A CNN-Based K-Mer Classification of Anti-Microbial Peptide Sequences

In this paper, a natural language processing approach to classification of Anti-Microbial Peptide (AMP) sequences from three kingdoms (in total) of eukaryotic (animals and plants) and prokaryotic (bacteria) organisms has been discussed. AMPs are peptides that are produced in all living organisms as an innate natural response against fungi, bacteria, viruses and other pathogens. They can be represented by sequences of the one-letter codes of their constituent amino acids. These sequences, when treated as words of the English language, can be used to predict various characteristics using NLP. The primary approach used in this research is to break down this “word” into three different types of K-mers (with K= 2, 3 and 4). These K-mers are then encoded into numerical features and fed as inputs to a Convolutional Neural Network (CNN) model that predicts their kingdoms. While the evaluation of deep learning models for DNA sequence prediction has been performed extensively, there is still a lack of researches on peptide sequence prediction using the NLPDL approach. Not only will this research help future researchers in performing sequence analysis of peptides using the said approach, but it will also give them insight on which values of K can be suitable for classification of K-mer sequences with best accuracy.