Computational prediction of continuous B-cell epitopes using random forest classifier

Immune responses protect against infection by microbes like viruses, bacteria, fungi and other parasites and against entry of non-microbial foreign matter. Any foreign matter which enters the body that can induce the immune system to produce a corresponding antibody is called an antigen. Antibodies are specific proteins produced to destroy antigen. A portion of antigen which can bond with the antigen binding site of the antibody is called B-cell epitope or antigenic determinant. These epitopes play a vital role in the development of peptide vaccines, in diagnosis of diseases, immune based cancer therapies and also for allergy research. Computational prediction of the presence and location of B-cell epitopes is a challenging task in the field of immunology. This paper proposes a new data mining model for prognostication of continuous B-cell epitopes by combining amino acid propensity scales and di, tri and tetra peptide immunogenicity scales. Principal Component Analysis, a dimensionality reduction technique was used to filter features that do not play a major role in epitope prediction. Finally the prediction was performed by Random Forest Classifier. Experimental results have shown that RF classifier improved the prediction accuracy compared to existing prediction algorithms like AAP, AAT-fs and BCPred.