Utility of a quantitative approach to microbial dysbiosis using machine learning in an African American cohort with self-reported hair loss

Microbial dysbiosis has been identified as a therapeutic target for several dermatologic conditions. However, the concept of dysbiosis is poorly defined, limiting reproducibility. We developed a reproducible metric for dysbiosis that can be applied to research within dermatology. Thirty-six individuals from an African-American cohort with self-reported hair loss without formal clinical diagnoses provided scalp swabs from both afflicted (hair loss) and normal (no hair loss) sites. The scalp microbiome was characterized via 16S rRNA gene sequencing. A dysbiosis score that considers the proportion of all taxa within the samples was calculated. Further, we identified the taxa most associated with dysbiosis using both a machine learning random forest classifier and a negative binomial mixed effects model to control for participant age. Sites of hair loss exhibited an increase in microbial diversity, in particular for individuals older than 40 years. We identified a core set of OTUs assigned to 7 genera that were significant contributors to increased scalp dysbiosis. This work demonstrates the utility of applying a quantitative approach to dysbiosis and provides a framework that can be applied to other microbiome-associated conditions.