A data-driven modeling framework for mapping genotypes to synthetic microbial community functions

Synthetic microbial communities offer valuable insights into the mechanisms that govern community functions, and they can be designed to achieve desired functions in order to address societal challenges in precision medicine and agriculture. Existing computational models for predicting synthetic community functions use species abundances as inputs; this makes it impossible to predict the effects of species not included in training data. We bridge this gap using a data-driven community genotype-function (dCGF) modeling framework. By lifting the representation of each species to a high-dimensional genetic feature (GF) space, dCGF learns a mapping from community GF matrices to community functions. Using in silico and experimental data, we demonstrate that dCGF can accurately predict community functions that are composed partly or entirely of new species. In addition, dCGF can generate hypotheses about the contribution of specific GFs to community functions. In sum, dCGF uses genetic information to model synthetic microbial communities in order to empower their model-driven design. A record of this paper's transparent peer review process is included in the supplemental information.