candidate distantly related metal ion-binding proteins can be used as real samples, and GANs can be used to augment these real samples to generate virtual samples; the virtual samples are then mixed with real samples to form a training set.

[0065] In another embodiment, after obtaining the top 50 distantly related metal-binding proteins, extracting and ranking too many proteins may introduce proteins with low distant homology similarity, thus affecting the accuracy of the training dataset. Therefore, this invention selects only high-value distantly related metal-binding proteins and uses them as a basis to generate new data with high similarity to the real data.

[0066] Specifically:

[0067] Generative Adversarial Networks (GANs) are introduced, consisting of two subnetworks: a generator and a discriminator. These two subnetworks compete with each other, continuously improving their abilities to ultimately generate data that is as realistic as possible. The learning process of a GAN is similar to a game between the generator and the discriminator: the generator attempts to make the discriminator unable to distinguish between real and fake data, while the discriminator strives to correctly differentiate between real and generated data. The goal of a GAN is to make the data generated by the generator increasingly resemble the distribution of real data, ultimately achieving an effect where the generated data is almost indistinguishable from real data.

[0068] In this invention, the generator's input data This represents all residues on 50 distantly homologous metal-binding proteins, with a total number of n residues, each consisting of a 40-dimensional vector, including M. PSSM and M WOP The score for each residue in the generator network is calculated based on its corresponding replacement residue. To reduce the impact of scores deviating from normal values ​​on the generator and discriminator networks, M is evaluated separately. PSSM and M WOP Perform a normalization operation. Then, set the set of residues generated by the generative adversarial network as the output set. The number of residues generated is 2n.

[0069] Thus, the hybrid dataset is obtained. The number is 3n. Mixed dataset. This will serve as the training set for building the model in the next step.

[0070] S203. A composite framework consisting of a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN) is adopted. The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using this training set to obtain the site prediction model.

[0071] This invention employs, as follows Figure 3 The prediction model architecture is shown. The training dataset for the specific metal ion binding site prediction model is... The dataset includes original real samples (distantly homologous metal-binding proteins) and generated virtual samples (highly similar samples generated using a generative adversarial network). The evolutionary information feature dimension for each residue of each protein in the dataset is 45 dimensions, i.e., M. PSSM and M WOP The score at this location (40 dimensions); local microenvironment characteristics, namely 5 physicochemical properties (5 dimensions), namely hydrophilicity, hydrophobicity, polarity, positive charge, and negative charge. These 5 physicochemical properties are the standard properties of amino acids.

[0072] The deep neural network learning framework used in this invention comprises two parts: a Bi-LSTM module and an FCNN module, as follows: Figure 3 As shown. The input module includes the Bi-LSTM module, which uses a Bidirectional Long Short-Term Memory (Bi-LSTM) network to learn and train prediction models for different types of metal ion binding sites. The FCNN module includes a set of fully connected neural networks (FCNN), primarily used to output the predicted probabilities for different types of metal ion binding sites. After the deep neural network learning composite framework is built, it needs to be trained. The binding sample is input into the composite framework to obtain the predicted binding site. The composite framework is trained with the goal of minimizing the deviation between the predicted binding site and the binding site label to obtain the site prediction model.

[0073] In one embodiment, Bi-LSTM is a recurrent neural network capable of retaining information previously processed by deep neural networks, making it suitable for processing sequence data with dependencies. This characteristic aligns with the prediction of functional sites in protein sequences. Bi-LSTM consists of two independent LSTM networks. These two networks operate independently, from the beginning to the end of the protein sequence (forward LSTM) and from the end to the beginning of the protein sequence (backward LSTM), respectively. The outputs of the two LSTMs are combined, allowing the neural network to consider contextual information simultaneously. Furthermore, Bi-LSTM includes units from the LSTM network capable of retaining long-term information, enabling the network to simultaneously mine both short-range and long-range dependencies in the sequence. This is highly effective for identifying metal ion binding sites in proteins that are geographically distant but spatially close or evolutionarily coupled. In this invention, the input feature dimension of the Bi-LSTM module is 45 dimensions, and the output feature dimension is set to 90 dimensions.

[0074] The FCNN module takes 90-dimensional input features and consists of three fully connected layers with 40, 20, and 10 dimensions respectively. Each layer uses ReLU as the activation function. The FCNN module ultimately outputs six probability values, including predicted probabilities for binding sites of zinc, calcium, magnesium, manganese, iron, and ferrous ions. The probability values ​​range from [0,1], with higher values ​​indicating a higher probability that the current location is a binding site for the corresponding metal ion.

[0075] S204. Use this prediction model to predict the metal ion binding sites of unknown protein sequences.

[0076] The unknown protein sequence is input into the prediction model to obtain the predicted binding site.

