A deep learning-assisted Biological Early Warning System using composite behavioral indicators in Daphnia magna with...

Biological Early Warning Systems (BEWS) complement chemical monitoring by detecting organismal responses to water quality disturbances. However, conventional BEWS mainly detect acute responses to high-level toxic exposure, whereas sublethal toxicity remains challenging to identify under field-derived water-matrix conditions. Here, we developed a deep learning-assisted BEWS integrating You Only Look Once (YOLO)-based trajectory generation with multi-parameter behavioral analysis and evaluated its applicability using Daphnia magna and field-collected river waters from different sampling sites. Daphnia behavior was continuously monitored over seven-day exposure periods, pre-exposure, exposure, and recovery phases. River water introduced into the Daphnia chambers was collected from representing distinct field-water matrices and physicochemical conditions. Five behavioral parameters, swimming speed, speed variance, turning angle, individual dispersion, and group dispersion, were quantified and normalized as Z-score. Composite behavioral indices were exploratorily evaluated to enhance early-warning alarm performance beyond the capabilities of single-parameter metrics. The YOLO-based approach achieved high precision (0.98), maintaining stable tracking and accurate discrimination of Daphnia individuals from environmental debris and biological interferences in the observation chambers. Among behavioral endpoints, swimming speed and group dispersion showed the clearest and most consistent responses to copper-induced stress. Several exploratorily evaluated composite indicators achieved a true positive rate of 1.0 and enabled early-warning at copper concentrations as low as 10 µg/L. These exploratory observations support the potential of the proposed framework for detecting sublethal stress across different field-collected river-water matrices. Overall, this study supports the applicability of integrating AI-driven tracking with composite behavioral analysis for BEWS under variable river-water matrix conditions.