Comprehensive function annotation of metagenomes and microbial genomes using a deep learning-based method

Comprehensive protein function annotation is key to understanding microbiome-related disease mechanisms in the host organisms. We have developed a new metagenome analysis workflow integrating de novo genome reconstruction, taxonomic profiling and deep learning-based functional annotations from DeepFRI. We validate DeepFRI functional annotations by comparing them to orthology-based annotations from eggNOG. Further, we demonstrate the usage of the workflow using 1,070 infant metagenome samples from the DIABIMMUNE cohort. We have generated a sequence catalogue comprising a collection of 7,174 metagenome-assembled genomes (MAGs), of which 2,255 are high quality near-complete bacterial genomes. These genomes encode 1.9 million non-redundant genes. We found high concordance (70%) between GO annotations predicted by DeepFRI and eggNOG. Importantly, we show that DeepFRI improved the annotation coverage, with nearly all the genes in the gene catalogue obtaining GO molecular function annotations, although the annotations are less specific compared to eggNOG. The pan-genome analysis of 42 bacterial species revealed a striking difference between DeepFRI and eggNOG annotations. eggNOG was shown to annotate more genes coming from well-studied organisms such as Escherichia coli while DeepFRI on the other hand is not sensitive to taxa. This work contributes to the understanding of the functional signature of the human gut microbiome. ### Competing Interest Statement The authors have declared no competing interest.