Predicting functions of uncharacterized gene products in microbial communities

Microbial communities are rich reservoirs for molecular functions that influence environmental and host-associated chemistry, with numerous roles in ecosystem maintenance, health, and disease. However, our knowledge of these molecular mechanisms is limited, due to the massive range of microbial genetic material in comparison to the limited throughput available for experimental characterization. Here, we developed a novel method to systematically predict functional capabilities of uncharacterized proteins in microbial communities by assessing high-dimensional community-wide data. We predicted potential functions for the majority of uncharacterized protein families (~70% of total) in 1,595 gut metagenomes (MGX) and 800 metatranscriptomes (MTX) from the Integrative Human Microbiome Project (HMP2). Using only MTX-based information, our approach achieved an average of approximately 0.7 AUC for Gene Ontology (GO) biological process term prediction. By aggregating predictions from other types of information (e.g. sequence similarity), the AUC was further improved to ~0.88. Further evaluations showed that our method recapitulated comparable, realistic prediction profiles from microbial communities when compared to state-of-the-art tools for function prediction in single-organisms. FUGAsseM predicted high-confidence functional annotations for >283,000 protein families (70% previously uncharacterized), including >20,000 novel protein families (i.e. those without substantial homology to isolate genomes). The functional capacity of microbial proteins was extensively unexplored by leveraging MTX-based coexpression for both common gut taxa and less-studied species in the human gut. Our method is generalizable to any type of microbial community, providing a new approach to predict microbial protein functions. We implemented it as an open-source tool, FUGAsseM (Function predictor of Uncharacterized Gene products by Assessing high-dimensional community data in Microbiomes), along with documentation available at http://huttenhower.sph.harvard.edu/fugassem. This study expands the functional landscape of the human microbiome and allows better exploitation of microbial proteins in any under-characterized communities.