Artificial intelligence-based gene and gene cluster function prediction method and device
By constructing a predictive model based on artificial intelligence, potential uric acid metabolism gene clusters in the human gut microbiota were screened out, solving the problems of time-consuming, labor-intensive and inaccurate traditional methods, and achieving efficient and accurate gene cluster function prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2022-12-13
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies struggle to efficiently mine gene clusters with unknown functions in the human gut microbiota, especially the uricase gene cluster. Traditional methods are time-consuming, labor-intensive, and inaccurate in predicting function.
Using an artificial intelligence-based approach, a predictive model was constructed, and a training set of existing functional protein information was used to screen for potential functional gene clusters, especially those related to uric acid metabolism.
This improved the efficiency and accuracy of gene cluster function prediction, led to the discovery of more potential uric acid metabolism-related gene clusters, and reduced the cost of subsequent biological experiments.
Smart Images
Figure CN116030881B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of artificial intelligence. Background Technology
[0002] The human gut microbiome is crucial for maintaining human health and is closely related to the development of various diseases, including type 2 diabetes, obesity, non-alcoholic fatty liver disease, and colorectal cancer. Thousands of microorganisms have been discovered in the human gut, and the number of genes encoded by gut microbiota is more than 150 times the number of human genes. Currently, the function of 90% of gut microbiome genes remains unknown. The rapid development of gut microbiota research is attributed to the application of technologies such as whole-genome and metagenomic sequencing, transcriptomics, and proteomics in the field of human microbiome research. This has enabled the discovery of a wealth of information about gut microbiota, including gene expression, protein abundance, and strain or species composition. This includes previously unknown biological information and functional data, which biologists refer to as "dark matter."
[0003] Thousands of prokaryotic genomes in microbial sequence databases encode various metabolic enzymes through biosynthetic gene clusters (BGCs), which are physically clustered (genomically close) sets of genes. Gut microbiota gene clusters express enzymes, catalyze intestinal digestion, metabolize cellulose to provide the host with vitamins, and provide nutrients such as fats and sugars. These functions are accomplished by functional gene clusters on the microbial genome. While some gene clusters have been resolved, and genes synthesizing small molecule products have been heterologously recombined or simulated in vitro using purified enzymes, the distribution and function of the vast majority of gene clusters, even those known to produce small molecules, remain incompletely elucidated. Furthermore, 90% of protein sequences in the human gut microbiome lack functional annotation. Moreover, due to computational annotation based on sequence similarity, misannotation and overprediction of enzyme functions frequently occur in public databases. Therefore, much information about gene clusters with unknown functions remains unknown, and the resolution of these unresolved functional gene clusters can provide insights for novel targeted microbiota interventions for disease treatment.
[0004] Currently, the screening and evaluation of functional gut microbiota strains mainly relies on a pathway of metagenomic sequencing data analysis – strain isolation – strain functional verification. While this approach is mature, it is time-consuming and labor-intensive. Therefore, there is a need to find rapid and efficient new research methods to apply to the vast amounts of biomic data available today, combined with in vitro and in vivo validation, to improve the efficiency of strain functional research. Natural products produced by bacteria during secondary metabolism possess rich chemical structures and biological activities, including various types of small molecule drug candidates such as antibiotics, anticancer drugs, and antiviral drugs, making them an important resource for novel drug development. In the bacterial genome, genes encoding the synthesis of various natural products exist in the form of biosynthetic gene clusters (BGCs), laying the theoretical foundation for the discovery of natural products from sequence to phenotype.
[0005] Taking uric acid metabolism as an example, most of the strains whose uricase genes have been annotated are from soil, plant pathogens and environmental microorganisms. Compared with the thousands of microorganisms in the human gut, the known uricase gene clusters in the human gut microbiota are very few, and potential strains involved in uric acid metabolism need to be discovered.
[0006] Current gene cluster databases are limited in information, and methods for gene function mining based on BLAST sequence alignment cannot effectively discover new gene functions and new metabolic pathways. There is a lack of methods for mining novel functions of gut microbiota gene clusters using artificial intelligence technology, highlighting the urgent need for methodological innovation. While BGC prediction tools based on traditional machine learning methods, such as ClusterFinder, employ methods like Hidden Markov Models (HMMs), their ability to detect novel BGCs is limited due to the inherent limitations of the algorithms themselves. Summary of the Invention
[0007] The present invention aims to at least partially solve one of the technical problems in the related art.
