Prediction of Bacterial Immunogenicity by Machine Learning Methods

Abstract

Prediction of bacterial immunogens is a prerequisite for the process of vaccine development through reverse vaccinology. The application of in silico methods allows significant reduction in time and cost for the discovery of potential vaccine candidates among proteins of a bacterial species. The steps in the prediction algorithm include collection of protein sequence datasets of known bacterial immunogens and non-immunogens, data preprocessing to transform the protein sequences into numerical matrices suitable for use as training and test sets for various machine learning methods, and derivation of predictive models. The performance of the derived models is evaluated by means of classification metrics.In this chapter, we present a protocol for predicting bacterial immunogenicity by applying machine learning methods. The protocol describes the process of model development from data collection and manipulation to training and validation of the derived models.

Keywords: Auto- and cross-covariance transformation; Classification models; E-descriptors; Immunogenicity prediction; Machine learning; WEKA.