Method for transforming any polypeptide sequence into antibacterial peptide

Abstract

The invention discloses a method for transforming any polypeptide sequence into antibacterial peptide, and belongs to the technical field of biological information. According to the method, a hypergraph neural network predictor for predicting the antibacterial activity fraction of polypeptide is developed, a pre-trained protein language model is subjected to fine tuning training deviating to capture antibacterial peptide sequence features, and a strategy network for selecting amino acid sites to be mutated in a polypeptide sequence is trained; for any initial polypeptide sequence, firstly, a to-be-mutated amino acid site is selected through a strategy network, then the amino acid type of the site after mutation is given by a fine-tuned protein language model, then the antibacterial activity score of the sequence after mutation is predicted by a hypergraph neural network predictor, the process is iteratively executed, and the antibacterial activity score of the sequence after mutation is predicted by a hypergraph neural network predictor. And finally, ideal antibacterial peptide sequence output is obtained. The method has the capability of transforming any polypeptide sequence into the antibacterial peptide, the antibacterial activity score of the transformed polypeptide is greatly improved, and the method can be applied to actual development of antibacterial drugs and lays a foundation for solving drug-resistant bacterial infection.

Description

Technical Field

[0001] The present invention relates to a method for transforming any polypeptide sequence into an antimicrobial peptide, and specifically to a method for transforming any polypeptide sequence into an antimicrobial peptide based on deep learning, and belongs to the field of bioinformatics technology. Background Art

[0002] Antimicrobial peptides are a class of short peptides that have a broad-spectrum inhibitory effect on bacteria and other pathogens. They mainly exert their antibacterial effect by destroying the bacterial cell membrane. With the development of deep learning technology, the development of computational models provides a more efficient way to design new antimicrobial peptides. Although the number of antimicrobial peptides included in the existing database has reached 30,000, compared with the theoretically vast sequence space, known antimicrobial peptides are still very rare, and not all of them have high antibacterial activity. The existing methods for designing new antimicrobial peptides can be divided into two types: modification methods to improve the activity of known antimicrobial peptides and generation methods for designing antimicrobial peptides from scratch. Due to the limited ability of the optimization algorithm, the modification method usually needs to start from the known antimicrobial peptide sequence, and modify the antimicrobial peptide sequence through one or more rounds of iterations under the guidance of the fitness function to improve its antimicrobial activity. The generation method is to train the generator with known antimicrobial peptide data, input random noise, and obtain a series of antimicrobial peptide sequences through the conversion function. Although the HydrAMP model (Nat Commun 2023, 14: 1453) is a generative method that can accomplish both unconstrained generation and analog generation by inputting known antimicrobial peptide sequences, it requires independent model settings when used. Currently, no model is capable of completing both tasks with the same settings under a unified framework. Summary of the invention

[0003] In response to the above problems, the present invention provides a method for transforming any polypeptide sequence into an antimicrobial peptide based on a deep neural network, so as to more efficiently obtain antimicrobial peptides with high antimicrobial activity.

[0004] like Figure 1 As shown, the method of transforming any polypeptide sequence into an antimicrobial peptide of the present invention comprises the following steps:

[0005] A. Develop a hypergraph neural network predictor to predict the antimicrobial activity scores of peptides;

[0006] B. Fine-tune the pre-trained protein language model using known antimicrobial peptide sequence data to make it more inclined to capture the characteristics of antimicrobial peptide sequences;

[0007] C. Train a strategy network to select amino acid sites to be mutated in the polypeptide sequence;

[0008] D. Transforming any polypeptide sequence into an antimicrobial peptide, including:

[0009] D1. Input any starting polypeptide sequence, and select the amino acid site to be mutated by the strategy network trained in step C;

[0010] D2. The protein language model fine-tuned in step B gives the amino acid type after the mutation at the site, and obtains the complete sequence after the mutation;

[0011] D3. The hypergraph neural network predictor developed in step A predicts the antibacterial activity score of the mutated sequence;

[0012] D4. Iterate steps D1, D2, and D3 until the ideal sequence output is obtained.