[0008] Therefore, the purpose of this invention is to propose an artificial intelligence-based method for predicting gene and gene cluster functions, which can be used to predict, identify and screen specific functional gene clusters.
[0009] To achieve the above objectives, a first aspect of the present invention proposes a method for predicting gene and gene cluster functions based on artificial intelligence, comprising:
[0010] Obtain all protein sequences from the target genome;
[0011] By utilizing the functional annotation information of proteins with existing target functions, a model training set is constructed, and a prediction model is trained.
[0012] The target function is predicted for all protein sequences using the prediction model, and the sequence information function score of the protein sequence is obtained.
[0013] The protein structure prediction model is used to predict the target function of the protein sequence, and the structural information and function score of the protein sequence are obtained.
[0014] The combined result of the sequence information function score and the structure information function score is used as the final score of the protein sequence. Based on the final scores of all protein sequences, all gene clusters on the target genome are evaluated, and the gene cluster with the highest score is selected as the candidate gene cluster.
[0015] In addition, the gene and gene cluster function prediction method based on artificial intelligence according to the above embodiments of the present invention may also have the following additional technical features:
[0016] Furthermore, in one embodiment of the present invention, after obtaining all protein sequences of the target genome, the method further includes:
[0017] The protein sequence is divided into protein sequence fragments of predetermined length.
[0018] Furthermore, in one embodiment of the present invention, the step of constructing a model training set and training a prediction model using existing functional annotation information of proteins with functions to be predicted includes:
[0019] From the gut microbiota database, protein sequences not related to the predicted function are filtered out. The gene sequence size of the protein sequences not related to the predicted function is reduced using a clustering tool and a machine learning clustering algorithm is used to select representative sequences as a negative sample dataset. Protein sequences with the predicted function are used as a positive sample dataset. The process also includes cutting the protein sequences not related to the predicted function and the protein sequences with the predicted function into protein sequence fragments of a predetermined length.
[0020] A prediction model is trained based on the negative sample dataset and the positive sample dataset.
[0021] Further, in one embodiment of the present invention, the step of evaluating all gene clusters on the target genome based on the final scores of all protein sequences and selecting the gene cluster with the highest score as a candidate gene cluster includes:
[0022] All protein sequences are sorted according to their final scores, and the K protein sequences with the highest scores are selected as anchors.
[0023] On the target genome, with each anchor point as the center, the functional evaluation score of the region corresponding to the anchor point is obtained by calculation and analysis based on the functional scores of adjacent protein sequences within a predetermined length range of the anchor point and the protein function annotation results.
[0024] Based on the functional assessment scores of all regions, the gene cluster with the highest functional assessment score was selected as the candidate gene cluster.
[0025] To achieve the above objectives, a second aspect of the present invention provides an artificial intelligence-based gene and gene cluster function prediction device, comprising the following modules:
[0026] The acquisition module is used to acquire all protein sequences of the target genome;
[0027] The training module is used to build a model training set by utilizing the functional annotation information of proteins with existing target functions, and to train a prediction model.
[0028] The sequence prediction module is used to predict the target function of all protein sequences using the prediction model, and to obtain the sequence information function score of the protein sequence.
[0029] The structure prediction module is used to predict the target function of the protein sequence by using a protein structure prediction model, and to obtain the structural information and function score of the protein sequence.
[0030] The gene cluster prediction module is used to combine the functional score of the sequence information and the functional score of the structure information as the final score of the protein sequence. Based on the final scores of all protein sequences, it evaluates all gene clusters on the target genome and selects the gene cluster with the highest score as the candidate gene cluster.
[0031] Furthermore, in one embodiment of the present invention, the acquisition module is further configured to:
[0032] The protein sequence is divided into protein sequence fragments of predetermined length.
[0033] Furthermore, in one embodiment of the present invention, the training module is further configured to:
[0034] From the gut microbiota database, protein sequences not related to the predicted function are filtered out. The gene sequence size of the protein sequences not related to the predicted function is reduced using a clustering tool and a machine learning clustering algorithm is used to select representative sequences as a negative sample dataset. Protein sequences with the predicted function are used as a positive sample dataset. The process also includes cutting the protein sequences not related to the predicted function and the protein sequences with the predicted function into protein sequence fragments of a predetermined length.
