Antibacterial peptide generation method, system, equipment and medium

Abstract

The invention discloses an antibacterial peptide generation method, system, equipment and medium, and relates to the technical field of antibacterial peptide generation, the method comprises the following steps: obtaining protein sequence data; inputting the protein sequence data into an antibacterial peptide generation model for training, and performing autoregression sampling on the protein sequence data through an LSTM module to generate an amino acid sequence; inputting the amino acid sequence into a Transform module, performing word segmentation according to characters, converting the amino acid sequence after word segmentation into an id sequence, and mapping the id sequence into an embedded vector; encoding the embedded vector through an encoder, and inputting global features extracted by encoding into a decoder to generate a new amino acid sequence; identifying and screening the antibacterial characteristics of the new amino acid sequence to obtain an antibacterial peptide sequence; inputting to-be-generated protein sequence data into the trained antibacterial peptide generation model to generate an antibacterial peptide sequence; the method improves the novelty and antibacterial property of the new polypeptide sequence.

Description

Technical Field

[0001] The present invention relates to the technical field of antimicrobial peptide production, and in particular to an antimicrobial peptide production method, system, equipment and medium. Background Art

[0002] Traditionally, new antimicrobial resistance peptides are discovered by extracting peptide molecules from animals and plants and then conducting wet experiments on certain antibacterial resistance. However, this method requires a lot of time and reagent costs, which seriously hinders the discovery of antimicrobial resistance peptides.

[0003] With the development of deep learning technology, generative models have demonstrated excellent performance in molecule generation and optimization, providing a new approach to addressing the traditional manual search for antimicrobial peptides (AMPs). Some researchers have employed LSTM-based models to generate new AMPs. PepGAN, proposed by some researchers, strikes a balance between covering active peptides and avoiding inactive ones. The AMP-GAN model, developed by others, utilizes an improved conditional generative adversarial network to design new AMPs. A conditional variational autoencoder, HydrAMP, has been proposed, which learns a low-dimensional, continuous peptide chemical space and captures its antimicrobial properties. While LSTM is suitable for training on small and medium-sized datasets, it exhibits overfitting on large datasets and has certain limitations on the length of molecular sequences. While LSTM offers advantages in terms of stability in molecule generation, it falls short in generating innovative and diverse molecules. In contrast, the Transformer model demonstrates superior performance on large datasets, exhibits little overfitting, and has no limitations on sequence length. Furthermore, the Transformer model outperforms LSTM in generating novel and diverse molecules, but requires a large number of long sequences, which is often difficult to achieve in practice.

[0004] In summary, the currently used LSTM and Transformer methods cannot generate AMPs with high novelty and antibacterial properties because the LSTM model itself cannot capture the long-range relationship between sequences. The Transformer method has stability defects in the generation process due to the lack of relevant large-scale data sets when generating AMPs. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, such as the inability to generate AMPs with high novelty and the defects in the stability of the generated AMPs, the present invention proposes a method, system, device and medium for generating antimicrobial peptides. By combining the advantages and disadvantages of the LSTM and Transformer methods, a LTAMP method is proposed, thereby solving the problems existing in the existing technology.

[0006] A method for producing an antimicrobial peptide comprises the following steps: Obtain protein sequence data; Inputting protein sequence data into an antimicrobial peptide generation model for training specifically includes: performing autoregressive sampling on the protein sequence data through an LSTM module to generate an amino acid sequence ID, and converting the generated amino acid sequence ID into an amino acid sequence; inputting the amino acid sequence into a Transformer module to perform character segmentation, converting the segmented amino acid sequence into an ID sequence, and mapping the ID sequence into an embedding vector; encoding the embedding vector through an encoder, inputting the global features extracted by the encoding into a decoder to generate a new ID sequence, and converting the new ID sequence into a new amino acid sequence; identifying and screening the antimicrobial properties of the new amino acid sequence to obtain an antimicrobial peptide sequence; The protein sequence data to be generated is input into the trained antimicrobial peptide generation model to generate an antimicrobial peptide sequence.