[0013] In the above method for transforming any polypeptide sequence into an antimicrobial peptide, the specific method of developing a hypergraph neural network predictor in step A is:

[0014] A1. Collect a data set including antimicrobial peptide sequences, non-antimicrobial peptide sequences and the antimicrobial activity scores corresponding to each sequence, and randomly divide them into training set, validation set and test set;

[0015] A2. Build a hypergraph neural network predictor, including:

[0016] A2-1. Each peptide sequence is divided into segments (lengths of 2, 3, and 4 amino acids) using a sliding window, and hypergraphs are constructed for the three different segment lengths of the peptides. The amino acid residues of the whole polypeptide are nodes V, the segmented fragments are hyperedges E, and the features of each residue output by the protein language model constitute the node feature matrix X;

[0017] A2-2. Extract feature Y from the peptide hypergraph through hypergraph convolution operation. The calculation method is:

[0018]

[0019] Where X is the node feature matrix of the hypergraph, H is the association matrix of the hypergraph, W is the hyperedge feature matrix (unless otherwise specified, W is a unit matrix), and D v and D e are the diagonal matrices of node and edge degrees, respectively, and Θ is a parameter learned during training. Features Y2, Y3, and Y4 are obtained from peptide hypergraphs of three different fragment lengths (2, 3, and 4 amino acids, respectively).

[0020] A2-3. The three hierarchical features Y2, Y3, and Y4 outputted from step A2-2 are merged to obtain the final feature Y of the polypeptide sequence. c , the final feature Y c Input into a fully connected neural network to predict the antibacterial activity score r of the peptide:

[0021] r=dense(Y c )

[0022] Among them, dense(·) represents the fully connected neural network calculation;

[0023] A3. Use the training set data in A1 to train the hypergraph neural network predictor constructed in A2, and the loss function (Loss) is the mean square error (MSE):

[0024]

[0025] For the i-th polypeptide, r i is the predicted antimicrobial activity score output from step A2-3, y i is the known antimicrobial activity score contained in the data set. The goal of training is to reduce the difference between the predicted antimicrobial activity score and the known antimicrobial activity score label so that the hypergraph neural network predictor can accurately predict the antimicrobial activity score of the polypeptide sequence. The validation set is used to select the predictor parameters, and the test set is used to test the predictor performance.

[0026] In the above method for transforming any polypeptide sequence into an antimicrobial peptide, the strategy network in step C is a fully connected neural network with a parameter of θ, and the method for training the strategy network is:

[0027] C1. Randomly generate a series of peptide sequences with lengths between 10 and 40 amino acids as training data sets, and randomly initialize the strategy network parameters θ;

[0028] C2. Convert the polypeptide sequence into a digital code X aa Input the strategy network and output the probability prob of each residue being predicted as a site to be mutated;

[0029] prob = softmax(X aa θ)

[0030] C3. Obtaining a specific amino acid site a to be mutated according to the probability sampling output in step C2;

[0031] a=multinomial(prob)

[0032] Among them, multinomial (·) represents the sampling operation;

[0033] C4. Replace the amino acid site a to be mutated outputted in step C3 with a mask character, input it into the language model fine-tuned in step B, obtain the mutated polypeptide sequence s, and then input it into the hypergraph neural network predictor developed in step A to obtain the antibacterial activity score r;

[0034] C5. Iterate steps C2, C3, and C4 T times to obtain a trajectory τ: (s1, a1, r1, …, s T , a T , r T ), where s is the polypeptide sequence obtained in step C4, a is the site to be mutated obtained in step C3, and r is the antibacterial activity score obtained in step C4; using the trajectory τ, the strategy network parameter θ is updated using the REINFORCE algorithm.

[0035] Preferably, the data set in step A1 comprises antimicrobial peptide sequences and their logMIC antimicrobial activity values ​​from the GRAMPA database, and the conversion method between logMIC and antimicrobial activity score y is y=1 / (1+10 logMIC-2 ). The data set also contains the same number of non-antimicrobial peptide sequences as the antimicrobial peptide sequences, which are selected from the Uniprot database, and the antimicrobial activity score label is filled with 0. In one embodiment of the present invention, the ratio of the training set, the validation set, and the test set is 8:1:1.

[0036] Preferably, the protein language model used to extract node features in step A2-1 is the Ankh model.

[0037] Specifically, the feature fusion operation in step A2-3 may include concatenation, cross product, Hadamard product, Kronecker product, and addition.

[0038] Preferably, the known antimicrobial peptide sequence data in step B are derived from 11 antimicrobial peptide databases, including APD3, DRAMP, DBAASP, SATPdb, XUAMP, ADAM, AMPfun, LAMP, CAMP, dbAMP, and YADAMP, and the CD-HIT tool is used to remove redundancy at a threshold of 0.6-0.9 to retain representative antimicrobial peptide sequences.

