A method and device for recommending mutable sites based on small-sample experimental data

Abstract

The present invention discloses a method and device for recommending mutatable sites from small-sample experimental data, comprising the following steps: obtaining small-sample experimental data, including the sequences and data of wild enzymes and mutants, and obtaining the optimal mutant sequence according to the data type of the mutants; predicting the mutant structure based on the optimal mutant sequence; predicting the substrate-binding pocket based on the mutant structure; predicting single mutation sites based on the protein structure model, and selecting residues with a distance from the center of the substrate-binding pocket less than a first threshold as the recommended single mutation sites; selecting sites from the small-sample experimental data, mutating each site in the site set into 19 other amino acids, and pairwise combining them to construct a double mutation set, predicting the mutation results, and obtaining the recommended double mutation sites according to the sorting results; selecting sites from the small-sample experimental data, obtaining the coordinates of the sites in the variant structure for clustering analysis, and selecting 1 site from each cluster to combine with other clusters to construct multi-mutations; predicting the mutation results of the multi-mutations, and obtaining the recommended multi-mutation sites according to the sorting results; the method of the present invention operates effectively under the condition of small-sample data; through the powerful generalization ability of the large language model, the structural and functional information in biomolecules can be captured, so as to effectively encode the sequence of the enzyme for more accurate inference.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics technology, and particularly to a method and device for recommending mutable sites from small sample experimental data. Background Art

[0002] Enzymes are biological catalysts composed of proteins or RNAs that can accelerate chemical reactions in organisms. By lowering the activation energy of the reaction, enzymes achieve efficient catalysis under milder conditions and have high specificity for substrates. They are widely present in organisms and regulate almost all life processes.

[0003] The limitations of wild enzymes are mainly reflected in their catalytic efficiency, stability, and substrate specificity. They usually can only work efficiently under the natural conditions of organisms and are sensitive to temperature, pH, and other environmental factors. In addition, the substrate range of wild enzymes is limited, and they often cannot meet the wide reaction requirements under industrial or experimental conditions.

[0004] Enzyme modification optimizes the performance of enzymes through techniques such as directed evolution or rational design, but faces many challenges. Especially when recommending mutation sites from small sample experimental data, the problem is more prominent. The structure of enzymes is complex, and a small number of mutations may lead to unpredictable changes in global performance. Due to the synergistic or antagonistic effects of enzyme mutation effects, it is difficult to accurately measure the global impact of mutations based on small-scale data. In addition, when the data samples are insufficient, the complex relationship between enzyme function and structure may not be captured, thus limiting the effect of mutation site optimization. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for recommending mutable sites from small sample experimental data to overcome the deficiencies in the prior art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] The present application discloses a method for recommending mutable sites from small sample experimental data, including the following steps:

[0008] S1. Obtain small sample experimental data, where the small sample experimental data includes the sequences and data of wild enzymes and mutants, obtain the optimal mutant sequence according to the data type of the mutants; predict the mutant structure according to the optimal mutant sequence; predict the substrate binding pocket according to the mutant structure;

[0009] S2. Predict single mutation sites based on the protein structure model to obtain a prediction result, and select the residues with a distance less than a first threshold from the center of the substrate binding pocket in step S1 as the recommended single mutation sites;

[0010] S3. Train a neural network model through a protein large language model based on the small sample experimental data in step S1;

[0011] S4. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are better than the wild enzyme data in the small sample experimental data, and the sites within a distance less than a second threshold from the center of the substrate binding pocket in step S1 to obtain a set of sites. Mutate each site in the set of sites into 19 other amino acids and pairwise combine them to construct a double mutant set. Use the neural network model in step S3 to predict the mutation results, sort according to the predicted results, and obtain the recommended double mutant sites according to the sorting results;

[0012] S5. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are better than the wild enzyme data in the small sample experimental data, obtain the coordinates of the sites in the mutant structure in step S1 for clustering analysis, and obtain several clusters;

[0013] S51. Select 1 site from each cluster and combine it with other clusters to construct a multi-mutation; use the neural network model in step S3 to predict the mutation results of the multi-mutation, sort according to the predicted results, and obtain the recommended multi-mutation sites according to the sorting results;

[0014] S52. Select the mutations with at least a 1.5-fold improvement compared to the wild enzyme from each cluster according to the sorted mutation results. The mutations selected from each cluster are combined with the mutations selected from other clusters as the recommended multi-mutation sites.

