Research On Antimicrobial Peptide Prediction Method Based On Sequence Multi-feature Joint Encoding

Antimicrobial peptides are alkaline substances with high bactericidal activity produced in organisms.As the best substitutes for antibiotics,they are receiving more and more attention in scientific research and clinical applications.Antimicrobial peptides have the advantages of quick effect,diverse action mechanisms,good resistance to drug resistance,and low toxicity,so they have broad application prospects in medicine,agriculture and other fields.In order to discover new antimicrobial peptides,it is costly and difficult to use wet experiment methods,and bioinformatics technology can effectively solve this problem.In bioinformatics,machine learning and deep learning methods have been widely used,but machine learning has the problem of relying on artificial feature engineering,so it has certain limitations.Deep learning methods can avoid tedious feature engineering,can automatically learn features from a large number of antimicrobial peptide sequences,and can more accurately predict whether new antimicrobial peptide sequences have antibacterial effects.Therefore,deep learning methods have attracted much attention in antimicrobial peptide prediction.At present,the amount of data on the functional types of antimicrobial peptides is small.If the antimicrobial peptides of specific functional types are predicted directly from the massive data,the prediction ability and generalization ability of the model will be reduced.In response to this problem,a two-level prediction model based on sequence multi-feature joint encoding was proposed to find antimicrobial peptides with specific functions.The specific research contents are as follows:(1)Antimicrobial peptide recognition model based on multi-feature joint encodingIn this study,an antimicrobial peptide recognition model based on multi-feature joint encoding was designed as the first-level model in the two-level prediction model for antimicrobial peptides.Firstly,the combination of position weight amino acid composition encoding,K-space amino acid composition encoding,N-gram encoding and original sequence digital encoding is used to code and perform feature embedding.This encoding combination takes into account both the physical and chemical properties of amino acids and the upstream and downstream Information,the features are enhanced,and then convolution,pooling and other operations are performed,and finally the prediction result is output.The first-level predictive model cleaned the data and was able to identify antimicrobial peptides with high accuracy,reducing the training and calculation of the second-level predictive model,and avoiding the interference and misjudgment of non-antimicrobial peptide data on subsequent models.(2)Antimicrobial peptide multi-label function prediction model based on temporal convolutional networkIn this study,a temporal convolutional network-based multi-label function prediction model for antimicrobial peptides was designed as the second-level model in the two-level prediction model for antimicrobial peptides.This model needs to solve the prediction problem of nine kinds of antimicrobial peptides whose data distribution is extremely unbalanced.Therefore,this paper uses the category weight method to oversample the samples,and generate An example of oversampling the amount of data in between.First,the multi-feature joint method of position weight amino acid composition encoding,K-space amino acid composition encoding,N-gram encoding and original sequence digital encoding is used for feature encoding,and feature embedding is performed,and then multi-scale convolution,time convolution,etc.operation,and finally output the prediction result.The second-level prediction focuses on the prediction of the functional type of antimicrobial peptides,avoiding the influence of confounding factors on the prediction results.The multi-feature joint encoding method designed in this study can combine multiple feature information and improve the feature representation ability.Moreover,the performance of the antimicrobial peptide secondary prediction model proposed in this study is superior to the current mainstream prediction methods,and it can identify antimicrobial peptides or predict antimicrobial peptides with specific functions with high accuracy in massive data.