Interpretable multi-omics machine learning reveals drought-driven shifts in plant-microbe interactions.

Area of Science:

• Plant science
• Microbiome research
• Genomics
• Metabolomics
• Machine learning

Background:

• Rhizosphere plant-microbe interactions are vital for plant growth, nutrient uptake, and stress tolerance.
• Integrating multi-omics data to understand plant-microbe symbiosis under drought stress is challenging.

Purpose of the Study:

• To integrate genomic, metabolomic, and microbiome data from soybean accessions under control and drought conditions.
• To identify environment-specific predictive features for plant phenotypes using machine learning.
• To understand drought adaptation mechanisms through plant-microbe interactions.

Main Methods:

• Integration of genomic, metabolomic, and microbiome data from 198 soybean accessions.
• Comparison of linear models (BLUP, GWAS) with nonlinear machine learning models.
• Application of SHapley Additive exPlanations (SHAP) for model interpretation and feature identification.

Main Results:

• Machine learning models outperformed linear models in capturing nonlinear relationships and variable selection.
• Daidzin (isoflavone derivative) and Candidatus Nitrosocosmicus were identified as major contributors to phenotypic variation under drought.
• SHAP-based networks revealed cross-omics interactions, including daidzin, gamma-aminobutyric acid (GABA), and Paenibacillus.

Conclusions:

• An interpretable machine learning approach successfully predicted plant phenotypes using multi-omics data.
• Identified biomarkers and interactions provide insights into plant adaptation to drought.
• Environment-dependent rhizosphere networks and symbiotic associations play a key role in drought resilience.