[0015] Preferably, the small sample experimental data is given by biological experimenters or obtained from a public mutation dataset.

[0016] Preferably, the data type includes enzyme activity data, ligand binding affinity data, or temperature tolerance data; the method for obtaining the optimal mutant sequence according to the data type of the mutants in step S1 is as follows: sort the data from good to bad according to the data type of the mutants, and select the sequence of the mutant with the best data.

[0017] Preferably, in step S1, AlphaFold, chai_lab, or helixfold3 is used to predict the protein structure corresponding to the optimal mutant sequence to obtain the mutant structure.

[0018] Preferably, in step S1, p2rank is used to predict the substrate binding pocket of the mutant structure.

[0019] Preferably, in step S2, the first threshold is taken as 15 angstroms.

[0020] Preferably, step S3 specifically includes the following steps:

[0021] S31. Employ the ESM or ProtTrans protein large language model;

[0022] S32. Through the protein large language model, convert the sequences in the small-sample experimental data into numbers as the training feature X, use the mutation result as Y, train the neural network model, and calculate the correlation between the mutation result in the test set and the predicted mutation result;

[0023] S33. Obtain the homologous sequences in the small-sample experimental data to fine-tune the protein large language model, train the neural network model, and calculate the correlation between the mutation result in the test set and the predicted mutation result;

[0024] S34. Compare the correlation results of S32 and S33, and select the neural network model with a larger correlation as the final neural network model.

[0025] Preferably, in step S4, the second threshold is taken as 15 Å, the mutation results are sorted from large to small, and top N is selected according to the sorting result as the recommended double mutation sites.

[0026] Preferably, in step S51, the mutation results are sorted from large to small, and top N is selected according to the sorting result as the recommended multi-mutation sites.

[0027] The present invention also discloses a device for recommending mutable sites from small-sample experimental data, including a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used for the method of recommending mutable sites from small-sample experimental data as described above.

[0028] Advantages of the present invention:

[0029] Small-sample adaptability: Traditional methods often require a large amount of experimental data to capture the complex relationship between enzyme function and structure, while the method of the present invention can also operate effectively under small-sample data conditions; Through the powerful generalization ability of the large language model, the structural and functional information in biomolecules can be captured, so as to effectively encode the sequences of enzymes for more accurate inference.

[0030] Efficient mutation screening: The method of the present invention can quantify the impact of mutations on enzyme performance through a neural network, significantly improving the efficiency of mutant site screening; Compared with traditional experimental exploration, the method of the present invention can propose more promising mutation combinations in a short time, thus reducing the blindness of experiments.

[0031] Rapid experimental verification: Through the predictive ability of the model, the activity of the target enzyme was increased by more than 6 times within 1 week of calculation time and 2 weeks of experimental time. This ability of rapid feedback not only has theoretical advantages in the method of the present invention, but also can show superior effects in actual experimental verification;

[0032] The features and advantages of the present invention will be described in detail through examples in conjunction with the accompanying drawings. Brief Description of the Drawings

[0033] Figure 1 is a schematic flow chart of a method for recommending mutable sites by small sample experimental data according to the present invention;

[0034] Figure 2 is the pocket information of the substrate binding pocket in the embodiment of the present invention;

[0035] Figure 3 is the result of predicting the mutation sites of the protein structure model in the embodiment of the present invention;

[0036] Figure 4 is the result of selecting residues within 15 Å from the pocket center in the embodiment of the present invention;

[0037] Figure 5 is the result of double mutation sites in the embodiment of the present invention;

[0038] Figure 6 is the clustering result in the embodiment of the present invention;

[0039] Figure 7 is the prediction result of step S51 in the embodiment of the present invention;

[0040] Figure 8 is all the mutation recommendation results in the embodiment of the present invention;

