Modeling GABA-mediated stress in pomegranate using integrated structural equation modeling and Machine Learning

The concurrent occurrence of abiotic stress represents one of the most serious challenges to global crop production, triggering complex, non-linear physiological disruptions that conventional statistical methods cannot adequately describe. This study investigates the potential of exogenous application of γ-aminobutyric acid (GABA) to mitigate combined drought and salinity stress in pomegranate (Punica granatum L.) over 45 days, using an integrative analytical approach that integrates Structural Equation Modeling (SEM) with machine learning (ML) methods. Random Forest (RF) and Support Vector Regression (SVR) were employed to examine associations among variables and capture non-linear predictive patterns, while SHapley Additive exPlanations (SHAP) enhanced model interpretability. Combined stress induced severe oxidative damage, significantly reducing the Fv/Fm, photosynthetic pigments, and stomatal traits. Application of 40 mM GABA effectively mitigated these effects, maintaining photosynthetic integrity and promoting secondary metabolite accumulation, including phenolics, flavonoids, and anthocyanins. SEM analysis revealed a strong negative effect of combined stress on pigment-related traits (β = -0.822), while GABA was associated with a positive effect (β = 0.352). ML models demonstrated exceptional predictive performance using SEM-derived latent variables, with SVR achieving R² values of 0.980 ± 0.003 for biochemical traits, 0.926 ± 0.011 for pigments, and 0.955 ± 0.013 for PSII efficiency. SHAP analysis revealed that combined stress (0.522) and GABA (0.263) were the most influential predictors of pigment retention within the ML approach. These findings demonstrate that integrating SEM with ML provides a useful and interpretable approach for understanding complex stress responses, as well as a predictive strategy for sustaining pomegranate tolerance and a generalizable approach to studying crop stress responses.