[0007] Furthermore, the method further includes preprocessing the protein sequence data after obtaining the protein sequence data; the preprocessing includes removing non-FASTA format characters and sequences of amino acid residues in the protein sequence data.

[0008] Furthermore, the antibacterial properties of the new amino acid sequences are identified and screened to obtain antimicrobial peptide sequences; specifically, the method includes using random forest RF and support vector machine SVM in QSAR modeling to functionally score the antibacterial properties of the new amino acid sequences and screen out sequences with QSAR scores greater than 0.5.

[0009] The present invention also includes an antimicrobial peptide production system, comprising: an acquisition unit, for acquiring protein sequence data; The model training unit is used to input protein sequence data into the antimicrobial peptide generation model for training, specifically including: performing autoregressive sampling on the protein sequence data through the LSTM module to generate an amino acid sequence ID, and converting the generated amino acid sequence ID into an amino acid sequence; inputting the amino acid sequence into the Transformer module to perform word segmentation by character, converting the amino acid sequence after word segmentation into an ID sequence, and mapping the ID sequence into an embedding vector; encoding the embedding vector through the encoder, inputting the global features extracted by the encoding into the decoder to generate a new ID sequence, and converting the new ID sequence into a new amino acid sequence; identifying and screening the antimicrobial properties of the new amino acid sequence to obtain an antimicrobial peptide sequence; The generation unit is used to input the protein sequence data to be generated into the trained antimicrobial peptide generation model to generate an antimicrobial peptide sequence.

[0010] The present invention also includes an antimicrobial peptide generation computer device, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the antimicrobial peptide generation method when executing the computer program.

[0011] The present invention also includes a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the steps of the antimicrobial peptide production method are executed.

[0012] The present invention provides a method for producing antimicrobial peptides, which has the following beneficial effects: The present invention proposes an LTAMP method by combining the respective advantages and disadvantages of the LSTM and Transformer methods. It combines the characteristics of LSTM that performs well in medium and short sequence lengths and the algorithm's performance does not improve as the scale of training data increases with the characteristics of the Transformer model that has no sequence length restrictions and is good at capturing relationships between long sequences, and its model performance will also improve accordingly as the scale of training data increases, to generate polypeptide sequences with different degrees of antibacterial properties; this method improves the novelty and antibacterial properties of the new polypeptide sequences. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 Schematic diagram of a process for producing antimicrobial peptides according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0015] The present invention proposes a method for generating antimicrobial peptides and designs an antimicrobial peptide model LTAMP based on a long short-term memory network and a Transformer model, which consists of three modules: LSTM, Transformer, and Identifying. In the process of generating polypeptide sequences, the AMP of short and medium sequences is used to train the LSTM network, and the medium and long sequences are used to train the Transformer model. In the generation stage, the LSTM is first used for autoregressive sampling to generate a legal polypeptide sequence, which is then input into the trained Transformer model to continue autoregressive generation. Finally, the generated new polypeptide sequence is used to predict the antimicrobial properties using the Identifying module. The LTAMP method combines two different learning models, which means that the output of the LSTM module is used as the input of the Transformer module to generate a new sequence, and finally the antimicrobial activity is identified by Identifying. The Identifying module includes two classifiers: QSAR random forest (RF) and QSAR support vector machine (SVM). Both classifiers are based on the characteristics of the pharmacophore, and the final output result is the average of the two prediction scores; such as Figure 1 As shown, the method includes:

[0016] S1. Data Acquisition and Processing: We first collected a large amount of protein sequence training data from the public bioinformatics databases Uniprot and PDB, including a large number of amino acid sequences, totaling 867,876. We then performed data preprocessing to filter out peptide sequences that did not meet the requirements, including removing non-FASTA format characters and sequences with more than 2,000 amino acid residues, and extracting useful sequence information from these databases. Through data preprocessing, we finally obtained sequences that could be used for model pretraining, including 694,301 valid sequences from the UniProt database and 173,575 valid sequences from the PDB database.

