Discovering antibiotics by integrating microbiome and metabolome data via high dimensional mediation analysis

Antibiotics resistance is a problem of growing importance. Discovery of new candidate compounds that are tolerable for humans is an important path to alleviate this problem as most screened compounds are toxic to humans. In this work we leverage publicly available data of paired microbiome and metabolome samples from human gut to discover compounds that has antibiotic activity at physiologically relevant concentrations to identify non-toxic candidates.

We built on the ecological theory that some bacteria produce compounds to inhibit competing bacteria, by constructing correlation networks and selecting negatively correlating taxa. To select metabolite features that might mediate this killing we fit high dimensional mediation models while controlling false discovery rates.

We test this proposition by selecting candidate compounds against Vancomycin Resistant Enterococcus (VRE). We selected 5 candidates of which 1 had previously been reported in the literature. We determined Minimum Inhibitory Concentrations and saw that 3 including the previously reported compound had inhibitory effects.

In conclusion, integration of parallel metabolome and microbiome observations can be used to discover compounds with antibiotic activity.