[0039] Preferably, the pre-trained protein language model in step B is the Ankh model, which is a T5 architecture model consisting of an encoder-decoder and is pre-trained on the UniRef50 dataset.

[0040] Specifically, the encoding conversion method of step C2 may include one-hot encoding, digital mapping encoding that corresponds each type of amino acid letter to a number, feature encoding including amino acid composition frequency and physical and chemical properties, and encoding using a pre-trained language model.

[0041] Preferably, the number of amino acid sites to be mutated outputted each time in step C3 is 1, and the number of iterations T in step C5 is 3-15.

[0042] Specifically, the criterion for "obtaining an ideal sequence output" in step D4 can be that the antibacterial activity score of the iterative sequence predicted by the hypergraph neural network predictor reaches the target value, or the number of mutations of the iterative sequence compared to the starting sequence reaches the target value, or a certain physical and chemical property of the iterative sequence meets the target value.

[0043] The present invention also provides a system for transforming any polypeptide sequence into an antimicrobial peptide, which is used to transform the polypeptide sequence using the above-mentioned method for transforming any polypeptide sequence into an antimicrobial peptide, such as Figure 2 As shown, including:

[0044] I. A mutation amino acid site designation module, which is used to select the amino acid site to be mutated for the input polypeptide sequence using a strategy network;

[0045] II. A mutant amino acid decoding module, which is used to perform mutation on the amino acid site to be mutated specified by the mutant amino acid site specifying module using the fine-tuned pre-trained protein language model, and output the mutated amino acid type and the complete polypeptide sequence after the mutation;

[0046] III. Sequence evaluation module, used to score the antibacterial activity of the mutated polypeptide sequence.

[0047] The method provided by the present invention is capable of transforming any polypeptide sequence into an antimicrobial peptide. When the existing antimicrobial peptide, signal peptide and random sequence are transformed by the method of the present invention, the antimicrobial activity score of the transformed polypeptide can be greatly improved, and it has the potential to play a role as an antimicrobial peptide. Therefore, the present invention can contribute to the practical development of antimicrobial drugs and lay the foundation for solving drug-resistant bacterial infections. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the method for transforming any polypeptide sequence into an antimicrobial peptide according to the present invention.

[0049] Figure 2 It is a system composition diagram for transforming any polypeptide sequence into an antimicrobial peptide according to the present invention.

[0050] Figure 3 It is a network architecture diagram of the hypergraph neural network predictor developed by the present invention.

[0051] Figure 4 It is the comparison result of the method of transforming any polypeptide sequence into antimicrobial peptides of the present invention and other methods in the same field.

[0052] Figure 5The method of the present invention for transforming any polypeptide sequence into an antimicrobial peptide is the result of transforming existing antimicrobial peptides, signal peptides and random sequences into antimicrobial peptides. DETAILED DESCRIPTION

[0053] The present invention will be further described below through specific embodiments in conjunction with the accompanying drawings.

[0054] Example 1 A method for transforming any polypeptide into an antimicrobial peptide

[0055] like Figure 1 As shown, the present invention provides a method for transforming any polypeptide sequence into an antimicrobial peptide, comprising:

[0056] 1) Develop a hypergraph neural network predictor for predicting the antimicrobial activity scores of peptides.

[0057] 1.1) The dataset contains 3265 antimicrobial peptide sequences and their logMIC antimicrobial activity values ​​from the GRAMPA database, and 3265 non-antimicrobial peptide sequences from the Uniprot database. logMIC-2 ) The antimicrobial activity score y of the antimicrobial peptide sequence was calculated and used as the label, and the labels of all non-antimicrobial peptide sequences were filled with 0. The data were randomly divided into a training set (5,224 items), a validation set (653 items), and a test set (653 items) in a ratio of 8:1:1.

[0058] 1.2) The architecture of the hypergraph neural network predictor is as follows Figure 3 shown.

[0059] Each peptide sequence was divided into segments (lengths of 2, 3, and 4 amino acids) using a sliding window, and hypergraphs were constructed for the three different segment lengths of the peptides. The amino acid residues of the entire polypeptide are nodes V, the segmented fragments are hyperedges E, and the features of each residue of the polypeptide output from the protein language model Ankh constitute the node feature matrix X.