[0017] S2. Model Generation and Training: We designed an antimicrobial peptide model (LTAMP) based on a long short-term memory (LSTM) network and a Transformer model. This model consists of a Transformer and LSTM component for peptide sequence generation, as well as an identification component. The output of the LSTM model is used as input to the Transformer model to generate new sequences. First, a long short-term memory (LSTM) network is trained on a portion of the data to generate valid peptide sequences. The LSTM model is suitable for small to medium-sized datasets (e.g., fewer than 128 characters) and can generate stable sequences. Simultaneously, a Transformer model is used to perform model prediction on the LSTM model output data to generate novel peptide sequences. The Transformer model is suitable for large datasets (e.g., greater than 500 characters) and can generate diverse sequences. Finally, the LTAMP model is constructed by combining the advantages of the LSTM and Transformer models. The specific implementation process involves: The trained LSTM model first samples new amino acid sequence IDs through autoregression, which are then converted into amino acid sequences using a dictionary of ID-amino acid character mappings. The amino acid sequence generated by the LSTM is then used as input to the trained Transformer model. The Transformer model first performs tokenization to segment the input amino acid sequence. It then uses an ID-to-amino acid dictionary to convert the amino acid sequence into an ID sequence. The ID is then directly converted to a vector, followed by encoding and decoding. Finally, a new ID sequence is directly generated, and the ID-to-amino acid dictionary conversion is performed to obtain the final peptide sequence. Next, illegal sequence filtering and identification are performed to identify peptide sequences with high antibacterial potential. The antibacterial properties of the generated peptide sequences are comprehensively predicted using the QSAR random forest (RF) and QSAR support vector machine (SVM) algorithms in the identification module. Quantitative structure-activity relationship (QSAR) is a method for predicting the biological activity of molecules using computational chemical descriptors.

[0018] Training: The present invention trained the Transformer and LSTM for 100 iterations using data from all 747,352 amino acid sequences (563,129 from Uniprot and 184,223 from the PDB). The model was then fine-tuned using peptide sequences from experimentally proven antimicrobial peptides, including 36,088 AMP sequences. This fine-tuning enabled the model to optimize for AMP sequences, learning their specific characteristics and generating peptides with antimicrobial properties. This improved the novelty and antimicrobial potency of the generated peptide sequences.

[0019] Model Evaluation and Validation Module: Finally, the present invention uses a global alignment algorithm to quantify the similarity between the generated sequences and known AMPs, assessing the diversity and novelty of the generated peptides. A predictive QSAR model is used to calculate functional scores for the generated compound sequences, screening for sequences with QSAR scores greater than 0.5.

[0020] Example: S1. First, data was collected from 12 AMP databases, including targets from different microorganisms and mechanisms, as well as the sequence length and molecular weight of these AMPs. These peptides with known antimicrobial activity were used to fine-tune the model and compare the generated peptide sequences. They were also used to train the identification module.

[0021] S2. By preprocessing the data, polypeptide sequences that do not meet the conditions are filtered out. Sequence information is extracted from the PDBv2 and UniProt2021-04 databases. After data preprocessing, 173,575 PDBv2 sequences and 694,301 UniProt sequences are obtained. These data will be used for pre-training of the LTAMP method to ensure that the model learns how to generate legal polypeptide sequences through a large number of polypeptide sequences. For the polypeptides that have been experimentally identified as having antibacterial activity in S1, they are used for further fine-tuning of the model, so that the model learns how to generate legal polypeptide sequences after pre-training, and then undergoes a fine-tuning process of truly active polypeptide sequences, so that the final model can generate legal polypeptide sequences with certain antibacterial properties. By comparing the amino acid distribution and subsequence structure of the generated sequences with known AMP sequences, and using a global algorithm to compare the similarities between different sequence sets, the quality and diversity of the generated AMP are tested.

[0022] S3. Generate peptides with varying degrees of antimicrobial properties by combining the LSTM and Transformer methods. Use the Identifying classifier (RF and SVM) modules to predict the antimicrobial properties of the generated AMP samples.

[0023] Using collected AMP datasets and non-AMP datasets (public databases), it was demonstrated that the LTAMP method can successfully train an efficient AMP generation model; the generated peptides were extensively compared with known AMPs, and the analysis results showed that the sequences generated by the LTAMP method are more diverse and innovative than the training data of traditional models, while still retaining the key characteristics of AMP.