[0041] Figure 9 is the verification of the activity of the mutant in the wet experiment in the embodiment of the present invention;

[0042] Figure 10 is a schematic structural diagram of a device for recommending mutable sites by small sample experimental data according to the present invention; Detailed Embodiments

[0043] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below through the accompanying drawings and examples. However, it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0044] As Figure 1As shown in the figure, a method for recommending mutatable sites from small sample experimental data according to the present invention includes the following steps:

[0045] S1. Obtain small sample experimental data, where the small sample experimental data includes the sequences and data of wild enzymes and mutants, and obtain the optimal mutant sequence according to the data type of the mutants; predict the mutant structure based on the optimal mutant sequence; predict the substrate binding pocket based on the mutant structure;

[0046] S2. Predict single mutant sites based on the protein structure model to obtain a prediction result, and select the residues with a distance less than a first threshold from the center of the substrate binding pocket in step S1 in the prediction result as the recommended single mutant sites;

[0047] S3. Based on the small sample experimental data in step S1, train a neural network model through a protein large language model;

[0048] S4. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are better than the wild enzyme data in the small sample experimental data, and the sites with a distance less than a second threshold from the center of the substrate binding pocket in step S1 to obtain a site set. Mutate each site in the site set into 19 other amino acids and construct a double mutant set by pairwise combination. Use the neural network model in step S3 to predict the mutation results, sort according to the predicted results, and obtain the recommended double mutant sites according to the sorting results;

[0049] S5. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are better than the wild enzyme data in the small sample experimental data, obtain the coordinates of the sites in the mutant structure in step S1 for clustering analysis, and obtain several clusters;

[0050] S51. Select 1 site from each cluster to combine with other clusters to construct a multi-mutation; use the neural network model in step S3 to predict the mutation results of the multi-mutation, sort according to the predicted results, and obtain the recommended multi-mutation sites according to the sorting results;

[0051] S52. Select the mutations with at least a 1.5-fold improvement compared to the wild enzyme according to the mutation results sorted from each cluster. The mutations selected from each cluster are combined with the mutations selected from other clusters as the recommended multi-mutation sites.

[0052] In a feasible embodiment, the small sample experimental data is given by biological experimenters or obtained from a public mutation dataset.

[0053] In a feasible embodiment, the data type includes enzyme activity data, ligand binding affinity data, or temperature tolerance data; the method for obtaining the optimal mutant sequence according to the data type of the mutant in step S1 is as follows: according to the data type of the mutant, sort the data from best to worst, and select the sequence of the mutant with the best data.

[0054] In a feasible embodiment, in step S1, AlphaFold, chai_lab, or helixfold3 is used to predict the protein structure corresponding to the optimal mutant sequence to obtain the mutant structure.

[0055] In a feasible embodiment, in step S1, p2rank is used to predict the substrate binding pocket of the mutant structure.

[0056] In a feasible embodiment, in step S2, the first threshold is set to 15 angstroms.

[0057] In a feasible embodiment, step S3 specifically includes the following steps:

[0058] S31: Use the ESM or ProtTrans protein large language model;

[0059] S32: Convert the sequences in the small sample experimental data into numbers as the training feature X through the protein large language model, use the mutation result as Y, train the neural network model, and calculate the correlation between the mutation result in the test set and the predicted mutation result;

[0060] S33: Obtain the homologous sequences in the small sample experimental data to fine-tune the protein large language model, train the neural network model, and calculate the correlation between the mutation result in the test set and the predicted mutation result;

[0061] S34: Compare the correlation results of S32 and S33, and select the neural network model with a larger correlation as the final neural network model.

[0062] In a feasible embodiment, in step S4, the second threshold is set to 15 angstroms, the mutation results are sorted from largest to smallest, and topN is selected according to the sorting result as the recommended double mutation sites.

[0063] In a feasible embodiment, in step S51, the mutation results are sorted from largest to smallest, and topN is selected according to the sorting result as the recommended multi-mutation sites.

[0064] Example:

[0065] Step 1: The mutation data of a certain enzyme is given by biological experimenters or obtained from a public mutation dataset (proteingym: https: / / proteingym.org / ).