[0060] The three peptide hypergraphs with different fragment lengths are respectively subjected to hypergraph convolution operations to extract features Y2, Y3, and Y4. The calculation method is:

[0061]

[0062] Among them, X is the hypergraph node feature matrix, H is the hypergraph association matrix, W is the unit matrix representing the hyperedge feature, and D v and D e are the diagonal matrices of node and edge degrees respectively, and Θ is the parameter learned during training.

[0063] The three levels of features Y2, Y3, and Y4 are spliced ​​together to obtain the final feature Y of the peptide sequence.c ,

[0064] Y c =concat(Y2, Y3, Y4)

[0065] Among them, concat(·) represents the concatenation operation.

[0066] Input into a two-layer fully connected neural network to predict the antibacterial activity score r of the peptide:

[0067] r=σ(σ(W c1 Y c +b c1 )W c2 +b c2 )

[0068] Where σ(·) is the activation function ReLU, W c1 , b c1 is the first layer network parameter, W c2 , b c2 are the parameters of the second layer network.

[0069] 1.3) Use the training set data to train the hypergraph neural network predictor, and the loss function (Loss) is the mean square error MSE:

[0070]

[0071] where r i is the predicted antibacterial activity score output for the i-th peptide, y i is the known antimicrobial activity fraction in the dataset.

[0072] The optimizer used in training was Adam, the number of training iterations (Epoch) was set to 20, and the learning rate was finally selected as 0.0001 and the batch size was 128 based on the results on the reference validation set. After the training was completed, the performance of the model was evaluated on the test set, with a Pearson correlation coefficient of 0.923 and a root mean square error (RMSE) of 0.163, indicating that the hypergraph neural network developed in the present invention can accurately predict the antibacterial activity score of a polypeptide sequence, thereby being used to develop a method for transforming any polypeptide into an antimicrobial peptide.

[0073] 2) The pre-trained protein language model Ankh was fine-tuned using 7316 known antimicrobial peptide sequence data from 11 antimicrobial peptide databases including APD3, DRAMP, DBAASP, SATPdb, XUAMP, ADAM, AMPfun, LAMP, CAMP, dbAMP, and YADAMP, so that it tends to capture the characteristics of antimicrobial peptide sequences.

[0074] 3) Train a strategy network to select amino acid sites to be mutated in the polypeptide sequence.

[0075] 3.1) Randomly generate 1000 peptide sequences with a length between 10 and 40 amino acids, use CD-HIT to remove duplicate sequences with a similarity threshold of 0.7, and use the remaining 966 random sequences as the training data set. At the same time, randomly initialize the strategy network parameters θ;

[0076] 3.2) Convert the polypeptide sequence into digital code X through digital mapping coding aa , that is, the 20 amino acid letters are represented by corresponding integers 1-20. Input X aa to the policy network. The policy network is a two-layer fully connected neural network with a parameter of θ. The dimension of each layer is 128, which is used to output the probability prob of each residue being predicted as a site to be mutated;

[0077] prob = softmax(X aa θ)

[0078] 3.3) Sampling the specific amino acid site a to be mutated according to probability prob;

[0079] a=multinomial(prob)

[0080] Among them, multinomial(·) represents the sampling operation.

[0081] 3.4) The amino acid site a to be mutated outputted in the above step is replaced with a mask character and inputted into the protein language model fine-tuned in step 2 to obtain the mutated polypeptide sequence s, which is then inputted into the hypergraph neural network predictor developed in step 1 to obtain the antibacterial activity score r;

[0082] 3.5) Iterate steps 3.2-3.4 for a total of 8 times to obtain a trajectory τ: (s1, a1, r1, ..., s8, a8, r8). Use the trajectory τ to update the policy network parameters θ using the REINFORCE algorithm.

[0083] The above training process was performed on all 966 sequences for a total of 40 rounds to obtain a completed policy network.

[0084] 4) Using the above developed modules, any polypeptide sequence can be transformed into an antimicrobial peptide, including:

[0085] 4.1) Input the polypeptide sequence to be modified, and select the amino acid site to be mutated by the strategy network;

[0086] 4.2) The fine-tuned protein language model is used to give the amino acid type after the mutation at the site, thereby obtaining the complete sequence after the mutation;

[0087] 4.3) Predicting the antibacterial activity score of the mutated sequence by the hypergraph neural network predictor;

[0088] Iterate steps 4.1, 4.2, and 4.3 until the antimicrobial activity score of the mutated sequence reaches 0.9, then terminate the transformation and collect the output candidate antimicrobial peptide sequences.