[0077] To verify the effectiveness of this method, 100 metal ion-binding proteins were randomly selected, and the number of homologous proteins found by different methods was tested. The results are as follows: Figure 3 As shown, when using only sequence alignment-based search methods, an average of approximately 43 homologous metal-binding proteins were found for each metal-binding protein; however, when using a distant homologous protein search method based on evolutionary information, an average of approximately 80 distant homologous metal-binding proteins were found for each metal-binding protein, representing an improvement of 86%. This clearly demonstrates that the present invention can effectively search for potential distant homologous proteins using evolutionary information.

[0078] Figure 4 A comparison of the number of metal-binding proteins found using sequence alignment and evolutionary homology-based search methods, respectively.

[0079] To further verify the effectiveness of the proposed method in identifying metal ion binding sites, this invention selects 100 proteins from different metal ion types, and then constructs different models using homologous metal ion-binding proteins, distantly homologous metal ion-binding proteins, and distantly homologous metal ion-binding proteins, along with generated high similarity data, and compares the prediction results.

[0080] Figure 5 The prediction sensitivity values ​​of three different models at six metal ion binding sites are presented. Sensitivity value is an important indicator for evaluating model performance, primarily measuring the model's ability to correctly identify positive samples (metal ion binding sites in this invention). The sensitivity value ranges from [0,1], with higher values ​​indicating stronger model recognition capabilities for metal ion binding sites. Figure 4 As shown, with the introduction of proteins bound by distantly homologous metal ions and the generation of highly similar data, the model's accuracy in identifying metal ion binding sites is significantly improved. The model's sensitivity values ​​for zinc, calcium, and magnesium ion binding sites all exceed 60%. Furthermore, this invention notes that this improvement is even more pronounced for ferric and ferrous ions, increasing from 25% to 43% (a 72% improvement) and from 22% to 45% (a 105% improvement), respectively. Considering the limited number of ferric and ferrous ion-binding proteins in existing data, using a generative adversarial network to increase the amount of training data greatly enhances the model's ability to correctly identify ferric and ferrous ions.

[0081] This invention further compares the overall prediction performance of models built using different protein data. Here, this invention uses the AUC value as a quantitative indicator. AUC (Area Under the Receiver Operating Characteristic Curve) is a core indicator for evaluating model performance. AUC can quantify the model's performance at different classification thresholds. The higher the AUC value, the better the model can separate positive and negative samples (i.e., the prediction probability of metal ion binding sites is generally higher than that of non-metal ion binding sites).

[0082] As shown in Table 1, the AUC values ​​of the model increased with the introduction of distantly homologous metal ion-binding proteins and the addition of high-similarity generated data. The AUC value of zinc ion binding sites increased from 0.713 to 0.735 and 0.746, respectively. The AUC values ​​of ferric and ferrous ions on homologous metal ion-binding proteins were only about 0.62–0.63, but eventually increased to 0.68–0.69. This indicates that both strategies can significantly improve the model's accurate identification of metal ion binding sites. This also demonstrates the effectiveness of the proposed method for predicting metal ion binding sites based on distantly homologous proteins and generative adversarial networks.

[0083] Table 1 compares the overall prediction performance of models built using different protein data.

[0084]

[0085] The above method first extracts evolutionary information features from known metal-binding proteins, avoiding the tedious process of manually designing features. For unknown protein sequences, homologous proteins are screened through multiple sequence alignment, and target feature vectors are extracted. Cosine similarity is then used to quickly screen candidate distantly homologous metal-binding proteins, significantly improving the efficiency of homologous protein screening. A training set is constructed based on the similarity values ​​used to screen candidate distantly homologous metal-binding proteins, ensuring the consistency of data quality. A combined framework of Bi-LSTM and FCNN is used to simultaneously capture long-term dependencies and local physicochemical features of sequences. The prediction model, trained on the training set, can directly output the probability value of metal-binding sites, realizing an end-to-end prediction process. This method transforms the complex process relying on expert experience into a standardized data-driven process, reducing the randomness caused by manual intervention and improving the model's ability to identify low-homology sequences through multi-dimensional feature fusion. While ensuring prediction accuracy, it significantly improves the repeatability and cross-platform applicability of experiments, ultimately achieving stable and efficient metal-binding site prediction.

[0086] Secondly, the present invention also provides a recognition device for protein-metal ion binding sites, such as... Figure 6 As shown, it includes:

[0087] The extraction module 601 is used to extract features from metal ion-binding proteins with known structures and functions to obtain evolutionary information features of multiple samples.

[0088] The screening module 602 is used to screen a set of homologous proteins for unknown protein sequences through multiple sequence alignment and extract the evolutionary information features of homologous proteins as target feature vectors; determine the similarity value between the target feature vector and the evolutionary information features of multiple samples through cosine similarity, and screen a preset number of candidate distantly homologous metal ion-binding proteins based on the similarity value; and construct a training set based on the candidate distantly homologous metal ion-binding proteins.

[0089] Module 603 is used to construct a composite framework that combines a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN). The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using this training set to obtain a site prediction model.

[0090] The prediction module 604 is used to predict the metal ion binding sites of the target protein using the prediction model.

