Highly accurate classification and discovery of microbial protein-coding gene functions using FunGeneTyper: an...

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• High-throughput DNA sequencing generates vast microbial gene data.
• Accurate functional assignment of novel protein-coding genes remains a significant challenge.
• Identifying antibiotic resistance genes (ARGs) and virulence factors is crucial for public health.

Purpose of the Study:

• To develop FunGeneTyper, an advanced framework for precise and fine-grained classification of ARGs and virulence factor genes.
• To leverage deep learning for enhanced functional gene annotation in microbial communities.
• To provide a versatile and accessible tool for researchers in metagenomics and biotechnology.

Main Methods:

• Development of FunGeneTyper, incorporating two novel deep learning models: FunTrans and FunRep.
• Creation of structured databases and supporting resources for robust gene classification.
• Utilizing an experimentally confirmed dataset of ARGs with remote homologous sequences for rigorous testing.

Main Results:

• FunGeneTyper achieved high accuracy (Accuracy > 0.99) and F1-scores (> 0.97) in classifying ARGs and virulence factors.
• Demonstrated superior performance in discovering novel ARGs across human gut, wastewater, and soil microbiomes compared to state-of-the-art tools.
• Outperformed sequence alignment-based and domain-based annotation approaches in ARG discovery.

Conclusions:

• FunGeneTyper offers a highly accurate and efficient solution for microbial gene function classification.
• The framework's lightweight, privacy-preserving, and plug-and-play nature enhances its accessibility and applicability.
• Widespread adoption of FunGeneTyper is anticipated to advance microbiome research, biotechnology, and bioinformatics.