PhaTYP: predicting the lifestyle for bacteriophages using BERT.

Area of Science:

• Computational biology
• Microbiology
• Bioinformatics

Background:

• Bacteriophages (phages) exhibit virulent or temperate lifestyles, crucial for understanding host interactions and applications like phage therapy.
• Experimental lifestyle annotation lags behind rapid phage genome sequencing, necessitating computational prediction tools.
• Existing computational tools struggle with short phage contigs common in metagenomic data.

Purpose of the Study:

• To develop an accurate computational tool, PhaTYP, for predicting bacteriophage lifestyles, with a focus on improving performance on short contigs.
• To address limitations in current phage lifestyle prediction methods, particularly for fragmented genomic data.

Main Methods:

• Developed PhaTYP, a novel tool for phage lifestyle prediction.
• Implemented self-supervised and fine-tuning training tasks to enhance prediction accuracy.
• Rigorously compared PhaTYP against four state-of-the-art methods: DeePhage, PHACTS, PhagePred, and BACPHLIP.

Main Results:

• PhaTYP demonstrated superior performance compared to existing methods in predicting phage lifestyles.
• PhaTYP achieved more stable and accurate predictions, especially on short phage contigs.
• The tool was successfully applied to analyze phage lifestyles within human neonates' gut metagenomic data.

Conclusions:

• PhaTYP offers a significant advancement in computational phage lifestyle prediction, particularly for challenging short sequence data.
• The tool enhances the analysis of phages in complex metagenomic datasets, contributing to microbial community understanding.
• PhaTYP is valuable for research in phage biology, phage therapy, and microbial ecology.