Protein homologous feature representation method based on sorting learning technology

Abstract

The invention discloses a protein homologous feature representation method based on a sorting learning technology. The method comprises the following steps: training a sorting model capable of correcting a PSI-BLAST result; the search sequence is input into PSI-BLAST, and a similar protein list is searched out through one-time iteration; reordering the similar protein list searched for the last time by using the ordering model; a default E-value of the PSI-BLAST is used for calculating the number of sequences needing to construct multi-sequence alignment, and a corresponding number of proteins are extracted from the reordering list in the previous step; by constructing the protein homologous feature representation method based on the sorting learning technology, on one hand, the problem of non-homologous noise in an existing homologous search method is solved through the sorting learning technology, on the other hand, improved homologous search methods are fused through the sorting learning technology, complementarity between different methods is brought into play, and the method is more efficient. And the quality of homologous features in the search result is further improved.

Description

Technical Field

[0001] The invention belongs to the field of bioinformatics, and in particular relates to a protein homology feature representation method based on ranking learning technology. Background Art

[0002] As a biological macromolecule, protein exists in every cell of the human body and is the main participant in the life activities of the human body. Whether we can truly understand the laws of protein operation in the human body is an important bottleneck for humans to understand the disease mechanism and achieve precision medicine. With the continuous maturity of sequencing technology, the cost of sequencing has dropped drastically, making a large amount of protein sequence data available to the world. As of 2022, the protein sequence data that have been measured in the UniProt database has reached an astonishing 230 million, and these data contain rich mysteries of life. However, for these protein sequence data that have been obtained, scientists currently know very little about their specific functions in living organisms, and only about 480,000 protein sequence data have been annotated with their corresponding functions. Although biologists can conduct in-depth research on the functions corresponding to protein sequence data from the perspective of biological experiments, humans cannot invest huge manpower and material resources in a short period of time to quickly analyze massive protein data, as they do with the new coronavirus protein. Therefore, the meaning of life contained in the massive protein sequence data cannot be recognized by humans, which also leads to the inability to release the huge scientific value.

[0003] In order to accelerate the understanding of the life significance contained in protein sequences, a large number of protein function prediction methods based on machine learning have emerged, aiming to determine whether a protein sequence has a specific biological function. However, for some unknown protein sequences with a similarity of less than 60% with known functional protein sequences, the accuracy of protein function prediction is still a challenging problem. In addition, innovative drug targets are the core of new drug research and development. At present, the vast majority of targets that can interact with drugs are proteins, and humans have discovered less than one thousand targets. Protein function research is the key step to further characterize the characteristics of targets. Therefore, how to more accurately predict the function corresponding to a protein sequence is an important research in the current field of bioinformatics.

[0004] Most of the existing protein function prediction methods use the search results of the BLAST method to construct protein homology features, thereby predicting the function of the protein sequence. However, since the BLAST method directly compares two protein sequences, it cannot capture some homologous proteins with low sequence similarity. Moreover, since the sequence representation of the BLAST method is relatively simple, it cannot effectively distinguish some protein sequences with high sequence similarity but non-homologous, resulting in a decrease in the accuracy of the BLAST method in predicting homology features of some protein sequences. Compared with some of the latest protein homology search methods, the prediction performance of the BLAST method has no advantage. Therefore, how to construct homology features for protein function prediction based on the latest homology search algorithm is an important prerequisite for improving the performance of protein function prediction.

[0005] Moreover, iterative search is a key technology to improve the performance of homology search methods. However, many literatures point out that this method has two serious problems in the iterative search process: 1) The search results contain non-homologous sequences; 2) Over-extension problem in sequence alignment. After analyzing these two problems on the database, it was found that the first problem of non-homologous sequences is widely present in various databases, while the over-extension problem only exists in the complete protein database. Summary of the invention

[0006] The purpose of the present invention is to provide a protein homology feature representation method based on ranking learning technology to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a protein homology feature representation method based on ranking learning technology, comprising the following steps: S1: Train a sorting model that can correct PSI-BLAST results; S2; input the search sequence into PSI-BLAST and search for a list of similar proteins using one iteration; S3: Use the ranking model to re-rank the similarity protein list from the previous search; S4: Use the default E-value of PSI-BLAST to calculate the number of sequences needed to construct a multiple sequence alignment, and extract the corresponding number of proteins from the re-ranked list in the previous step; S5: Use the protein list in the previous step to construct a multiple sequence alignment matrix MSA; S6: Use the multiple sequence alignment matrix MSA to construct the position-specific matrix PSSM; S7: input the PSSM into PSI-BLAST for the next iteration search; S8: Repeat S3-S7 for subsequent iterations.

