Learning the mechanism of collective microbial function via random community-media pairing

Microbial functions often emerge at community level and involve complex interactions. Despite the complexity, previous work shows that regression over randomly-sampled datasets is capable of functional predictions. This project moves beyond prediction and asks if regression can also be informative about mechanisms underlying collective functions. For this, we propose extending the random sampling method to vary the growth environment, cultivating each community in the spent medium of another randomly constructed community. We illustrate the approach using a model-based analysis, showing that this random pairing protocol can succeed at assigning species to the correct reaction pathway steps and identifying species essential to the collective function. More generally, our work illustrates that the utility of machine learning-based approaches can be greatly enhanced by a synergistic experimental design.