[0089] Comparison of Example 2 with other methods

[0090] The method provided by the present invention performs better than other known methods in the art in the task of transforming known antimicrobial peptides to enhance their antimicrobial activity. The input known antimicrobial peptide data are 200 sequences with a minimum inhibitory concentration greater than 10 μM. Since other methods are sequence optimization methods, the hypergraph neural network predictor provided by the present invention is used as an optimization guide for scoring. For each method, all sequences generated during 8 rounds of optimization were collected, and the top 200 sequences with the highest predicted antimicrobial activity scores were retained after the optimization. Six widely used antimicrobial peptide classifiers, AMPScannerV2, CAMPR4-RF, CAMPR4-SVM, CAMPR4-ANN, MACREL, and ampir, were used to predict the probability that these sequences are antimicrobial peptides, and the average predicted value of each sequence was calculated and its distribution was plotted as shown in the figure. Figure 4 As shown, the predicted probability distribution of the evolved sequences of the present invention is higher than that of other methods.

[0091] Example 3: Transforming any polypeptide into an antimicrobial peptide

[0092] The method of the present invention can be used to transform any polypeptide sequence into an antimicrobial peptide. The following are examples of transforming existing antimicrobial peptides, signal peptides and random sequences into antimicrobial peptides. The input sequences are 200 known antimicrobial peptides with a minimum inhibitory concentration greater than 10 μM, 708 bacterial signal peptides, and 200 randomly generated polypeptides. Considering the difficulty of synthesis and the potential impact of disulfide bonds on the activity of antimicrobial peptides, the starting polypeptides selected in this example exclude sequences containing cysteine ​​(C). The above sequences are input into this method for 8 rounds of iterations, a total of 10 parallel transformations. After each input sequence is transformed, the sequence with the best antimicrobial activity score predicted by the hypergraph neural network module is collected, and its score distribution is as follows Figure 5 Compared with the antimicrobial activity scores of the input starting sequences, the three groups of different peptides all achieved significant improvements in antimicrobial activity scores. Furthermore, the top 10 candidate antimicrobial peptides with the highest scores after modification were selected from the three groups of starting peptides for experimental testing of antimicrobial activity. The names and sequences of the 30 antimicrobial peptides are shown in Table 1.

[0093] Table 1. Sequences of 30 candidate antimicrobial peptides

[0094]

[0095]

[0096] The bacterial species used in the minimum inhibitory concentration determination experiment include two standard strains of Gram-negative bacteria and two standard strains of Gram-positive bacteria, namely Escherichia coli ATCC25922, Pseudomonas aeruginosa PAO1, Staphylococcus aureus ATCC6538, and Bacillus subtilis ATCC6633. The Hancock microbroth dilution method was used to determine the minimum inhibitory concentration of the candidate antimicrobial peptides. The freeze-dried strains of the four standard strains were revived, activated by overnight culture, and streaked onto non-resistant LB (Haibo Biotechnology, #HB0129-2) plates for culture. Four single colonies with similar morphology were picked and placed in 5 mL LB medium (Haibo Biotechnology, #HB0218), cultured at 37°C in a shaker at 220 rpm for about 3 hours, and then diluted to make the OD 625 The values ​​are in the range of 0.08-0.13.

[0097] The candidate antimicrobial peptide powder was synthesized by Wuhan Dangang Biotechnology Co., Ltd. through solid phase chemical synthesis, and the molecular weight was correct by mass spectrometry. The antimicrobial peptide samples were purified by high performance liquid chromatography, and the purity was above 95%. 0.1 mg of each candidate antimicrobial peptide powder was dissolved in sterile PBS buffer (pH 7.4) to prepare a candidate antimicrobial peptide stock solution with a concentration of 512 μM.