[0024] Based on the same inventive concept, the present invention also proposes an antimicrobial peptide production system, comprising: The acquisition unit is used to obtain protein sequence data.

[0025] The model training unit is used to input protein sequence data into the antimicrobial peptide generation model for training, specifically including: performing autoregressive sampling on the protein sequence data through the LSTM module to generate an amino acid sequence ID, and converting the generated amino acid sequence ID into an amino acid sequence; inputting the amino acid sequence into the Transformer module to perform word segmentation by character, converting the amino acid sequence after word segmentation into an ID sequence, and mapping the ID sequence into an embedding vector; encoding the embedding vector through the encoder, inputting the global features extracted by the encoding into the decoder to generate a new ID sequence, and converting the new ID sequence into a new amino acid sequence; identifying and screening the antimicrobial properties of the new amino acid sequence to obtain an antimicrobial peptide sequence.

[0026] The generation unit is used to input the protein sequence data to be generated into the trained antimicrobial peptide generation model to generate an antimicrobial peptide sequence.

[0027] The present invention also provides a computer device for generating antimicrobial peptides, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the antimicrobial peptide generation method when executing the computer program.

[0028] The present invention also provides a readable storage medium storing a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the antimicrobial peptide production method.

[0029] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for producing an antimicrobial peptide, characterized in that: The following steps are involved: Obtain protein sequence data; Inputting protein sequence data into an antimicrobial peptide generation model for training specifically includes: performing autoregressive sampling on the protein sequence data through an LSTM module to generate an amino acid sequence ID, and converting the generated amino acid sequence ID into an amino acid sequence; inputting the amino acid sequence into a Transformer module to perform character segmentation, converting the segmented amino acid sequence into an ID sequence, and mapping the ID sequence into an embedding vector; encoding the embedding vector through an encoder, inputting the global features extracted by the encoding into a decoder to generate a new ID sequence, and converting the new ID sequence into a new amino acid sequence; identifying and screening the antimicrobial properties of the new amino acid sequence to obtain an antimicrobial peptide sequence; The protein sequence data to be generated is input into the trained antimicrobial peptide generation model to generate an antimicrobial peptide sequence.

2. The method for producing an antimicrobial peptide according to claim 1, wherein: The method further includes preprocessing the protein sequence data after obtaining the protein sequence data; the preprocessing includes removing non-FASTA format characters and sequences of amino acid residues in the protein sequence data.

3. The method for producing an antimicrobial peptide according to claim 1, wherein: The antibacterial properties of the new amino acid sequences are identified and screened to obtain antibacterial peptide sequences; specifically, the antibacterial properties of the new amino acid sequences are functionally scored using random forest RF and support vector machine SVM in QSAR modeling to screen out sequences with QSAR scores greater than 0.

5.

4. An antimicrobial peptide production system, characterized in that: include: an acquisition unit, for acquiring protein sequence data; The model training unit is used to input protein sequence data into the antimicrobial peptide generation model for training, specifically including: performing autoregressive sampling on the protein sequence data through the LSTM module to generate an amino acid sequence ID, and converting the generated amino acid sequence ID into an amino acid sequence; inputting the amino acid sequence into the Transformer module to perform word segmentation by character, converting the amino acid sequence after word segmentation into an ID sequence, and mapping the ID sequence into an embedding vector; encoding the embedding vector through the encoder, inputting the global features extracted by the encoding into the decoder to generate a new ID sequence, and converting the new ID sequence into a new amino acid sequence; identifying and screening the antimicrobial properties of the new amino acid sequence to obtain an antimicrobial peptide sequence; The generation unit is used to input the protein sequence data to be generated into the trained antimicrobial peptide generation model to generate an antimicrobial peptide sequence.

5. A computer device for generating antimicrobial peptides, characterized in that: include: A memory, a processor, and a computer program stored in the memory, wherein when the processor executes the computer program, the steps of the antimicrobial peptide production method according to any one of claims 1 to 3 are implemented.

6. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the antimicrobial peptide production method according to any one of claims 1 to 3.

Images