[0008] Preferably, the S1 specifically comprises the steps of: A feature extraction system is used to extract multiple features from protein sequences, and the first iteration results of PSI-BLAST are scored using a distance measurement algorithm. These scores, together with the similarity scores given by PSI-BLAST, are input into the ranking learning method for training. Finally, a ranking model that corrects the order of PSI-BLAST results is trained.

[0009] Preferably, the feature extraction system is a PSE-IN-ONE system.

[0010] Preferably, in S3, the sorting model is the LTR-Resort sorting model Preferably, the characteristics include physicochemical properties, sequence composition, and secondary structure information of the sequence.

[0011] Preferably, the S2 specifically comprises the steps of: In the initial search stage, the protein sequence Q to be searched is input into PSI-BLAST for processing. PSI-BLAST uses the site-specific information of the sequence to search the database for a list of proteins L similar to the input sequence Q, generate similarity scores and E-values, and evaluate the degree of match of each similar sequence. The results of the initial search will serve as the basis for subsequent iterations and provide data support for subsequent sorting and screening. Based on this step, proteins similar to the target sequence are preliminarily determined.

[0012] Preferably, the S3 specifically comprises the steps of: After obtaining the initial similar protein list L, it is re-ranked using the previously trained ranking model to obtain a new ranked list L'.

[0013] Preferably, the S4 specifically comprises the steps of: In the reordered list L', the default E-value calculation method of PSI-BLAST is used to determine the number of sequences N required to construct a multiple sequence alignment. The first N protein sequences with the smallest E-value are extracted to construct a multiple sequence alignment matrix, which provides a basis for subsequent iterative searches and ensures that the selected sequences have similarity and biological relevance.

[0014] Preferably, the S5 specifically comprises the steps of: After extracting the first N protein sequences from the reordered list, these sequences are used to construct the multiple sequence alignment matrix MSA. The conserved regions and variation points between sequences are discovered through alignment. The process of constructing MSA involves global alignment between sequences to maximize sequence consistency and alignment. The generated MSA matrix contains the alignment information of all selected sequences.

[0015] Preferably, the S7 specifically comprises the steps of: The generated PSSM is input into PSI-BLAST for the next iterative search, and the PSSM is used for iterative search to identify proteins similar to the target sequence and generate new search results.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention constructs a protein homology feature representation method based on ranking learning technology. On the one hand, the ranking learning technology is used to reduce the non-homologous noise problem in the existing homology search method. On the other hand, the ranking learning technology is used to integrate the improved homology search method to give play to the complementarity between different methods and further improve the quality of homology features in the search results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0019] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0020] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0021] See also Figure 1 As shown, the present invention provides the following technical solutions: like Figure 1 As shown, in the present invention, the ranking learning technology is used to filter non-homologous noise. In the improvement research of the protein homology search method, the protein sequence spectrum is mainly used instead of the original protein sequence for homology search. Compared with the simple description of the protein sequence, the protein sequence spectrum can describe the mutation probability at the protein site from an evolutionary perspective, which becomes an important prerequisite for improving the matching accuracy between protein sequences. However, since the protein sequence spectrum often contains non-homologous noise problems, the performance of the homologous protein search method is limited. Therefore, this study intends to split the advanced protein homology search framework and reconstruct the construction part of the protein sequence spectrum. In the protein sequence spectrum construction part, the sequence similarity and feature similarity will be fused based on the ranking learning technology, so as to enhance the ability to distinguish non-homologous proteins and ultimately enhance the accuracy of the homology search method.

[0022] In addition, in the present invention, if Figure 1 As shown in the figure, the improved homology search method is fused by ranking learning technology. Since the complementarity between different homology search methods has been proved by relevant research, the final performance of directly fusing homology search methods is constrained by the upper limit of the performance of each homology search method. Because, although there is a certain complementarity between homology search methods, for some difficult-to-predict protein sequences, the prediction performance after fusion is improved, but it is still difficult to meet the actual needs.

