Data-driven machine learning modeling reveals the impact of micro/nanoplastics on microalgae and their key underlying mechanisms

Micro- and nano-plastics (MNPs) pose a growing threat to freshwater microalgae, leading to water quality and biodiversity. Traditional experiments often encounter difficulties in terms of cost, time, and capturing complex interactions when exploring this critical issue. To overcome these limitations, we applied eight machine learning models to predict MNPs' effects on microalgae activity using literature data from the past decade. Of these, Extreme Gradient Boosting (XGB), optimized via Bayesian methods with 5-fold cross-validation, performed best (R² = 0.89, RMSE = 0.09) without overfitting. Key predictors included reactive oxygen species (ROS) production, MNP type and size, photosystem II activity, and microalgae species. Notably, MNP size and algal species had the most direct influence on activity, while ROS levels played a central role in mediating toxicity. Variance partitioning confirmed ROS as the most critical factor, enhancing the explanatory power when combined with other variables. Our findings also identified polyvinylchloride (PVC), particularly at sizes under 160 μm, as the most harmful plastic type. Chlorella pyrenoidosa emerged as the most sensitive species. These insights offer valuable guidance for improving MNP pollution management, developing bioremediation strategies, and refining ecological risk assessments in aquatic ecosystems.