DCVBin: a novel binning method for single-sample metagenomes based on DNA language model and variational autoencoder.

Area of Science:

• Metagenomics
• Bioinformatics
• Genomics

Background:

• Metagenomic DNA contigs binning is crucial for reconstructing genomes.
• Existing methods struggle with single-sample binning due to limited coverage data.
• This limitation hinders detailed metagenomic analysis at the individual sample level.

Purpose of the Study:

• To develop a novel single-sample metagenomic contigs binning method.
• To improve genome reconstruction accuracy in scenarios with limited coverage.
• To integrate advanced language modeling with traditional binning techniques.

Main Methods:

• Proposed DCVBin, a method incorporating semantic features from a DNA language model.
• Utilized variational autoencoder to integrate DNA language model features with 4-mer frequencies.
• Employed k-means clustering, with cluster number determined by single-copy genes.

Main Results:

• DCVBin demonstrated high-accuracy single-sample metagenomic binning across six datasets.
• Outperformed existing state-of-the-art methods in single-sample binning tasks.
• A DCVBin-integrated framework accurately predicted colorectal cancer from gut metagenomes.

Conclusions:

• DCVBin significantly advances single-sample metagenomic binning capabilities.
• The method holds promise for disease diagnostics using metagenomic data.
• Identified potential microbial biomarkers for colorectal cancer detection.