Carnelian: alignment-free functional binning and abundance estimation of metagenomic reads

Abstract

Accurate assignment of metagenomic reads to their functional roles is an important first step towards gaining insights into the relationship between the human microbiomeincluding the collective genesand disease. Existing approaches focus on binning sequencing reads into known taxonomic classes or by genes, often failing to produce results that generalize across different cohorts with the same disease. We present Carnelian, a highly precise and accurate pipeline for alignment-free functional binning and abundance estimation, which leverages the recent idea of even-coverage, low-density locality sensitive hashing. When coupled with one-against-all classifiers, reads can be binned by molecular function encoded in their gene content with higher precision and accuracy. Carnelians minutes-per-metagenome processing speed enables analysis of large-scale disease or environmental datasets to reveal disease- and environment-specific changes in microbial functionality previously poorly understood. Our pipeline newly reveals a functional dysbiosis in patient gut microbiomes, not found in earlier metagenomic studies, and identifies a distinct shift from matched healthy individuals in Type-2 Diabetes (T2D) and early-stage Parkinsons Disease (PD). We remarkably identify a set of functional markers that can differentiate between patients and healthy individuals consistently across both the datasets with high specificity.