PhaGenus: genus-level classification of bacteriophages using a Transformer model.

Area of Science:

• Microbiology and Virology
• Bioinformatics and Computational Biology
• Genomics and Metagenomics

Background:

• Bacteriophages (phages) are crucial for microbial community regulation and hold promise for combating antibiotic resistance.
• High-throughput sequencing accelerates phage discovery, but taxonomic classification of phage contigs remains challenging due to genetic diversity and limited annotation.
• Existing tools struggle to balance prediction accuracy and speed for phage taxonomic classification.

Purpose of the Study:

• To develop an advanced computational model for genus-level taxonomic classification of phage contigs.
• To improve the accuracy and efficiency of phage identification, particularly for novel or divergent sequences.

Main Methods:

• Developed PhaGenus, a learning-based model employing a Transformer architecture.
• The model learns associations between protein clusters for classification.
• Evaluated PhaGenus on diverse datasets, including low-similarity and uncharacterized phage sequences.

Main Results:

• PhaGenus achieved superior performance over state-of-the-art methods, especially on low-similarity datasets (>= 13.7% improvement).
• Demonstrated significant effectiveness in identifying previously uncharacterized genera (8.52% improvement).
• Showcased high accuracy and increased contig classification rates on infant gut and GOV2.0 datasets.

Conclusions:

• PhaGenus offers a robust solution for genus-level phage taxonomic classification.
• The model enhances the ability to analyze microbial communities and discover novel phages.
• PhaGenus represents a significant advancement in phage bioinformatics tools.