Predicting the Aquatic Toxicity of Pharmaceutical and Personal Care Products: A Multitasking Modeling Approach

Pharmaceutical and Personal Care Products (PPCPs) have become a significant environmental concern due to their widespread use, persistence, and potential toxicity, often referred to as forever chemicals. This study aims to develop and validate robust in silico models for predicting the aquatic toxicity of PPCPs. To do so, we resorted to the ECOTOX database and employed a Python-based tool to prepare and curate the dataset. Multitasking Quantitative Structure–Toxicity Relationship (mt-QSTR) models were then developed employing the Box–Jenkins moving average approach, incorporating both linear and non-linear frameworks based on diverse feature selection algorithms and machine learning techniques. To further improve the external predictivity, a consensus modeling approach was also implemented. The most accurate model achieved an overall predictive accuracy exceeding 85%, providing valuable insights into the structural features influencing PPCP toxicity. Key factors contributing to high aquatic toxicity included high lipophilicity, mass density, molecular mass, and reduced electronegativity. This work offers a foundation for designing safer PPCPs with reduced environmental impact, aligning with sustainable chemical development goals.