Prediction of butyrate production in microbial communities using machine learning

Butyrate is a beneficial metabolite produced by the gut microbiome with disproportionate in vivo concentrations linked to disease, particularly inflammation-based disorders. We trained a machine learning model to predict microbial community butyrate production using published in vitro data. We explored different training and testing regimes, demonstrating that training on communities of intermediate complexity outperformed training on simple or highly complex communities while training on data distributed across communities of varying complexity was the most effective strategy. Our work aims to optimize experimental data collection strategies. Future modeling work will incorporate mechanistic metabolic information into predictive models to improve performance.