Machine-Learning Classification Suggests That Many Alphaproteobacterial Prophages May Instead Be Gene Transfer Agents

Area of Science:

• Genomics and Bioinformatics
• Microbiology and Virology
• Molecular Biology

Background:

• Bacterial and archaeal genomes frequently contain viral DNA regions, but not all represent active viruses.
• Gene transfer agents (GTAs) are hypothesized to be former viruses that facilitate DNA exchange within bacterial populations.
• Rhodobacter capsulatus GTA (RcGTA)-like genes are common in Alphaproteobacteria but often misannotated as viral proteins.

Purpose of the Study:

• To develop a computational tool for accurately distinguishing RcGTA-like genes from viral homologs.
• To assess the prevalence of RcGTA-like elements in alphaproteobacterial genomes.
• To improve the identification of mobile genetic elements in microbial genomes.

Main Methods:

• Development of a support vector machine (SVM) classifier based on amino acid composition differences.
• Training the classifier on a manually curated dataset of homologous viruses and GTAs.
• Application of the classifier to analyze 1,423 alphaproteobacterial genomes.

Main Results:

• The SVM classifier accurately differentiates RcGTA-like genes from viral homologs.
• RcGTA-like "head-tail" gene clusters were identified in 57.5% of the examined alphaproteobacterial genomes.
• Over half of predicted in silico prophages were reclassified as GTAs, indicating widespread misidentification.

Conclusions:

• The developed classifier is an effective tool for identifying RcGTA-like elements in newly sequenced genomes.
• GTAs are significantly more prevalent in Alphaproteobacteria than previously recognized.
• This work highlights the need for refined methods to distinguish viral elements from functional genetic elements like GTAs.