[0035] A prediction model is trained based on the negative sample dataset and the positive sample dataset.
[0036] Furthermore, in one embodiment of the present invention, the gene cluster prediction module is further configured to:
[0037] All protein sequences are sorted according to their final scores, and the K protein sequences with the highest scores are selected as anchors.
[0038] On the target genome, with each anchor point as the center, the functional evaluation score of the region corresponding to the anchor point is obtained by calculation and analysis based on the functional scores of adjacent protein sequences within a predetermined length range of the anchor point and the protein function annotation results.
[0039] Based on the functional assessment scores of all regions, the gene cluster with the highest functional assessment score was selected as the candidate gene cluster.
[0040] To achieve the above objectives, a third aspect of the present invention provides a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements an artificial intelligence-based gene and gene cluster function prediction method as described above.
[0041] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements an artificial intelligence-based gene and gene cluster function prediction method as described above.
[0042] The gene and gene cluster function prediction method based on artificial intelligence proposed in this invention focuses on predicting and classifying all known genomic and protein function types simultaneously, rather than predicting and classifying each function of interest, such as the uric acid-lowering function, as is the case with other models. Instead, it constructs a targeted dataset and sequence prediction model to predict, identify, and screen gene clusters with specific functions. Attached Figure Description
[0043] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0044] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for predicting gene and gene cluster functions, as provided in an embodiment of the present invention.
[0045] Figure 2 The flowchart illustrates the complete AI-based gene and gene cluster function prediction method provided in this embodiment of the invention.
[0046] Figure 3 This is a schematic diagram of the gene cluster for predicting uric acid metabolism provided in an embodiment of the present invention.
[0047] Figure 4 This is a schematic diagram of a gene and gene cluster function prediction device based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation
[0048] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0049] The following describes an embodiment of the artificial intelligence-based gene and gene cluster function prediction method of the present invention with reference to the accompanying drawings.
[0050] Example 1
[0051] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for predicting gene and gene cluster functions, as provided in an embodiment of the present invention.
[0052] like Figure 1 As shown, this AI-based method for predicting gene and gene cluster functions includes the following steps:
[0053] S101: Obtain all protein sequences of the target genome;
[0054] Furthermore, in one embodiment of the present invention, after obtaining all protein sequences of the target genome, the method further includes:
[0055] The protein sequence is cut into protein sequence fragments of predetermined length.
[0056] S102: Utilize the functional annotation information of proteins with existing target functions to construct a model training set and train a prediction model.
[0057] Furthermore, in one embodiment of the present invention, a model training set is constructed using existing functional annotation information of proteins with functions to be predicted, and a prediction model is trained, including:
[0058] From the gut microbiota database, protein sequences not related to the predicted function are filtered out. The gene sequence size of the protein sequences not related to the predicted function is reduced using clustering tools and representative sequences are selected using machine learning clustering algorithms as negative sample datasets. Protein sequences with the predicted function are used as positive sample datasets. This also includes cutting the protein sequences not related to the predicted function and the protein sequences with the predicted function into protein sequence fragments of predetermined lengths.
[0059] The prediction model is trained using negative and positive sample datasets.
[0060] First, a gut microbiota database was constructed. Information on known species in the human gut was collected and organized from genomic and gut microbiota-related databases such as NCBI Genome, IMG / M, and GMRepo. This included the species' genome sequence, protein annotations, nutritional and metabolic types, ability to be cultured independently, aerobicness, and relative abundance distribution in healthy individuals and patients. This information was used for subsequent artificial intelligence model construction and strain screening.
[0061] Then, the model training and testing datasets were constructed. For functions of interest, enzyme information involved in metabolism was obtained from metabolic pathways in KEGG, and relevant enzyme protein sequence information was obtained from the Uniprot database; in addition, functional sequences and metabolic pathway information reported in previous studies were supplemented based on literature review results.
