High-accuracy SNV calling for bacterial isolates using deep learning with AccuSNV.

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Accurate bacterial single nucleotide variant (SNV) detection is crucial for microbial evolution studies, transmission tracing, and identifying antimicrobial resistance.
• Existing SNV calling tools struggle with bacterial genome complexity, leading to high false positive rates and requiring manual filtering, which is challenging with large datasets.

Purpose of the Study:

• To develop AccuSNV, a novel deep learning-based tool for high-precision and automated bacterial SNV calling.
• To improve the accuracy and efficiency of SNV detection in bacterial whole-genome sequencing data.

Main Methods:

• AccuSNV utilizes a convolutional neural network (CNN) that integrates alignment information across multiple bacterial samples.
• The tool was evaluated against seven popular SNV calling tools using simulated data from six bacterial species with varying sequencing depths and divergence levels.
• Real-world validation was performed on curated bacterial datasets containing known SNVs.

Main Results:

• AccuSNV demonstrated superior performance compared to seven other SNV calling tools in both simulated and real-world datasets.
• The deep learning approach effectively leverages across-sample patterns to enhance SNV calling precision.
• The tool achieved consistent high accuracy across diverse bacterial species, sequencing depths, and evolutionary divergence levels.

Conclusions:

• AccuSNV offers a highly precise and automated solution for bacterial SNV calling, addressing limitations of traditional methods.
• Its user-friendly downstream analysis modules (mutation annotation, phylogenetic inference, dN/dS calculation) make it broadly accessible.
• AccuSNV significantly advances the ability to study bacterial evolution, transmission, and antimicrobial resistance through accurate genomic analysis.