[0098] Sterilized polypropylene 96-well culture plates (Greiner, #655201) were used for the determination. Three rows of parallel controls were set for each candidate antimicrobial peptide, with columns 1-10 containing bacterial solution and candidate antimicrobial peptide solution after gradient dilution, column 11 containing growth control, i.e., no antimicrobial peptide solution was added, and column 12 containing sterile control, i.e., only culture medium was added. First, 100 μL of LB culture medium was added to each of columns 1-11, and 200 μL of LB culture medium was added to column 12. After the candidate antimicrobial peptide stock solution was diluted 8 times with LB culture medium, 100 μL was added to each of the wells in column 1, 100 μL was transferred to column 2 and mixed, and 100 μL was transferred from column 2 to column 3 and mixed, and this step was repeated in sequence to complete the double gradient dilution (concentrations were 128, 64, 32, 16, 8, 4, 2, 1, 0.5, and 0.25 μM, respectively), and 100 μL was discarded after mixing in column 10. Take 400 μL of the adjusted bacterial solution and dilute it 100 times in 39.6 mL of LB medium, and add 100 μL of bacterial solution to each well of columns 1-11. Place the 96-well culture plate in a 37°C incubator and culture for 16 hours. Directly observe the bacterial growth in each well of the cell culture plate. The lowest antimicrobial peptide solution concentration value at which no obvious bacterial growth is observed by the naked eye is the minimum inhibitory concentration value of the antimicrobial peptide under this condition. The results are shown in Table 2.

[0099] Table 2. Minimum inhibitory concentrations of 30 candidate antimicrobial peptides against four strains (in μM)

[0100]

[0101]

[0102] Taking the minimum inhibitory concentration of at least one strain as less than or equal to 128 μM as the standard for antimicrobial peptides, in the above table, Perceptide01-10 is modified from known antimicrobial peptides, still maintains antimicrobial activity, and has been improved to a certain extent compared with before the modification. 8 of the 10 peptides Perceptide11-20 modified from signal peptides are antimicrobial peptides, and 6 of the 10 peptides Perceptide21-30 modified from random sequences are antimicrobial peptides. This shows the effectiveness of the method provided by the present invention to transform sequences from different sources into antimicrobial peptides, which can be used in a wider range of antimicrobial peptide modification designs to obtain antimicrobial peptides with ideal properties.

Claims

1. A method for transforming any polypeptide sequence into an antimicrobial peptide, comprising the following steps: A. Develop a hypergraph neural network predictor for predicting the antimicrobial activity scores of peptides; B. Fine-tune the pre-trained protein language model using known antimicrobial peptide sequence data to make it more inclined to capture the characteristics of antimicrobial peptide sequences; C. training a strategy network for selecting amino acid sites to be mutated in a polypeptide sequence; D. Transforming any polypeptide sequence into an antimicrobial peptide, including: D1. Input any starting polypeptide sequence, and select the amino acid site to be mutated by the strategy network trained in step C; D2. The protein language model fine-tuned in step B gives the amino acid type after the mutation at the site, and obtains the complete sequence after the mutation; D3. The hypergraph neural network predictor developed in step A predicts the antibacterial activity score of the mutated sequence; D4. Iterate steps D1, D2, and D3 until the ideal antimicrobial peptide sequence output is obtained.

2. The method according to claim 1, characterized in that Step A: The method for developing a hypergraph neural network predictor is: A1. Collect a data set including antimicrobial peptide sequences, non-antimicrobial peptide sequences and the antimicrobial activity scores corresponding to each sequence, and randomly divide them into training set, validation set and test set; A2. Build a hypergraph neural network predictor, including: A2-1. Each peptide sequence is divided into segments with sliding windows of 2, 3, and 4 amino acids in length, and hypergraphs are constructed for the peptides after segmentation with three different segment lengths. Among them, each amino acid residue of the whole polypeptide is a node V, the segmented fragments are hyperedges E, and the features of each residue output by the protein language model of the polypeptide constitute the node feature matrix X; A2-2. Extract feature Y from the peptide hypergraph through hypergraph convolution operation. The calculation method is: Among them, X is the node feature matrix of the hypergraph, H is the association matrix of the hypergraph, W is the hyperedge feature matrix, and D v and D e are the diagonal matrices of node and edge degrees, respectively, and Θ is the parameter learned during training. Features Y2, Y3, and Y4 are obtained from three peptide hypergraphs with different fragment lengths. A2-3. The three hierarchical features Y2, Y3, and Y4 outputted from step A2-2 are merged to obtain the final feature Y of the polypeptide sequence. c , the final feature Y c Input into a fully connected neural network to predict the antibacterial activity score r of the peptide: r=dense(Y c ) Among them, dense(·) represents the fully connected neural network calculation; A3. Use the training set data in A1 to train the hypergraph neural network predictor constructed in A2, and the loss function Loss is the mean square error MSE: For the i-th polypeptide, r i is the predicted antimicrobial activity score output from step A2-3, y i is the known antimicrobial activity score contained in the data set; the training goal is to reduce the difference between the predicted antimicrobial activity score and the known antimicrobial activity score label, so that the hypergraph neural network predictor can accurately predict the antimicrobial activity score of the polypeptide sequence; the validation set is used to select the hypergraph neural network predictor parameters, and the test set is used to test the performance of the hypergraph neural network predictor.

