Prophage loci predictor for bacterial genomes

Abstract

This paper proposes a new algorithm for prophage loci prediction in bacteria. Prophages are defined in Bioinformatics as viral nucleotide sequences that are found intermixed with host nucleotide sequences in bacteria. The proposed algorithm uses machine learning patterns and processing methodologies in order to provide a highly efficient system for loci prediction, thereby reducing the time-space complexity required unlike others of its class. In the training phase, a pattern database is constructed from raw nucleotide sequences of both bacteria and viruses obtained from a training set. In the prediction phase, the aforementioned database is used along with Particle Swarm Optimization (PSO) to predict the probable loci of prophages in a test set of bacterial nucleotide sequences. Testing this method on raw sequences consisting of both partial and complete nucleotide sequences of various bacteria has yielded good results in predicting the loci of prophages in them. This algorithm and connected processes compare favorably in terms of predictive performance with others of its class such as PhiSpy and ProphET, while outperforming others in terms of raw processing speed, suggesting that a data-centric approach can yield comparable results while using a fraction of the resources.