Bacteriophage classification for assembled contigs using graph convolutional network.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacteriophages (phages) are the most abundant biological entities, crucial for microbial biology.
• High-throughput sequencing, especially metagenomics, rapidly reveals new phages, but taxonomic classification lags significantly.
• Existing alignment-based tools struggle with classifying diverse and abundant phage contigs from metagenomic data.

Purpose of the Study:

• To develop a novel semi-supervised learning model for accurate taxonomic classification of phage contigs.
• To address the challenges posed by high phage diversity, abundance, and limited known phage classifications.

Main Methods:

• Developed PhaGCN, a semi-supervised learning model for phage contig taxonomic classification.
• Constructed a knowledge graph integrating DNA sequence features (via CNN) and protein sequence similarity (via gene-sharing network).
• Utilized graph convolutional networks to leverage both labeled and unlabeled data for enhanced learning.

Main Results:

• PhaGCN demonstrated effective taxonomic classification of phage contigs.
• The model performed favorably against existing phage classification tools on both simulated and real sequencing data.
• The approach successfully combined sequence features and network similarity for improved classification.

Conclusions:

• PhaGCN offers a robust solution for the taxonomic classification of bacteriophages from metagenomic data.
• The model's performance indicates its potential to advance phage bioinformatics and microbial ecology.
• The developed method aids in understanding the vast, largely undiscovered phage population.