RepBin: Constraint-Based Graph Representation Learning for Metagenomic Binning

Mixed communities of organisms are found in many environments -- from the human gut to marine ecosystems and can have profound impact on health environment. Metagenomics studies genomic material such through high-throughput sequencing that yields DNA subsequences for subsequent analysis. A fundamental problem standard workflow, called binning, is discover clusters, subsequences, associated with constituent organisms. Inherent noise various biological constraints need be imposed them skewed cluster size distribution exacerbate difficulty this unsupervised learning problem. In paper, we present a new formulation using graph where nodes edges represent homophily information. addition, model providing heterophilous signal about cannot clustered together. We solve binning by developing algorithms (i) representation preserves both relations heterophily (ii) constraint-based clustering method addresses problems distribution. Extensive experiments, real synthetic datasets, demonstrate our approach, RepBin, outperforms wide variety competing methods. Our methods, may useful other domains as well, advance state-of-the-art metagenomics learning.