Machine-Learning-Based Prediction of Algal Density Using Algal Volatile Organic Compounds for Bloom Early Warning

Harmful algal blooms (HABs) pose severe threats to aquatic ecosystems, yet rapid and accurate prediction of algal density remains challenging. As integrated metabolites are released throughout the algal growth, algal volatile organic compounds (AVOCs) may signal bloom onset earlier than conventional indicators. This study introduced a novel approach combining proton transfer reaction time-of-flight mass spectrometry (PTR-TOF-MS) and interpretable machine learning to predict algal density through AVOCs. By analyzing an AVOC data set of Microcystis aeruginosa (n = 814) and Chlorella vulgaris (n = 834), the extreme gradient boosting model demonstrated rapid and accurate prediction of algal density (R²: 0.95-0.98), outperforming most existing models (R²: 0.38-0.99) reliant on environmental parameters. Butanal and 2-octenal were identified as biomarkers, with species-specific concentration thresholds of butanal (158.41 ppbv for Microcystis aeruginosa; 165.61 ppbv for Chlorella vulgaris) and 2-octenal (9.02 and 6.99 ppbv, respectively), below which algal density rapidly increased. Transcriptomic and enzymatic analysis revealed that metabolic reprogramming during exponential growth, characterized by enhanced photosynthesis, suppressed carbohydrate catabolism, and inhibited fatty acid degradation, collectively contributed to decreased butanal and 2-octenal production. Field validation in a natural lake preliminarily demonstrated the model's potential for HAB monitoring (67-81% bloom risk). This work established AVOCs as dynamic indicators of algal physiology and mechanistically linked metabolic shifts to bloom dynamics, offering a transformative tool for aquatic ecosystem management.