Improved Antimicrobial Peptide Prediction Based On Multi-label Classification Algorithm

Antimicrobial peptide prediction plays an important role in the field of drug molecular design.It is an important research direction of bioinformatics,including the prediction of antimicrobial peptide resistance and antimicrobial peptide activity.Among them,the traditional prediction models mostly ignore the internal association information of resistance genes,which affects the prediction accuracy.However,in the activity prediction,the traditional multi-label model has low prediction performance because of its high feature dimension and relatively small number of instances.In view of the above two problems,this paper carries out the improvement of multi-label classification algorithm and its application in antimicrobial peptide prediction.The specific research contents are as follows:Considering the problem that the traditional resistance prediction model mostly ignores the internal correlation information of resistance genes,which affects the prediction accuracy,a new feature extraction method and a cascade forest antimicrobial peptide resistance prediction model are proposed by making full use of the amino acid composition information in the position specific score matrix(PSSM)and adding the amino acid composition and eight kinds of physical and chemical properties information.The experimental results show that the proposed method can effectively predict the resistance of antimicrobial peptides.Compared with the existing antimicrobial peptide resistance prediction algorithms,the proposed algorithm improves the prediction accuracy.In order to enhance the generalization ability of antimicrobial peptide resistance prediction model,especially for the unbalanced distribution of positive and negative samples in antimicrobial peptides,this paper further optimizes the model,and takes fish antimicrobial peptides as an example to verify the experimental results,which shows that the model has a certain generalization ability.Aiming at the problem of high feature dimension and relatively small number of instances in activity prediction,feature selection of high-dimensional data is firstly carried out by introducing weight matrix.Secondly,in order to solve the problem of missing labels,high-order correlation between labels is fully used to complete the missing labels.Finally,in order to solve the problem of overfitting caused by relatively small number of examples,the regularization term is introduced to construct the activity prediction model of antimicrobial peptides.The proposed model is validated in the comprehensive antimicrobial peptide datasets,and the results show that the proposed model could effectively predict the antimicrobial peptide activity.