[0066] Obtain the optimal mutant sequence from the data type of the mutants of a certain enzyme (if it is enzyme activity, select the sequence corresponding to the largest enzyme activity value; if it is binding affinity, select the sequence corresponding to the largest binding affinity; if it is temperature tolerance, select the sequence corresponding to the highest temperature).

[0067] Predict the mutant structure from the optimal mutant sequence; predict the substrate binding pocket from the mutant structure.

[0068] Tools that can predict protein structures include AlphaFold (https: / / golgi.sandbox.google.com / ), chai_lab(https: / / lab.chaidiscovery.com / dashboard); helixfold3(https: / / paddlehelix.baidu.com / app / all / helixfold3 / forecast). AlphaFold is selected in this invention to predict the structure; p2rank (https: / / github.com / rdk / p2rank) is used in this invention to predict the pocket, and the pocket information is as Figure 2 shown

[0069] Since the experimental result this time is enzyme activity data, the selected good sites below refer to the sites with good enzyme activity.

[0070] Step 2: Predict single mutation sites based on the protein structure model, and select the residues less than 15 angstroms away from the center of the predicted substrate binding pocket in S1 as the recommended single mutation sites

[0071] This invention uses a structure model (https: / / github.com / drorlab / gvp-pytorch) to predict single mutation sites, and part of the prediction results are as follows:

[0072] As Figure 3 shown, the structure model predicts mutation sites (a total of 367);

[0073] As Figure 4 shown, select the residues within 15 angstroms from the pocket center;

[0074] Step 3: Use a protein large language model to train a neural network based on small sample experimental data

[0075] 1: Protein large language model This invention uses ESM, specifically esm2_t33_650M_UR50D

[0076] 2: The homologous sequences for fine-tuning the large language model are obtained using Jackhmmer in the UniRef90 database (11,297 homologous sequences). The execution command is:

[0077] jackhmmer -o / dev / null -A nit.sto

[0078] --noali --F1 0.0005 --F2 5e-05 --F3 5e-07 --incE 0.0001 -E0.0001 --cpu 25

[0079] -N 1 uniref90.fasta

[0080] 3: The neural network uses gaussian_process from the sklearn package. The Gaussian kernel is a composite kernel, specifically WhiteKernel and RationalQuadratic

[0081] 4: The Pearsonr method from the scipy package is used for relevance calculation

[0082] Do not fine-tune the protein large language model

[0083] Model score: 0.37; Relevance between the test set and experimental results: 0.52

[0084] Fine-tune the protein large language model

[0085] Model score: 0.48; Relevance between the test set and experimental results: 0.69

[0086] Therefore, the fine-tuned protein large language model is selected to train the Gaussian model;

[0087] Step 4: Select sites with improved performance compared to the wild enzyme and sites within 15 Å of the center of the substrate-binding pocket from the mutation data, mutate each site to 19 different amino acids, combine them in pairs, and use the neural network model in Step 3 to predict the mutation results. Sort the predicted mutation results from largest to smallest and select the top 5 as the recommended double-mutation sites.

[0088] As Figure 5 shown, the total number of double-mutation combinations is 267,501. Sort and remove duplicate sites, and select the top 5;

[0089] Step 5: Select sites with improved performance compared to the wild enzyme from the mutation data, and obtain the coordinates of these sites from the predicted structures in S1 for clustering analysis, asFigure 6 As shown, the structural clustering results;

[0090] Step 5 (1): Select 1 locus from each cluster and combine it with other clusters to construct multiple mutations. Use the neural network model in Step 3 to predict the mutation results of the multiple mutants, and sort them from largest to smallest according to the predicted mutation results, and select the top 5 as the recommended multiple mutation sites; Model prediction, sorting and selecting multiple mutation combinations as Figure 7 shown;

[0091] Step 5 (2): From the clustering results, select the mutations with at least a 1.5-fold improvement in the mutation results compared to the wild type from each cluster according to the mutation results. The mutations selected from each cluster are combined with the mutations selected from other clusters as the recommended multiple mutation sites.

