Summary
A bioRxiv preprint describes DeepMM, a method that detects suspected misassembly junctions and splits contigs before microbial genomes are grouped into bins. In benchmark datasets, this reduced gene annotations that disagreed with sample-matched reference genomes.
A bioRxiv preprint describes DeepMM, a method for finding and splitting suspected misassembly points in microbial DNA contigs before those sequences are grouped into metagenome-assembled genomes (MAGs). Across benchmark datasets, the researchers report that this correction reduced gene and virulence-factor annotations that disagreed with sample-matched reference genomes.
The study highlights an assembly problem that can persist even when a MAG appears nearly complete by standard quality checks: DNA from different species can be joined within one contig, a continuous assembled sequence. Such a chimeric contig may then carry genes that do not fit the organism assigned to the genome.
Why marker-based quality checks can miss chimeras
Metagenomic sequencing captures DNA from a community of organisms. Assembly software joins short sequencing reads into longer contigs, and binning methods group contigs thought to come from the same organism. The resulting MAGs let researchers study microbes that may be difficult to grow in the laboratory.
Common estimates of MAG completeness and contamination use marker genes—genes expected to occur in particular groups of organisms. As the authors note, these checks can indicate that a genome is near-complete while failing to detect foreign DNA joined inside a contig. The issue matters because gene content is often used to infer what a microbe can do.
The researchers examined 63 samples across eight Critical Assessment of Metagenome Interpretation (CAMI) datasets. At least one interspecies chimeric contig was found in 444 of 640 near-complete MEGAHIT MAGs (69.4%) and 141 of 687 near-complete metaSPAdes MAGs (20.5%). Among affected MAGs, chimeric contigs accounted for a median 3.9% and 1.1% of total genome length, respectively. The authors found that these contigs carried genes and virulence-factor annotations discordant with the MAG's reference genome.
DeepMM corrects contigs before binning
DeepMM uses spatial patterns in short-read alignments to identify candidate misassembly junctions. It splits a contig at a supported candidate junction before binning, aiming to prevent joined sequences from being treated as part of the same genome.
In the CAMI assemblies, the researchers report higher contig-level AUPRC and AUROC for DeepMM than for existing tools across all eight datasets. These are measures of how well a method distinguishes candidate misassemblies from other contig regions. In two real-world gut cohorts evaluated against long-read assemblies, DeepMM had the highest F1 scores and competitive AUPRC and AUROC. The abstract also reports that correction increased the number of CheckM2-defined near-complete MAGs in CAMI and four real-world cohorts.
For CAMI MEGAHIT MAGs, DeepMM correction reduced reference-discordant genes on chimeric contigs by 35% and reference-absent virulence-factor calls by 43%. The corresponding reductions after correction with metaMIC were 23% and 20%. These gene and virulence-factor comparisons use sample-matched reference genomes; the authors note that the results depend on which reference set is used.
The findings suggest that checking and correcting contig structure can complement marker-based MAG quality control. The work is a bioRxiv preprint, and its reported annotation discrepancies are defined by comparison with reference genomes rather than by direct measurement of microbial activity.