Predicting promoters in phage genomes using machine learning models

Abstract

The renewed interest in phages as antibacterial agents has led to the exponentially growing number of sequenced phage genomes. Therefore, the development of novel bioinformatics methods to automate and facilitate phage genome annotation is of utmost importance. The most difficult step of phage genome annotation is the identification of promoters. As the existing methods for predicting promoters are not well suited for phages, we used machine learning models for locating promoters in phage genomes. Several models were created, using different algorithms and datasets, which consisted of known phage promoter and non-promoter sequences. All models showed good performance, but the ANN model provided better results for the smaller dataset (92% of accuracy, 89% of precision and 87% of recall) and the SVM model returned better results for the larger dataset (93% of accuracy, 91% of precision and 80% of recall). Both models were applied to the genome of Pseudomonas phage phiPsa17 and were able to identify both types of promoters, host and phage, found in phage genomes.