Artificial Intelligence–Driven Graph Methods for Microbiome Research: A Systematic Review

ABSTRACT

The complexity of high‐throughput sequencing data in microbiome research necessitates advanced analytical methods to decipher microbial community structures and interactions. Graph‐based methods, which model taxa as nodes and their relationships as edges, have emerged as a powerful framework for this purpose. However, the rapid evolution of these techniques, from classical network analysis to modern graph representation learning, has created a fragmented and complex methodological landscape. This systematic review provides a comprehensive synthesis and structured overview of the field. We systematically review the literature and propose a three‐part classification to categorize the primary contribution of each study: (1) novel methods for microbial network representation, (2) applications of classical graph analysis, and (3) the emerging use of graph representation learning for predictive modeling. We summarize the most frequent applications within each class, including keystone species identification, community clustering, and graph‐based disease classification. Furthermore, we identify critical cross‐cutting challenges, such as the lack of standardized tools and the need for improved model interpretability. By structuring the current landscape of graph‐based methods, this review serves as a crucial guide for researchers to select appropriate models and highlights key gaps to direct future research.