[0023] Therefore, the present invention is based on the improved homology search method in the first part for fusion. After solving the non-homologous noise problem in the homology search method, the ranking learning technology is used to further improve the quality of homology features in the search results from the perspective of complementarity of different methods.

[0024] In addition, in the present invention, a more accurate iterative search method is provided, and the specific process is as follows: The protein homology feature representation method based on ranking learning technology includes the following steps: S1: Train a sorting model that can correct PSI-BLAST results; S2; input the search sequence into PSI-BLAST and search for a list of similar proteins using one iteration; S3: Use the ranking model to re-rank the similarity protein list from the previous search; S4: Use the default E-value of PSI-BLAST to calculate the number of sequences needed to construct a multiple sequence alignment, and extract the corresponding number of proteins from the re-ranked list in the previous step; S5: Use the protein list in the previous step to construct a multiple sequence alignment matrix MSA; S6: Use the multiple sequence alignment matrix MSA to construct the position-specific matrix PSSM; S7: input the PSSM into PSI-BLAST for the next iteration search; S8: Repeat S3-S7 for subsequent iterations.

[0025] In addition, in the present invention, specifically, in one embodiment, the detailed steps of the above S1 are: S1: First, use the PSE-IN-ONE system to extract multiple features from the protein sequence, including the physicochemical properties of the sequence, sequence composition, secondary structure information, etc. Set a feature vector F to represent the multiple features of each protein sequence, and these feature vectors will be used for subsequent ranking learning. Then, the distance metric algorithm (Euclidean distance) is used to calculate the similarity scores between proteins to form a similarity matrix D. Next, score the results of the first iteration of PSI-BLAST to obtain a similarity score S. The feature vector F and the similarity score S are input together into the ranking learning method for training to obtain a ranking model θ that corrects the order of the PSI-BLAST results. This model can combine multiple protein features and PSI-BLAST scores to improve the accuracy of the ranking.

[0026] The formula is as follows: ; ; ; ; Regarding the above S2 step, specifically: S2: In the initial search phase, the protein sequence Q to be searched is input into PSI-BLAST for processing. PSI-BLAST uses the site-specific information of the sequence to search the database for a list of proteins L that are similar to the input sequence Q. This process generates similarity scores and E-values ​​to evaluate the degree of match for each similar sequence. The results of the initial search will serve as the basis for subsequent iterations, providing data support for further sorting and screening. Through this step, proteins similar to the target sequence can be preliminarily determined, laying the foundation for further analysis and sorting. The formula is as follows: ; Regarding the above S3 step, specifically: S3: After obtaining the initial similar protein list L, use the previously trained ranking model θ to re-rank it. The ranking model θ comprehensively considers the multiple protein features extracted by the PSE-IN-ONE system and the similarity scores of PSI-BLAST, so that it can more accurately reflect the actual similarity between proteins. By re-ranking, a new ranking list L' is obtained, and the order of protein sequences in this list is more consistent with actual biological significance. The purpose of this step is to improve the accuracy of multiple sequence alignment and position-specific matrix construction in subsequent steps through more accurate ranking, thereby enhancing the effect of the overall iterative search. The formula is as follows: ; Regarding the above S4 step, specifically: S4: In the reordered list L', use the default E-value calculation method of PSI-BLAST to determine the number of sequences N required to construct a multiple sequence alignment. E-value is a statistical indicator to measure the significance of the result. A smaller E-value indicates a higher similarity. By extracting the first N protein sequences with the smallest E-value, it can be ensured that the selected sequences have high similarity and biological relevance. These extracted sequences will be used to construct a multiple sequence alignment matrix, providing a basis for subsequent iterative searches. The key to this step is to screen out the most relevant sequences through a reasonable E-value threshold and improve the quality of the multiple sequence alignment. The formula is as follows: ; Regarding the above S5 step, specifically: S5: After extracting the first N protein sequences from the reordered list, these sequences are used to construct the multiple sequence alignment matrix MSA. Multiple sequence alignment (MSA) is an important tool for analyzing the homology between multiple sequences. Through alignment, conserved regions and variation points between sequences can be discovered. The process of constructing MSA involves global alignment between sequences to maximize sequence consistency and alignment. The generated MSA matrix will contain the alignment information of all selected sequences, which is essential for the subsequent site-specific matrix construction. High-quality multiple sequence alignments can further improve the accuracy and reliability of iterative searches; Regarding the above S6 step, specifically: S6: Using the constructed multiple sequence alignment matrix MSA, generate the position-specific scoring matrix PSSM (Position-Specific Scoring Matrix). PSSM is a matrix generated based on multiple sequence alignment, which reflects the frequency and conservation of amino acids at each position. By counting the frequency of occurrence of amino acids at each position and combining the background distribution, the specificity score of each position can be calculated. The generated PSSM provides more detailed site information, so that subsequent PSI-BLAST searches can use this information for more accurate matching. By constructing PSSM, the sensitivity of the search model to sequence specificity and conserved regions is improved, thereby improving the accuracy of the overall search. The formula is as follows: PSSM=Construct-PSSM(MSA); S7: The generated PSSM is input into PSI-BLAST for the next iteration search. Using PSSM for iterative search can more accurately identify proteins similar to the target sequence, thereby optimizing the search results. i+1 The steps of re-ranking, E-value calculation, multiple sequence alignment and position-specific matrix construction will be repeated. By repeating these steps, the search results are continuously optimized and refined. The iterative search process continues until the predetermined stop condition is met, such as reaching the maximum number of iterations or the search results are stable. Through multiple iterations, the accuracy and reliability of the search are gradually improved, and finally highly related protein sequences are obtained. The formula is as follows: .

