Annotation-free prediction of microbial dioxygen utilization

Aerobes require dioxygen (O2) to grow; anaerobes do not. But nearly all microbes — aerobes, anaerobes, and facultative organisms alike — express enzymes whose substrates include O2, if only for detoxification. This presents a challenge when trying to assess which organisms are aerobic from genomic data alone. This challenge can be overcome by noting that O2 utilization has wide-ranging effects on microbes: aerobes typically have larger genomes, encode more O2-utilizing enzymes, and tend to use different amino acids in their proteins. Here we show that these effects permit high-quality prediction of O2 utilization from genome sequences, with several models displaying >70% balanced accuracy on a ternary classification task wherein blind guessing is only 33.3% accurate. Since genome annotation is compute-intensive and relies on many assumptions, we asked if annotation-free methods also perform well. We discovered that simple and efficient models based entirely on genome sequence content — e.g. triplets of amino acids — perform about as well as intensive annotation-based algorithms, enabling the rapid processing of global-scale sequence data to predict aerobic physiology. To demonstrate the utility of efficient physiological predictions we estimated the prevalence of aerobes and anaerobes along a well-studied O2 gradient in the Black Sea, finding strong quantitative correspondence between local chemistry (O2:sulfide concentration ratio) and the composition of microbial communities. We therefore suggest that statistical methods like ours can be used to estimate, or “sense,” pivotal features of the environment from DNA sequencing data. Importance We now have access to sequence data from a wide variety of natural environments. These data document a bewildering diversity of microbes, many known only from their genomes. Physiology — an organism’s capacity to engage metabolically with its environment — may provide a more useful lens than taxonomy for understanding microbial communities. As an example of this broader principle, we developed algorithms that accurately predict microbial dioxygen utilization directly from genome sequences without first annotating genes, e.g. by considering only the amino acids in protein sequences. Annotation-free algorithms enabled rapid characterization of natural samples, demonstrating a quantitative correspondence between sequences and local O2 levels. These results suggest that DNA sequencing can be repurposed as a multi-pronged chemical sensor, estimating concentrations of O2 and other key facets of complex natural settings.