Replidec - Use naive Bayes classifier to identify virus lifecycle from metagenomics data

Abstract

MotivationViruses are the most abundant biological entities on earth. The majority of these entities are bacterial viruses or phages which specifically infect bacteria. Phages can use different replication strategies to invade their hosts including lytic, lysogenic, chronic cycle and pseudolysogeny. While the determination of the replication strategy used by phages is important to explore the phage-bacteria relationships in different ecosystems there are not many tools that can predict this in metagenomic data. In addition, most of the tools available can only predict lytic and lysogenic cycles. To address this issue, we have developed a new software called Replidec to identify three most common phage replication cycles (virulent, temperate, chronic) in viral sequences. ResultsReplidec uses Naive Bayes classifier combined with alignment-based methods to improve the prediction accuracy in metagenomic data. We test Replidec on viral genomes with known replication cycle and simulated metagenomic sequences. Replidec perform relatively good both in isolated genomes (F1 score: 92.29% {+/-} 0.81; mcc: 89.14% {+/-} 1.22) and simulated metagenomic sequences(F1 score: 87.55% {+/-} 2.12; mcc: 88.23% {+/-} 2.55). Moreover, Replidec can also accurately predict the replication cycle in small viral fragments([~]3000bp). In conclusion, Replidec can achieve the best performance in simulated metagenomic data compared to most prediction softwares including BACPHLIP.