Funneliformismosseae symbiosis and TiO2 nanoparticles under drought in Melissa officinalis L.: a machine learning modeling approach

Plant-mycorrhizal symbiosis under combined abiotic stress and nanoparticle exposure remains underexplored. The complexity of such multifactorial systems, combined with the labor-intensive nature of physiology study, challenges conventional analytical approaches. Machine learning can contribute to further elucidating relations among physiological variables that traditional methods may miss. This offers new avenues for predicting plant responses under multifactorial stress. The symbiotic interaction of Melissa officinalis with arbuscular mycorrhizal fungus (AMF) under TiO 2 nanoparticle (NP) exposure and drought was assessed. Drought-stressed plants were treated with TiO 2 NP (0, 250, 500 ppm) with or without Funneliformis mosseae inoculation. Twenty machine learning algorithms were applied alongside conventional analyses. SMOGN augmentation was applied to address the limited sample size of the parameters, and imbalance in shoot fresh weight data distribution. Under drought, AMF + 250 ppm TiO 2 NP was most effective, increasing total fresh weight by 21.7%, total chlorophyll by 10.0%, shoot length by 30.2%, and protein content by 14.7%, while reducing malondialdehyde (MDA) by 39.1% compared to the control. The non-AMF plants treated with 250 ppm TiO 2 NP under well-watered conditions showed the highest ascorbate peroxidase activity (∼2-fold). Peroxidase activity increased by 53.0-210.0%, superoxide dismutase (SOD) by 7.6-159.4%, chlorophyll b by up to 37.9%, and the chlorophyll b/a ratio by 14–46% across treatments compared to the control. Bayesian Ridge regression achieved high predictive accuracy for shoot fresh weight (R² = 0.9194), identifying chlorophyll a and b as the most influential predictors. TiO 2 nanoparticles enhanced antioxidant defense, while mycorrhizal symbiosis promoted growth. It is suggested that their synergy contributed to mitigating oxidative stress and sustaining growth under drought. The machine learning approach applied demonstrated the potential utility of the algorithms for plant physiology studies.