3. The method according to claim 2, characterized in that The data set in step A1 contains the antimicrobial peptide sequences and their logMIC antimicrobial activity values ​​from the GRAMPA database. The conversion method between logMIC and antimicrobial activity score y is y = 1 / (1+10 logMIC -2 ); the dataset also contains the same number of non-antimicrobial peptide sequences from the Uniprot database as the antimicrobial peptide sequences, and the antimicrobial activity score labels are filled with 0.

4. The method according to claim 2, characterized in that The protein language model used to extract node features in step A2-1 is the Ankh model, and the feature fusion operation in step A2-3 is selected from one or more of concatenation, cross product, dot product, Kronecker product, and addition.

5. The method according to claim 1, characterized in that In step B, the known antimicrobial peptide sequence data are derived from 11 antimicrobial peptide databases, including APD3, DRAMP, DBAASP, SATPdb, XUAMP, ADAM, AMPfun, LAMP, CAMP, dbAMP, and YADAMP, and the CD-HIT tool is used to remove redundancy with a threshold of 0.6-0.9 to retain representative antimicrobial peptide sequences; the pre-trained protein language model is the Ankh model.

6. The method according to claim 1, characterized in that The policy network described in step C is a fully connected neural network with a parameter of θ. The method for training the policy network is: C1. Randomly generate a series of peptide sequences with lengths between 10 and 40 amino acids as training data sets, and randomly initialize the strategy network parameters θ; C2. Convert the polypeptide sequence into a digital code X aa Input the strategy network and output the probability prob of each residue being predicted as a site to be mutated; prob=softmax(X aa θ) C3. Obtaining a specific amino acid site a to be mutated according to the probability sampling output in step C2; a=multinomial(prob) Among them, multinomial (·) represents the sampling operation; C4. Replace the amino acid site a to be mutated outputted in step C3 with a mask character, input it into the language model fine-tuned in step B, obtain the mutated polypeptide sequence s, and then input it into the hypergraph neural network predictor developed in step A to obtain the antibacterial activity score r; C5. Iterate steps C2, C3, and C4 T times to obtain a trajectory τ: (s1, a1, r1, ..., s T , a T , r T ), where s is the polypeptide sequence obtained in step C4, a is the site to be mutated obtained in step C3, and r is the antibacterial activity score obtained in step C4; using the trajectory τ, the strategy network parameter θ is updated using the REINFORCE algorithm.

7. The method according to claim 6, characterized in that The encoding conversion method of step C2 is selected from one of the following methods: one-hot encoding, digital mapping encoding of each type of amino acid letter corresponding to a number, including feature encoding of amino acid composition frequency and physical and chemical properties, and encoding using a pre-trained language model.

8. The method according to claim 1, characterized in that The standard of the ideal antimicrobial peptide sequence in step D4 is that the antimicrobial activity score of the iterative sequence predicted by the hypergraph neural network predictor reaches the target value, or the number of mutations of the iterative sequence compared with the starting sequence reaches the target value, or a certain physical and chemical property of the iterative sequence meets the target value.

9. A system for transforming any polypeptide sequence into an antimicrobial peptide, for implementing the method for transforming any polypeptide sequence into an antimicrobial peptide according to any one of claims 1 to 8, comprising: I. A mutation amino acid site designation module, which is used to select the amino acid site to be mutated for the input polypeptide sequence using a strategy network; II. A mutant amino acid decoding module, which is used to perform mutation on the amino acid site to be mutated specified by the mutant amino acid site specifying module using the fine-tuned pre-trained protein language model, and output the mutated amino acid type and the complete polypeptide sequence after the mutation; III. Sequence evaluation module, used to score the antibacterial activity of the mutated polypeptide sequence.

Images