q-RASAR modeling of antibiotics-microplastics mixtures: Towards predictive aquatic toxicology and resistance risk
Shubha Das1, Shruti Mukhopadhyay2, Mainak Chatterjee3
1Drug Discovery and Development Laboratory (DDD Lab), Department of Pharmaceutical Technology, Jadavpur University, Kolkata 700032, India.
Abstract:
Microplastics (MPs) and antibiotics are emerging pollutants that frequently co-occur in aquatic environments, where their interactions intensify ecotoxicological risks and may accelerate the spread of antibiotics resistance. Experimental assessment of such sorption-driven environmental behavior of mixture is costly, time-intensive, and ethically constrained, underscoring the need for predictive computational approaches. We have reported a Partial Least Squares (PLS)-based quantitative Read-Across Structure-Activity Relationship (q-RASAR) framework to evaluate the sorption-driven toxicity of antibiotics-microplastics mixtures. The distribution coefficient (log Kd) was selected as the endpoint, reflecting partitioning between aqueous and plastic phases, a key determinant of environmental persistence and bioavailability. A curated dataset of antibiotics-microplastics mixtures was used to generate mixture descriptors based on additivity, squared, and norm-based rules. Descriptors reduction and q-RASAR integration yielded a final model with five hybrid descriptors (structural + similarity-based). The model exhibited strong internal robustness (Q2LOO = 0.761) and high external predictivity (Q2F1 = 0.832; MAEtest = 0.152). Applicability domain and Y-randomization confirmed statistical soundness. Mechanistic analysis indicated that hydrogen-bond donors, oxygen-based polar functionalities, and terminal unsaturation enhance adsorption, while steric hindrance and charge asymmetry reduce binding affinity. To address limited experimental data, we further designed 111 hypothetical mixtures and validated predictions using the Prediction Reliability Indicator (PRI) tool. Most predictions were classified as "Good" and within the applicability domain, confirming the framework's scalability and reliability. By combining mechanistic interpretability with strong predictive power, the framework enables high-throughput virtual screening and supports early-stage environmental risk assessment, data-gap filling, and regulatory decision-making in line with OECD guidelines.
