From Prediction to Mechanism: Explainable AI for Plant-Microbiome Multi-Omics Integration and Biological Discovery

Predicting complex phenotypes from plant-microbiome multi-omics data has advanced significantly with machine learning and deep learning. However, traditional black-box models remain limited by their focus on predictive performance rather than mechanistic biological understanding. To bridge this gap, explainable artificial intelligence (XAI) has emerged as a key framework to transform biological data analysis from correlation mapping to hypothesis-driven discovery. This review synthesizes current advances in XAI methodologies, including post hoc feature attribution techniques and intrinsically interpretable neural architectures, and their applications in plant-microbiome interactions. We highlight how XAI enables multi-scale insights by mapping model predictions to functional genes, metabolic pathways, and microbial interaction networks. Furthermore, we address critical challenges in distinguishing statistical importance from true biological causality and advocate for a closed-loop framework that continuously integrates computational reasoning with wet-lab experimental validation. Finally, we explore future avenues, including biologically informed neural networks, causal AI, and multi-omics knowledge graphs. By facilitating the transition from predictive analytics to biologically interpretable and experimentally testable hypotheses, XAI may contribute to mechanistic discovery in plant-microbiome systems.