[0092] Clustering results and experimental data

[0093] clust1=['K320S-1.10']

[0094] clust2=['V199H-1.20','L194M-1.08','A201Y-2.60','Q207N-1.02']

[0095] clust3=['K97N-1.18','D40G-1.05','I117L-1.12']

[0096] clust4=['M287L-1.79','A286T-1.23','A288H-2']

[0097] clust5=['A179C-2.09','A176C-2.13','M180Q-2.51']

[0098] Select the mutations with at least a 1.5-fold improvement compared to the wild type from each cluster and combine them as the recommended multiple mutation sites are:

[0099] K320S_A201Y_K97N_A288H_M180Q

[0100] A201Y_K97N_A288H_M180Q

[0101] A201Q_K97N_A288H_M180Q

[0102] A201Y_K97N_A288H_A179C

[0103] A201Y_K97N_M287L_M180Q

[0104] All the finally recommended mutations are as Figure 8 shown;

[0105] Next, wet experiments verify the activity of the recommended mutable sites.

[0106] The wet experiments verify the activity of the mutants as Figure 9 shown; where WT represents the wild protease; the abbreviation of at enzyme, L60I represents the mutation of the 60th amino acid of the protease from amino acid L to amino acid I [numbering starts from 1];

[0107] In the field of enzyme engineering, although traditional methods such as directed evolution and rational design have made significant progress, they still face many challenges in optimizing enzyme performance. Especially when recommending mutation sites only relying on small sample experimental data, the relationship between the complex structure and function of the enzyme often makes the mutation effect difficult to predict. Some mutations may lead to significant changes in global performance, and these changes are difficult to capture in small-scale data sets. In addition, the mutation effects of enzymes often have synergistic or antagonistic effects, and traditional methods are difficult to evaluate the impact of these interactions on overall activity or stability. This makes it more difficult to accurately optimize mutation sites in the case of limited data; To overcome these limitations, the present invention proposes an innovative method for recommending mutable sites from small sample experimental data. Compared with traditional methods, this method exhibits multiple advantages:

[0108] Small sample adaptability: Traditional methods often require a large amount of experimental data to capture the complex relationship between enzyme function and structure, while the method of the present invention can also operate effectively under small sample data conditions. Through the powerful generalization ability of the large language model, the structural and functional information in biomolecules can be captured, so as to effectively encode the enzyme sequence for more accurate inference.

[0109] Efficient mutation screening: The method of the present invention can quantify the impact of mutations on enzyme performance through neural networks, significantly improving the efficiency of screening mutation sites. Compared with traditional experimental exploration, the method of the present invention can propose more promising mutation combinations in a short time, thus reducing the blindness of experiments.

[0110] Rapid experimental verification: Through the prediction ability of the model, the present invention increased the activity of the target enzyme by more than 6 times within 1 week of computing time and 2 weeks of experimental time. This ability of rapid feedback makes the method of the present invention not only have theoretical advantages, but also show superior effects in actual experimental verification.

[0111] An embodiment of the apparatus for recommending mutable sites from small-sample experimental data according to the present invention can be applied to any device with data processing capabilities, and such a device with data processing capabilities can be a device or apparatus such as a computer. The apparatus embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically defined apparatus, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for running. In terms of the hardware level, as Figure 10 shown, it is a hardware structure diagram of any device with data processing capabilities where the apparatus for recommending mutable sites from small-sample experimental data according to the present invention is located. In addition to Figure 10 the processor, memory, network interface, and non-volatile memory shown, the device with data processing capabilities where the apparatus in the embodiment is located usually includes other hardware according to the actual functions of the device with data processing capabilities, which will not be elaborated here. The implementation processes of the functions and roles of each unit in the above apparatus are specifically described in the implementation processes of the corresponding steps in the above method, which will not be elaborated here.

[0112] For the apparatus embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The apparatus embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0113] The embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements the apparatus for recommending mutable sites from small-sample experimental data in the above embodiment.

