Integrating chromatin conformation information in a self-supervised learning model improves metagenome binning

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Metagenome binning groups DNA fragments by genome of origin, crucial after assembly.
• Current methods struggle with binning completeness in low-sample datasets due to limited co-abundance data.

Purpose of the Study:

• To develop a novel, reference-independent algorithm to enhance metagenome binning completeness.
• To leverage chromatin conformation data for improved genomic binning.

Main Methods:

• Developed metaBAT-LR, a self-supervised algorithm utilizing Hi-C sequencing data.
• Built a model from high-quality genome bins to predict scaffold origins.
• Applied predictions to merge incomplete bins and recruit unbinned scaffolds.

Main Results:

• metaBAT-LR successfully improved binning completeness on synthetic and real-world datasets.
• The algorithm demonstrated effectiveness in merging bins and recruiting unbinned scaffolds.
• Benchmarking revealed metaBAT-LR identified unique bins missed by other tools.

Conclusions:

• metaBAT-LR significantly enhances metagenome binning completeness, particularly for challenging datasets.
• The algorithm offers a valuable new approach for genomic analysis in microbial communities.
• metaBAT-LR is open-source, promoting wider accessibility and application in research.