A conceptual framework for revealing rare bacterial species in the gut microbiome through guided data transformation: Beyond enterotypes

Abstract The gut microbiota is increasingly recognized as a rich source of biological data that offers critical information on host health, including information on pathological conditions such as cancer, diabetes and other metabolic disorders. Inferring host health from gut bacterial composition using statistical analytical methods remains a challenge. Here, we demonstrate that groups of bacterial species with high abundance and variance (referred to as dominant bacterial species and often associated with enterotypes) exert a disproportionately large influence on gut microbiome analyses, thereby obscuring the contribution of less abundant species (referred to as rare bacterial species). To address this limitation, we propose a guided data transformation highlighting rare bacterial species while minimizing the impact of dominant bacterial species on microbiome statistical analyses. This transformation (i) yields transformed data for which clustering is more closely associated with host health and (ii) helps to improve the performance of supervised machine learning algorithms in high‐dimensional settings (). By applying the guided data transformation to three real data sets, we demonstrate that this method (i) improves unsupervised analysis, (ii) supervised machine learning algorithms in high‐dimensional settings () and (iii) suggests that enterotypes may act as a confounding variable in predicting host health or at least hide biological information.