REMAG: recovery of eukaryotic genomes from metagenomic data using non-contrastive self-supervised learning

REMAG recovers more near-complete eukaryotic genomes than existing tools in metagenomic data, outperforming current methods in simulated and real datasets.

• Why it matters: Eukaryotic MAG recovery lags behind prokaryotes due to reliance on prokaryotic reference databases and genome size limitations, hindering understanding of microbial communities.
• What they did: The study developed REMAG, a tool that uses a genomic foundation model, a dual-encoder Siamese network trained with Barlow Twins loss, and Leiden clustering to recover eukaryotic genomes from long-read metagenomic data, tested on simulated and real datasets.
• The result: REMAG demonstrates superior recovery of high-quality eukaryotic genomes, enabling more comprehensive analysis of microbial communities and scalable binning as dataset sizes increase.