Support vector machine for prediction of antimicrobial peptides in legumes

Abstract

Resistance to chemical antibiotics is an unsolved and growing problem. A new generation of native peptide molecules, also known as antimicrobial peptides (AMPs) may be a natural alternative to chemical antibiotics and a potential area of research under applied biotechnology. In the present study, a systematic attempt has been made to develop a direct method for predicting AMPs of legumes using Support Vector Machine (SVM). The SVM based method with polynomial kernel function with degree 2 was found to be the best model for classification of legume AMPs with accuracy and Mathews Correlation Coefficient of 96.4% and 0.931, respectively. The best performance was obtained at threshold 0.5, where the sensitivity, specificity were 1.000 and 0.929, respectively. The ROC curve was plotted and area under curve (AUC) was found to be 0.964 with standard error of 0.041, which indicated a good prediction performance. It is anticipated that the current prediction method would be a useful tool for the systematic analysis of genome data. AMPs identified from the studies may be used to confer disease resistance in other crops as transgenics, thus opening unsuspected alternative to provide agronomically relevant levels of disease control worldwide