[0091] Using the aforementioned apparatus, evolutionary information features of samples are first obtained by extracting features from known metal-ion-binding proteins, avoiding the tedious process of manually designing features. For unknown protein sequences, homologous proteins are screened through multiple sequence alignment, and target feature vectors are extracted. Cosine similarity is then used to quickly screen candidate distantly homologous metal-ion-binding proteins, significantly improving the efficiency of homologous protein screening. A training set is constructed based on the similarity values ​​used to screen candidate distantly homologous metal-ion-binding proteins, ensuring data quality consistency. A composite framework of Bi-LSTM and FCNN is employed, simultaneously capturing long-term sequence dependencies and local physicochemical features. The prediction model, trained on the training set, can directly output the probability value of metal-ion binding sites, realizing an end-to-end prediction process. This method transforms the complex process relying on expert experience into a standardized data-driven process, reducing the randomness caused by manual intervention and improving the model's ability to identify low-homology sequences through multi-dimensional feature fusion. While ensuring prediction accuracy, it significantly improves experimental repeatability and cross-platform applicability, ultimately achieving stable and efficient metal-ion binding site prediction.

[0092] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1The steps of the method for identifying protein-metal ion binding sites are provided.

[0093] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above-mentioned functions. Figure 1 The steps of the method for identifying protein-metal ion binding sites are provided.

[0094] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0096] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0098] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for identifying protein-metal ion binding sites, characterized in that, The method includes: Feature extraction was performed on metal ion-binding proteins with known structures and functions to obtain evolutionary information features of multiple samples. For unknown protein sequences, a set of homologous proteins is screened through multiple sequence alignment, and the evolutionary information features of the homologous proteins are extracted as target feature vectors. The similarity value between the target feature vector and the evolutionary information features of multiple samples is determined by cosine similarity, and a predetermined number of candidate distant homologous metal ion-binding proteins are screened based on the similarity value. A training set is constructed based on the candidate distant homologous metal ion-binding proteins. A composite framework combining a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN) is employed. The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using the training set to obtain a site prediction model. The prediction model is used to predict the metal ion binding sites of unknown protein sequences. The evolutionary information features include the mean and standard deviation vectors of the position-specific scoring matrix and the weighted observation matrix, and the feature extraction formula is as follows: ； ； in, and These are the location-specific scoring matrix and the weighted observation matrix, respectively. and Represent the positions in the position-specific scoring matrix and the weighted observation matrix, respectively. The amino acids were replaced with amino acids The score, , Expressed as protein sequence length, This represents a set of 20 amino acids; Statistical analysis of the 20 amino acids in each metal ion-bound protein was performed separately. and The average value and standard deviation of the corresponding positions are used to obtain evolutionary information features; The formula for calculating the similarity value is: ； ； in, Represents the similarity value. and Representing vectors and In the The numerical value of the position. This represents the average vector of the position-specific scoring matrices for 20 amino acids. This represents the standard deviation vector of the position-specific scoring matrix for 20 amino acids. This represents the average vector of the weighted observation matrix of 20 amino acids. This represents the standard deviation vector of the weighted observation matrix of the 20 amino acids. Represents any protein in the set of homologous proteins. This represents any protein in a set of metal-ion-bound proteins. , , , These are the weighting coefficients; The construction of the training set based on candidate distantly homologous metal ion-binding proteins includes: Using candidate distantly homologous metal ion-binding proteins as real samples, virtual samples are generated by data augmentation of the real samples through a generative adversarial network (GAN). The training set is formed by mixing virtual samples with real samples.

2. The method according to claim 1, characterized in that, The training set includes samples of metal ion-protein binding and binding site labels. The complex framework is trained using this training set to obtain a site prediction model, including: The binding sample is input into the composite frame to obtain the predicted binding site; The composite framework was trained to obtain a site prediction model with the goal of minimizing the deviation between the predicted binding site and the binding site label.

3. A device for recognizing protein-metal ion binding sites, characterized in that, The apparatus comprising the method of claim 1, wherein the apparatus includes: The extraction module is used to extract features from metal ion-binding proteins with known structures and functions, and obtain evolutionary information features from multiple samples. The screening module is used to screen a set of homologous proteins for unknown protein sequences through multiple sequence alignment and extract the evolutionary information features of homologous proteins as target feature vectors; it determines the similarity value between the target feature vector and the evolutionary information features of multiple samples through cosine similarity, and screens a preset number of candidate distant homologous metal ion-binding proteins based on the similarity value; and it constructs a training set based on the candidate distant homologous metal ion-binding proteins. A construction module is used to construct a composite framework that combines a bidirectional long short-term memory network (Bi-LSTM) and a fully connected neural network (FCNN). The input features include evolutionary information features and physicochemical property features, and the output is the probability value of different metal ion binding sites. The composite framework is trained using the training set to obtain a site prediction model. The prediction module is used to predict the metal ion binding sites of the target protein using the prediction model.

4. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 2.

5. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 2.

Images