Alphabet reduction and distributed vector representation based method for classification of antimicrobial peptides

Antimicrobial peptides(AMPs) also known as host defence peptides are an essential part of innate immunity. AMPs are emerging as promising agents to multidrug resistant pathogens owing to their size, toxicity and biological activities. Effective identification of AMPs using computational method will be helpful in designing new antimicrobial agents for further study. Sequence based analysis for AMPs have been there for a while, where different methods have been proposed using amino acid composition and pseudo amino acid composition methods for inferring the activity of AMPs. In this paper, we demonstrate the use of machine learning models using alphabet reduction and distributed vector representation for classifying a sequence as AMP and non-AMP. The alphabet reduction is based on various physico-chemical properties of peptide sequences such as hydropathy index, contact energies between amino acids, conformation similarity, substitution matrix, amino acid charges etc. Alphabet reduction along with distributed vector representation (ProtVec) gives promising results and is also found to be computationally inexpensive as compared to other sequence based methods. Antimicrobial peptides are predicted using a binary classifier which gives a n a ccuracy of 97.94% a long with MCC 0.94.