[0027] In addition, the present invention provides the following embodiments: S1: Data preprocessing Data collection: Collect a large amount of protein sequence data from databases such as UniProt and PDB to ensure the diversity and representativeness of the data.

[0028] Data cleaning: Remove duplicate, incomplete or low-quality sequences to ensure the accuracy and reliability of input data.

[0029] Step 2: Feature extraction system preparation PSE-IN-ONE system configuration: The PSE-IN-ONE system is configured to ensure the extraction of multiple features including physicochemical properties of the sequence (e.g., hydrophobicity, charge distribution), sequence composition (e.g., amino acid frequency), secondary structure information (e.g., helices, folded regions), etc.

[0030] Feature selection: According to the specific task requirements, select the most critical feature set for protein homology detection from the multiple extracted features.

[0031] Step S3: Sorting model training Data annotation: The results of the first iteration of PSI-BLAST were annotated manually or using known homology relationships as training data.

[0032] Model training: Using the labeled data and features extracted from the PSE-IN-ONE system, the sorting model is trained using sorting learning techniques (such as Pointwise, Pairwise, or Listwise methods). During the training process, hyperparameters such as learning rate and regularization parameters can be adjusted to optimize model performance.

[0033] Step S4: Model evaluation and selection Cross-validation: Use cross-validation methods (such as K-fold cross-validation) to evaluate the generalization ability of the model and select the model with the best performance as the final ranking model.

[0034] Performance evaluation indicators: Use indicators such as accuracy, recall, and F1 score to comprehensively evaluate the performance of the model.

[0035] Step S5: Iteration control: Set a reasonable number of iterations or convergence conditions to avoid overfitting or waste of computing resources.

[0036] Feature updating: In each iteration, in addition to using the PSSM generated by PSI-BLAST, the protein sequence features in the list re-ranked by the ranking model can also be considered for inclusion in the feature extraction of subsequent iterations to update the feature matrix and improve the accuracy of detection.

[0037] Step S6: Result analysis Results summary: The results of each iteration were summarized to analyze the correction effect of the sorting model on the PSI-BLAST results and the quality of the construction of the multiple sequence alignment matrix (MSA).

[0038] Biological significance interpretation: Combined with biological knowledge, the detected homologous proteins are analyzed in terms of function, structure, etc. to verify their homology.

[0039] Step S7: Independent Verification Independent test set validation: Use a test set that is completely independent of the training data to validate the method to ensure the robustness and generalization ability of the method.

[0040] Performance comparison: This method was compared with other existing protein homology detection methods to evaluate their advantages and disadvantages.

[0041] Step S8: Application scenario expansion Disease-related protein detection: Apply this method to the detection of disease-related proteins to provide new clues for disease diagnosis and treatment.

[0042] Drug design: Use the detected homologous protein information to assist in drug target discovery and drug design.

[0043] Step S9: Continuous optimization and updating Algorithm optimization: With the development of sorting learning technology and bioinformatics, the algorithm is continuously optimized and updated to improve detection efficiency and accuracy.

[0044] Database update: Regularly update the protein database to ensure the timeliness and accuracy of the data.

