Rapid Inference of Direct Interactions in Large-Scale Ecological Networks from Heterogeneous Microbial Sequencing Data

Abstract

The availability of large-scale metagenomic sequencing data can facilitate the understanding of microbial ecosystems in unprecedented detail. However, current computational methods for predicting ecological interactions are hampered by insufficient statistical resolution and limited computational scalability. They also do not integrate metadata, which can reduce the interpretability of predicted ecological patterns. Here, we present FlashWeave, a computational approach based on a flexible Probabilistic Graphical Model framework that integrates metadata and predicts direct microbial interactions from heterogeneous microbial abundance data sets with hundreds of thousands of samples. FlashWeave outperforms state-of-the-art methods on diverse benchmarking challenges in terms of runtime and accuracy. We use FlashWeave to analyze a cross-study data set of 69,818 publicly available human gut samples and produce, to the best of our knowledge, the largest and most diverse network of predicted, direct gastrointestinal microbial interactions to date. FlashWeave is freely available for download here: https://github.com/meringlab/FlashWeave.jl.