Accurate prediction of metagenome-assembled genome completeness by MAGISTA, a random forest model built on alignment-free intra-bin statistics.

Background: Although the total number of microbial taxa on Earth is under debate, it is clear that only a small fraction of these has been cultivated and validly named. Evidently, the inability to culture most bacteria outside of very specific conditions severely limits their characterization and further studies. In the last decade, a major part of the solution to this problem has been the use of metagenome sequencing, whereby the DNA of an entire microbial community is sequenced, followed by the in silico reconstruction of genomes of its novel component species. The large discrepancy between the number of sequenced type strain genomes (around 12,000) and total microbial diversity (10-10 species) directs these efforts to de novo assembly and binning. Unfortunately, these steps are error-prone and as such, the results have to be intensely scrutinized to avoid publishing incomplete and low-quality genomes.

Results: We developed MAGISTA (metagenome-assembled genome intra-bin statistics assessment), a novel approach to assess metagenome-assembled genome quality that tackles some of the often-neglected drawbacks of current reference gene-based methods. MAGISTA is based on alignment-free distance distributions between contig fragments within metagenomic bins, rather than a set of reference genes. For proper training, a highly complex genomic DNA mock community was needed and constructed by pooling genomic DNA of 227 bacterial strains, specifically selected to obtain a wide variety representing the major phylogenetic lineages of cultivable bacteria.

Conclusions: MAGISTA achieved a 20% reduction in root-mean-square error in comparison to the marker gene approach when tested on publicly available mock metagenomes. Furthermore, our highly complex genomic DNA mock community is a very valuable tool for benchmarking (new) metagenome analysis methods.