[0062] Then, non-target function-related protein sequences are filtered out from the gut microbiota database. Considering the large amount of data, sequence clustering tools such as CDHIT are used to reduce the size of the gene sequences. Furthermore, machine learning clustering algorithms such as K-means and DBSCAN are used to select representative sequences through clustering, which serve as the negative sample dataset for model construction.
[0063] To construct a model for predicting functions of interest, known functional sequences and enzyme sequences along metabolic pathways are used as positive samples. Since different enzymes have different protein sequence lengths, the sequences are cut into equal-length segments, with each segment consisting of a short sequence of a specific length, such as 50 amino acids, forming a separate sample. Finally, 20% of the protein sequences from each of the positive and negative samples are randomly selected to form the model test dataset, and the remaining 80% constitute the model training dataset.
[0064] Next, model training and performance evaluation are performed. Model training can leverage deep learning frameworks like PyTorch or the scikit-learn machine learning library, combining various artificial intelligence methods to build sequence function prediction models. By integrating multiple trained models, the accuracy of sequence function discrimination can be further improved. For example, traditional machine learning models, including logistic regression, random forests, and gradient boosting trees, as well as deep learning models such as CNNs, LSTMs, and BERT, can be incorporated to transform the prediction of functional sequences into a sequence classification problem.
[0065] After the model is trained, the genomes of all strains need to be scanned sequentially, and the functional scores of all protein sequences on the genome sequence of each strain need to be calculated. Since long protein sequences are segmented, the scores of all segmented fragments need to be considered when calculating the functional score of a protein sequence.
[0066] S103: Predict the target function of all protein sequences using a prediction model, and obtain the sequence information function score of the protein sequences;
[0067] S104: Use protein structure prediction models to predict the target function of protein sequences and obtain structural information and functional scores of protein sequences.
[0068] While modeling and predicting potential functional protein sequences based on amino acid sequences to identify them doesn't consider three-dimensional protein structure information, deep learning protein structure prediction tools can be used to further predict the structure of the potential sequences selected in S103. Then, based on the prediction results, methods from the field of computer-aided drug design can be used to evaluate the docking of the protein with uric acid molecules. The evaluation results are then fused with the prediction results from S103 based on one-dimensional sequence information, thereby further reducing the number of potential uricases and lowering the cost of subsequent biological experiments.
[0069] S105: The combined result of sequence information function score and structural information function score is used as the final score of the protein sequence. Based on the final scores of all protein sequences, all gene clusters on the target genome are evaluated, and the gene cluster with the highest score is selected as the candidate gene cluster.
[0070] Furthermore, in one embodiment of the present invention, based on the final scores of all protein sequences, all gene clusters on the target genome are evaluated, and the gene cluster with the highest score is selected as a candidate gene cluster, including:
[0071] All protein sequences are sorted according to their final scores, and the K protein sequences with the highest scores are selected as anchors.
[0072] On the target genome, with each anchor point as the center, the functional scores of adjacent protein sequences within a predetermined length range of the anchor point and the protein function annotation results are calculated and analyzed to obtain the functional evaluation score of the region corresponding to the anchor point.
[0073] Based on the functional assessment scores of all regions, the gene cluster with the highest functional assessment score was selected as the candidate gene cluster.
[0074] By using artificial intelligence models to score the functional scores of protein sequences encoded by bacterial strains, potential uric acid metabolism gene clusters can be further mined from the genome. First, protein sequences are sorted and screened based on their predicted functional scores, and the K highest-scoring protein sequences are selected as anchor points. Then, within a fixed length range (e.g., 20 kb) centered on each anchor point, other genome-encoded protein sequences are searched. The functional scores of neighboring protein sequences and protein functional annotation results are used for computational analysis to obtain an overall assessment of the function of that region. Finally, based on the functional assessment scores of different regions, the top-performing regions are selected as potential gene clusters for biological validation.
[0075] The above is a complete flowchart of the artificial intelligence-based gene and gene cluster function prediction method. Figure 2 This is a schematic diagram of the technical route of the present invention.
