Metabolite Prediction in the Human Microbiome

• Activity Number: 221 - Advanced Statistical Methods for Microbiome Data Analysis
• Type: Topic Contributed
• Date/Time: Monday, July 31, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Himel Mallick* and Eric A. Franzosa and Lauren J. Mclver and Soumya Banerjee and Alexandra Sirota-Madi and Aleksandar D. Kostic and Clary B. Clish and Hera Vlamakis and Ramnik Xavier and Curtis Huttenhower
• Companies: Harvard University and Harvard University and Harvard University and University of Oxford and Evelo Biosciences and Harvard University and Broad Institute and Broad Institute and Massachusetts General Hospital and Harvard T.H. Chan School of Public Health
• Keywords: Microbiome ; Metabolomics ; Metagenomics ; Biomarker Discovery ; Predictive Modeling ; Penalized Regression

Abstract

We describe MelonnPan, a computational method to predict metabolite pools given shotgun metagenomic information by incorporating biological prior knowledge in the form of microbial gene families or pathway abundance profiles. MelonnPan uses an elastic net regularization technique to infer which gene families are predictive and then combines these gene families to estimate the composite metabolomes. This allows us to accurately infer a community-wide metabolic network by drawing on a pool of up to one million microbial genes. Focusing on two independent gut microbiome datasets comprising over 150 patients with inflammatory bowel disease and control participants, we demonstrate that our framework successfully recovers observed community metabolic trends in over 90% of the metabolites, including the prediction of metabolic shifts associated with bile acids, fatty acids, steroids, prenol lipids, and sphingolipids. Our results thus demonstrate that with sufficient dimensionality reduction, this 'predictive metabolomic' approach should provide useful insights into the thousands of community profiles for which only metagenomes are currently available.