CoCoBin: Graph-Based Metagenomic Binning via Composition–Coverage Separation

Abstract Motivation Metagenomic binning is a critical step in metagenomic analysis, aiming to cluster contigs from the same genome into coherent groups. In contemporary workflows, most binning tools begin with the assembly of shotgun metagenomic sequencing data. The assembled contigs are then grouped into bins representing individual microbial genomes or species, typically using taxonomy-independent methods. Although several methods exist, metagenomic binning remains a challenging yet mandatory task, particularly in the context of complex and highly diverse microbial communities. Results We propose CoCoBin, a novel metagenomic binning tool explicitly designed for the effective binning of metagenomic contigs. In this study, we introduced an innovative approach for calculating contig similarity by separating composition and coverage information. The method begins by (1) assigning contigs into a cluster based on length ranges, (2) calculating contig similarity based on composition features (e.g., k-mer frequencies), and (3) calculating contig difference based on coverage features. These similarity measures are then integrated to construct a graph, where nodes represent contigs and edges represent the similarities between them. Finally, the Louvain algorithm is applied to the graph to cluster closely related contigs. CoCoBin was compared against several state-of-the-art binning tools: BusyBee Web, CONCOCT, MaxBin 2.0, MetaBAT 2, and MetaDecoder on nine simulated datasets, five mock community datasets, and one real dataset. The AMBER tool used to evaluate the binning results across all datasets shows that CoCoBin achieved the best performance regarding the number of bins identified, followed by its performance on the F1 score. Availability The source code of CoCoBin is available at https://github.com/cucpbioinfo/CoCoBin Contact duangdao.w@chula.ac.th Supplementary information Supplementary data are available at Bioinformatics online.