An individual response-based machine learning regression method to predict ecotoxicity of chemical mixtures in the absence and presence of dissolved organic matter

The coexistence of chemical pollutants elicits combined toxicity in organisms. However, the traditional risk assessment paradigms, namely Concentration Addition (CA) and Independent Action (IA), are limited to a known mode of action (MOA). Evaluating mixtures with unknown MOAs poses a significant challenge. Additionally, water environmental factors, which are seldom considered, significantly influence the behavior and effects of these mixtures, necessitating the refinement of ecotoxicological risk assessment methodologies. We propose to predict the mixture effects of chemical compounds in both the absence and presence of environmental factors based on the responses of individual chemicals using machine learning (ML) methods. To develop and validate this new methodological approach, we selected two antibiotics, ciprofloxacin (CIP) and oxytetracycline (OTC), as a case to examine their joint toxicity to Escherichia coli, Chlorella pyrenoidosa, and Daphnia magna in the absence and presence of dissolved organic matter (DOM). The concentration-response curve (CRC) derived from the neural network (NN) model was closer to the observed CRC than the CRCs from the CA and IA models, with all falling within the 95 % confidence interval of the observed CRC from a full concentration perspective. The average absolute difference in effect conenctrations between the NN model prediction and experimental observation was 11.9 %; while for the CA and IA models, the corresponding absolute differences were 34.3 % and 30.1 %, respectively. The proposed NN model, in terms of joint effect concentrations, was more accurate for predicting joint toxicity than the reference models, regardless of mixture ratios, species types, and DOM effects. This result underscores the need to consider the practicality of the individual response-based ML method in the mixture toxicity assessment of pollutants with unknown MOAs in the environment.