[0114] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0115] The foregoing is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, or improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for recommending mutable sites from small-sample experimental data, characterized in that, It includes the following steps: S1. Obtain small-sample experimental data, where the small-sample experimental data includes the sequences and data of wild enzymes and mutants, and obtain the optimal mutant sequence according to the data type of the mutants; predict the mutant structure based on the optimal mutant sequence; predict the substrate-binding pocket based on the mutant structure; S2. Predict single mutation sites based on the protein structure model to obtain a prediction result, and select the residues with a distance less than threshold one from the center of the substrate-binding pocket in step S1 as the recommended single mutation sites in the prediction result; S3. Based on the small-sample experimental data in step S1, train a neural network model through a protein large language model; S4. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are superior to the wild enzyme data in the small-sample experimental data, and the sites with a distance less than threshold two from the center of the substrate-binding pocket in step S1 to obtain a site set. Mutate each site in the site set into 19 other amino acids and construct a double mutation set by pairwise combination. Use the neural network model in step S3 to predict the mutation results, sort according to the predicted results, and obtain the recommended double mutation sites according to the sorting results; S5. According to the data type of the mutants, select the sites corresponding to the mutant sequences that are superior to the wild enzyme data in the small-sample experimental data, obtain the coordinates of the sites in the mutant structure in step S1, and perform clustering analysis to obtain several clusters; S51. Select 1 site from each cluster and combine it with other clusters to construct a multi-mutation; use the neural network model in step S3 to predict the mutation results of the multi-mutation, sort according to the predicted results, and obtain the recommended multi-mutation sites according to the sorting results; S52. Select the mutations with a mutation result at least 1.5 times improved compared to the wild enzyme from each cluster according to the sorting of the mutation results. The mutations selected from each cluster are combined with the mutations selected from other clusters as the recommended multi-mutation sites.

2. The method for recommending mutable sites from small sample experimental data according to claim 1, wherein: The small-sample experimental data is given by biological experimenters or obtained from a public mutation dataset.

3. A method for recommending mutable sites from small sample experimental data as described in claim 1, characterized in that: The data type includes enzyme activity data, ligand-binding affinity data, or temperature tolerance data; the method for obtaining the optimal mutant sequence according to the data type of the mutants in step S1 is as follows: sort the data from best to worst according to the data type of the mutants, and select the sequence of the mutant with the best data.

4. A method for recommending mutable sites from small sample experimental data according to claim 1, characterized in that: In step S1, AlphaFold, chai_lab, or helixfold3 is used to predict the protein structure corresponding to the optimal mutant sequence to obtain the mutant structure.

5. A method for recommending mutable sites from small sample experimental data as described in claim 1, characterized in that: In step S1, p2rank is used to predict the substrate-binding pocket of the mutant structure.

6. The method for recommending mutable sites from small sample experimental data according to claim 1, wherein: In step S2, the threshold one is taken as 15 Å.

7. A method for recommending mutable sites from small sample experimental data as claimed in claim 1, characterized in that: Step S3 specifically includes the following steps: S31. Use the ESM or ProtTrans protein large language model; S32. Convert the sequences in the small-sample experimental data into numbers as the training feature X through the protein large language model, use the mutation result as Y, train a neural network model, and calculate the correlation between the mutation results in the test set and the predicted mutation results; S33. Obtain homologous sequences in the small-sample experimental data to fine-tune the protein large language model, train a neural network model, and calculate the correlation between the mutation results and the predicted mutation results in the test set; S34. Compare the correlation results of S32 and S33, and select the neural network model with a larger correlation as the final neural network model.

8. A method for recommending mutable sites from small-sample experimental data as claimed in claim 1, characterized in that: In step S4, the threshold two is taken as 15 Å, the mutation results are sorted from largest to smallest, and top N is selected as the recommended double mutation sites according to the sorting results.

9. A method for recommending mutable sites from small sample experimental data as claimed in claim 1, characterized in that: In step S51, the mutation results are sorted from largest to smallest, and top N is selected as the recommended multi-mutation sites according to the sorting results.

10. A device for recommending mutable sites based on small-sample experimental data, characterized in that: It includes a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement the method for recommending mutatable sites from small-sample experimental data according to any one of claims 1-9.

Images