Network based differential abundance analysis: bridging community interactions and host microbiome dynamics

Abstract

Differential abundance analysis is a critical task in microbiome research, aiming to identify microbial features (e.g., Amplicon Sequence Variant (ASV), Operational Taxonomic Unit (OTU), taxa) that vary across conditions. Despite significant advancements, current leading methods (e.g., Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC), ANCOM-BC2) face challenges in robustness and reproducibility, limiting their utility in complex ecological datasets. In this work, we propose a novel network-based approach for differential abundance analysis that integrates microbial interactions to improve accuracy and interpretability. Using simulated data generated from five empirical datasets by a third-party simulator, independent of all methods tested, our approach consistently outperforms ANCOM-BC and ANCOM-BC2 in terms of F1 scores. Beyond numerical performance, our method uses network analysis to uncover drivers of differential abundance, offering insights into microbial interactions and causal links with environmental or pathological factors. For example, we identify potential endogenous ecological drivers and exogenous influences that traditional binary classifications might overlook. This capability broadens the scope of microbiome research, enabling a deeper understanding of microbial ecology and its connection to host health and environmental conditions. Our findings highlight the potential of network-based approaches to advance both the methodological and biological frontiers of differential abundance analysis.