Statistical design of a synthetic microbiome that clears a multi-drug resistant gut pathogen

Given a set of parts and no pre-determined instruction manual on how to put these parts together, how can we create a system that works?

The concept of microbiome design is not new, but ways of approaching the problem have focused on just two paths: ‘bottom-up’ and ‘top-down’ approaches. The ‘bottom-up’ approach is to construct communities one bacterium at a time with the hope that properties of individual bacteria still hold when placed in community settings. The ‘top-down’ approach is to take intact whole microbiomes that perform a function (like a whole gut microbiome that suppresses the invasion of pathogens) and systematically reduce the community to its core members. Despite the success of these approaches in specific cases, neither ultimately yields principles by which communities can be designed in a truly forward manner. In other words, transitioning microbiome design from a discipline of trial-and-error to one of true engineering is still lacking. In this paper, we laid out a statistical framework for inferring a generative model of microbiome design using no information about the underlying properties of individual bacteria, the capacity of bacteria to interact with each other, or mechanism by which they can execute a desired function. We instead created a Design-Build-Test-Learn-Distill method where many synthetic microbiomes were designed, built, and tested for a specific function; then a machine-learning model was learned and distilled to learn key constraints that directly indicated what community to build. Our test function was to kill a multi- drug resistant pathogen called Klebsiella pneumoniae.

As the paper shows, we successfully created a 15-member community (SynCom15) that kills K. pneumoniae in mice better than a whole stool transplant. This work has nucleated much interest in our capacity to design synthetic biological systems using statistics and has formed a major area of interest for our laboratory.