How to efficiently predict the structure of microbial communities

Our body is home to many microbes, which function as a community to help us fight against bad microbes, train our immune system and even influence our mental health. However, predicting the behavior of these microbial communities is challenging due to the presence of a large number of inter-species interactions. Traditionally, these interactions are estimated using a bottom-up approach by examining all microbes alone followed by all pairs of microbes, all triads of microbes, and so on. This prediction method however becomes increasingly intractable as the number of species in the community increases and becomes practically impossible for a real microbial community that has hundreds of microbes. To solve this problem, we proposed a reverse, top-down approach to estimate the interactions. As opposed to the traditional, bottom-up approach, our method requires examining only monocultures and leave-one-out sub communities. Leave-one-out sub communities are obtained by dropping one species at a time from the original community. This enables our method to be applied to real microbial communities that have hundreds of microbes. Finally, to rigorously validate our method, we applied it to millions of computer generated synthetic communities and two real communities and found that our method performs impressively well in each case.