[0045] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus. Memory, used to store computer programs; The processor is used to implement the protein homology feature representation method based on ranking learning technology of the present invention when executing the program stored in the memory.

[0046] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the above terminal and other devices. The memory may include a random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. Optionally, the memory may also be at least one storage system located away from the aforementioned processor.

[0047] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0048] In addition, to achieve the above-mentioned purpose, an embodiment of the present invention further proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for diabetic retinopathy lesion segmentation based on a heterogeneous convolutional perception network of an embodiment of the present invention.

[0049] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable vehicles (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] The embodiments of the present invention are described with reference to flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0051] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction system, which is implemented in the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0053] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "" and / or "" indicate that either one of the two can be selected, or both can be selected. Moreover, the terms "" include "" "" contain "" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "" include a..."" does not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

[0055] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Protein homology feature representation method based on ranking learning technology, Features: The steps include: S1: Train a sorting model that can correct PSI-BLAST results; S2; input the search sequence into PSI-BLAST and search for a list of similar proteins using one iteration; S3: Use the ranking model to re-rank the similarity protein list from the previous search; S4: Use the default E-value of PSI-BLAST to calculate the number of sequences needed to construct a multiple sequence alignment, and extract the corresponding number of proteins from the re-ranked list in the previous step; S5: Use the protein list in the previous step to construct a multiple sequence alignment matrix MSA; S6: Use the multiple sequence alignment matrix MSA to construct the position-specific matrix PSSM; S7: input the PSSM into PSI-BLAST for the next iteration search; S8: Repeat S3-S7 for subsequent iterations.

2. The protein homology feature representation method based on ranking learning technology according to claim 1, Features: The S1 specifically comprises the steps of: A feature extraction system is used to extract multiple features from protein sequences, and the first iteration results of PSI-BLAST are scored using a distance measurement algorithm. These scores, together with the similarity scores given by PSI-BLAST, are input into the ranking learning method for training. Finally, a ranking model that corrects the order of PSI-BLAST results is trained.

3. The protein homology feature representation method based on ranking learning technology according to claim 2, Features: The feature extraction system is a PSE-IN-ONE system.

4. The protein homology feature representation method based on ranking learning technology according to claim 3, Features: In S3, the sorting model is the LTR-Resort sorting model.

5. The protein homology feature representation method based on ranking learning technology according to claim 4, Features: The characteristics include physicochemical properties of the sequence, sequence composition, and secondary structure information.

6. The protein homology feature representation method based on ranking learning technology according to claim 5, Features: The S2 specifically includes the steps of: In the initial search stage, the protein sequence Q to be searched is input into PSI-BLAST for processing. PSI-BLAST uses the site-specific information of the sequence to search the database for a list of proteins L similar to the input sequence Q, generate similarity scores and E-values, and evaluate the degree of match of each similar sequence. The results of the initial search will serve as the basis for subsequent iterations and provide data support for subsequent sorting and screening. Based on this step, proteins similar to the target sequence are preliminarily determined.

7. The protein homology feature representation method based on ranking learning technology according to claim 6, Features: The S3 specifically includes the steps of: After obtaining the initial similar protein list L, it is re-ranked using the previously trained ranking model to obtain a new ranked list L'.

8. The protein homology feature representation method based on ranking learning technology according to claim 7, Features: The S4 specifically comprises the steps of: In the reordered list L', the default E-value calculation method of PSI-BLAST is used to determine the number of sequences N required to construct a multiple sequence alignment. The first N protein sequences with the smallest E-value are extracted to construct a multiple sequence alignment matrix, which provides a basis for subsequent iterative searches and ensures that the selected sequences have similarity and biological relevance.

9. The protein homology feature representation method based on ranking learning technology according to claim 8, Features: The S5 specifically includes the steps of: After extracting the first N protein sequences from the reordered list, these sequences are used to construct the multiple sequence alignment matrix MSA. The conserved regions and variation points between sequences are discovered through alignment. The process of constructing MSA involves global alignment between sequences to maximize sequence consistency and alignment. The generated MSA matrix contains the alignment information of all selected sequences.

10. The protein homology feature representation method based on ranking learning technology according to claim 9, Features: The S7 specifically includes the steps of: The generated PSSM is input into PSI-BLAST for the next iterative search, and the PSSM is used for iterative search to identify proteins similar to the target sequence and generate new search results.

Images