PPA-GCN: A Efficient GCN Framework for Prokaryotic Pathways Assignment.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing generates vast microbial genomic data.
• Accurate gene functional and metabolic pathway assignment is crucial for downstream analysis.
• Current metabolic pathway assignment rates for prokaryotes are below 48%, hindering research.

Purpose of the Study:

• To develop a high-precision framework for prokaryotic metabolic pathway assignment.
• To address the bottleneck of low assignment rates in newly sequenced prokaryotic genomes.

Main Methods:

• Developed PPA-GCN, a prokaryotic pathway assignment framework utilizing graph convolutional networks.
• Integrated KEGG information and genomic characteristics, including gene synteny, for network construction.
• Employed self-supervised learning principles to enhance the framework's learning capacity.

Main Results:

• PPA-GCN achieved significantly improved assignment rates: 84.8% for *Flavobacterium*, 77.0% for *Pseudomonas*, and 71.0% for *Staphylococcus*.
• Initial assignment rates for these genera were considerably lower (27.7%, 49.5%, and 30.1%, respectively).
• The framework demonstrated strong fault tolerance and applicability to microbial genera with sufficient genomic data.

Conclusions:

• PPA-GCN effectively enhances metabolic pathway assignment accuracy in prokaryotic genomes.
• The framework offers novel insights and potential for deep learning applications in functional genomics.
• PPA-GCN is a valuable tool for interpreting functional annotations and can be extended to other prokaryotic genera.