AMAP: Hierarchical multi-label prediction of biologically active and antimicrobial peptides

Due to increase in antibiotic resistance in recent years, development of efficient and accurate techniques for discovery and design of biologically active peptides such as antimicrobial peptides (AMPs) has become essential. The screening of natural and synthetic AMPs in the wet lab is a challenge due to time and cost involved in such experiments. Bioinformatics methods can be used to speed up discovery and design of antimicrobial peptides by limiting the wet-lab search to promising peptide sequences. However, most such tools are typically limited to the prediction of whether a peptide exhibits antimicrobial activity or not and they do not identify the exact type of the biological activities of these peptides. In this work, we have designed a machine learning based model called AMAP for predicting biological activity of peptides with a specialized focus on antimicrobial activity prediction. AMAP used multi-label classification to predict 14 different types of biological functions of a given peptide sequence with improved accuracy in comparison to existing state of the art techniques. We have performed stringent performance analyses of the proposed method. In addition to cross-validation and performance comparison with existing AMP predictors, AMAP has also been benchmarked on recently published experimentally verified peptides that were not a part of our training set. We have also analyzed features used in this work and our analysis shows that the proposed predictor can generalize well in predicting biological activity of novel peptide sequences. A webserver of the proposed method is available at the URL: http://faculty.pieas.edu.pk/fayyaz/software.html#AMAP.