[0076] Example 2
[0077] like Figure 3 As shown, taking the prediction of uric acid metabolism gene clusters as an example, a learning model is constructed based on known uricase genes and protein sequences to predict potential uricase genes in human intestinal strains and the strains' ability to lower uric acid. Protein sequences are segmented into fixed-length fragments, and positive and negative samples are divided using existing protein functional annotation information to construct a model training set, thereby training the prediction model. For the genome of each bacterial strain, the trained model is used to predict the functional score of each fragment in uric acid metabolism. Then, the scores of all fragments of a protein are calculated to obtain the protein's score. This score can be further combined with protein structure prediction models (such as Alphafold2) to predict the protein sequence structure, using the binding ability of protein structure judgment and specific molecules (such as uric acid analysis) to supplement the overall functional score of the protein. For a complete genome, the positional relationship of proteins on the genome is considered to evaluate the functional scores of multiple proteins within a large region, thereby obtaining the gene cluster score. All potential gene clusters on the genome are scanned, and the gene cluster with the highest score is selected as the candidate gene cluster.
[0078] The gene and gene cluster function prediction method based on artificial intelligence proposed in this invention focuses on predicting and classifying all known genomic and protein function types simultaneously, rather than predicting and classifying each function of interest, such as the uric acid-lowering function, as is the case with other models. Instead, it constructs a targeted dataset and sequence prediction model to predict, identify, and screen gene clusters with specific functions.
[0079] Compared with traditional sequence alignment BLAST methods and other machine learning-based algorithms, the artificial intelligence gene cluster prediction method based on this invention can effectively and specifically mine gene sequences and gene clusters with unknown functions on microbial genomes, discover more novel gene sequences and gene clusters with specific functions of interest to researchers, thereby helping to discover new functional strains.
[0080] To achieve the above embodiments, the present invention also proposes an artificial intelligence-based gene and gene cluster function prediction device.
[0081] Figure 4 This is a schematic diagram of the structure of an artificial intelligence-based gene and gene cluster function prediction device provided in an embodiment of the present invention.
[0082] like Figure 4 As shown, the AI-based gene and gene cluster function prediction device includes: an acquisition module 100, a training module 200, a sequence prediction module 300, a structure prediction module 400, and a gene cluster prediction module 500, wherein...
[0083] The acquisition module is used to acquire all protein sequences of the target genome;
[0084] The training module is used to build a model training set by utilizing the functional annotation information of proteins with existing target functions, and to train a prediction model.
[0085] The sequence prediction module is used to predict the target function of all protein sequences through a prediction model and obtain the sequence information function score of the protein sequence.
[0086] The structure prediction module is used to predict the target function of a protein sequence using a protein structure prediction model, and to obtain the structural information and functional score of the protein sequence.
[0087] The gene cluster prediction module is used to combine the sequence information function score and the structure information function score as the final score of the protein sequence. Based on the final scores of all protein sequences, it evaluates all gene clusters on the target genome and selects the gene cluster with the highest score as the candidate gene cluster.
[0088] Furthermore, in one embodiment of the present invention, the acquisition module is also used for:
[0089] The protein sequence is cut into protein sequence fragments of predetermined length.
[0090] Furthermore, in one embodiment of the present invention, the training module is also used for:
[0091] From the gut microbiota database, protein sequences not related to the predicted function are filtered out. The gene sequence size of the protein sequences not related to the predicted function is reduced using clustering tools and representative sequences are selected using machine learning clustering algorithms as negative sample datasets. Protein sequences with the predicted function are used as positive sample datasets. This also includes cutting the protein sequences not related to the predicted function and the protein sequences with the predicted function into protein sequence fragments of predetermined lengths.
[0092] The prediction model is trained using negative and positive sample datasets.
[0093] Furthermore, in one embodiment of the present invention, the gene cluster prediction module is also used for:
[0094] All protein sequences are sorted according to their final scores, and the K protein sequences with the highest scores are selected as anchors.
[0095] On the target genome, with each anchor point as the center, the functional scores of adjacent protein sequences within a predetermined length range of the anchor point and the protein function annotation results are calculated and analyzed to obtain the functional evaluation score of the region corresponding to the anchor point.
[0096] Based on the functional assessment scores of all regions, the gene cluster with the highest functional assessment score was selected as the candidate gene cluster.
[0097] To achieve the above objectives, a third aspect of the present invention provides a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the artificial intelligence-based gene and gene cluster function prediction method as described above.
[0098] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the artificial intelligence-based gene and gene cluster function prediction method as described above.
