A Constrained NMF Approach to Analyze Quantitative Metagenomic Data*

In this paper, we propose a new method for inferring the metabolic potential of microbial ecosystems based on gene frequencies generated from shotgun metagenomic data. Our approach is based on Non-Negative Matrix Factorization with constraints accounting for prior biological knowledge of bacterial metabolism. The problem is solved using efficient accelerated projected gradient methods. The approach is illustrated on a toy model and on real data on fiber metabolism by the gut microbiota in humans. We show how this approach leads to the inference of biologically relevant gene clusters.