[0099] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0100] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0101] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for predicting gene and gene cluster function based on artificial intelligence, characterized in that, Includes the following steps: Obtain all protein sequences from the target genome; By utilizing the functional annotation information of proteins with existing target functions, a model training set is constructed, and a prediction model is trained. The target function is predicted for all protein sequences using the prediction model, and the sequence information function score of the protein sequence is obtained. The protein structure prediction model is used to predict the target function of the protein sequence, and the structural information and function score of the protein sequence are obtained. The combined result of the sequence information functional score and the structure information functional score is used as the final score of the protein sequence. All protein sequences are sorted according to their final scores, and the K protein sequences with the highest scores are selected as anchor points. On the target genome, centered on each anchor point, functional evaluation scores are calculated and analyzed based on the functional scores of neighboring protein sequences within a predetermined length range of the anchor point and the protein functional annotation results to obtain the functional evaluation score of the region corresponding to the anchor point. Based on the functional evaluation scores of all regions, the gene cluster with the highest functional evaluation score is selected as a candidate gene cluster. A deep learning protein structure prediction tool is used to predict the structure of the selected potential sequences. Then, based on the prediction results, relevant methods in the field of computer-aided drug design are used to evaluate the docking of the protein with uric acid molecules.
2. The method according to claim 1, characterized in that, After obtaining all protein sequences of the target genome, the process also includes: The protein sequence is divided into protein sequence fragments of predetermined length.
3. The method according to claim 1, characterized in that, The process of constructing a model training set and training a prediction model using existing functional annotation information of proteins with functions to be predicted includes: From the gut microbiota database, protein sequences not related to the predicted function are filtered out. The gene sequence size of the protein sequences not related to the predicted function is reduced using a clustering tool and a machine learning clustering algorithm is used to select representative sequences as a negative sample dataset. Protein sequences with the predicted function are used as a positive sample dataset. The process also includes cutting the protein sequences not related to the predicted function and the protein sequences with the predicted function into protein sequence fragments of a predetermined length. A prediction model is trained based on the negative sample dataset and the positive sample dataset.
4. A gene and gene cluster function prediction device based on artificial intelligence, characterized in that, Includes the following modules: The acquisition module is used to acquire all protein sequences of the target genome; The training module is used to build a model training set by utilizing the functional annotation information of proteins with existing target functions, and to train a prediction model. The sequence prediction module is used to predict the target function of all protein sequences using the prediction model, and to obtain the sequence information function score of the protein sequence. The structure prediction module is used to predict the target function of the protein sequence by using a protein structure prediction model, and to obtain the structural information and function score of the protein sequence. The gene cluster prediction module is used to combine the functional scores of the sequence information and the functional scores of the structure information as the final score of the protein sequence. Based on the final scores of all protein sequences, the module sorts them and selects the K protein sequences with the highest scores as anchor points. On the target genome, centered on each anchor point, the module calculates and analyzes the functional scores of neighboring protein sequences within a predetermined length range of the anchor point, along with protein function annotation results, to obtain the functional evaluation score of the region corresponding to the anchor point. Based on the functional evaluation scores of all regions, the module selects the gene cluster with the highest functional evaluation score as a candidate gene cluster. A deep learning protein structure prediction tool is used to predict the structure of the selected potential sequences. Then, based on the prediction results, relevant methods in the field of computer-aided drug design are used to evaluate the docking of the protein with uric acid molecules.
5. The apparatus according to claim 4, characterized in that, The acquisition module is also used for: The protein sequence is divided into protein sequence fragments of predetermined length.
6. The apparatus according to claim 4, characterized in that, The training module is also used for: From the gut microbiota database, protein sequences not associated with the predicted function are filtered out. These sequences are then used to reduce their gene sequence size using clustering tools, and representative sequences are selected using machine learning clustering algorithms to form a negative sample dataset. Protein sequences with known predicted functions are used as the positive sample dataset. This process also includes... The protein sequences not related to the predicted function and the protein sequences with the predicted function are segmented into protein sequence fragments of predetermined lengths. A prediction model is trained based on the negative sample dataset and the positive sample dataset.
7. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the artificial intelligence-based gene and gene cluster function prediction method as described in any one of claims 1-3.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the artificial intelligence-based gene and gene cluster function prediction method as described in any one of claims 1-3.