A global Quasi-SMILES model based on the Monte Carlo algorithm for assessing the multi-organism aquatic ecotoxicity of personal care products

Personal care products (PCPs) are widely used for external applications on the body, and their increased consumption has raised concerns about their potential environmental impact, particularly in aquatic ecosystems. Evaluating the aquatic ecotoxicity of PCPs is essential, but the process is a long and difficult task. Thus, it is crucial to employ tools for rapid screening. The quantitative structure-activity relationship (QSAR) approach can leverage existing data to identify potentially hazardous PCPs quickly. This study uses QSAR models to assess the aquatic ecotoxicity of 159 PCPs across three organisms’ algae, crustaceans, and fish providing a broader ecological perspective than traditional methods, which typically focus on a single organism. A QSAR model was implemented using CORAL software, which utilizes the SMILES format to predict aquatic toxicity. However, traditional SMILES do not incorporate experimental context, limiting prediction accuracy. To address this, the Quasi-SMILES method extends the traditional SMILES notation by incorporating experimental conditions related to three key organisms of the aquatic trophic level algae (Pseudokirchneriella subcapitata), crustacean (Daphnia magna), and fish (Pimephales promelas) thus enabling more accurate predictions of chemical behavior under diverse environmental conditions. Using random data splitting and multiple objective functions, 40 models were developed based on the Monte Carlo method. The model that combined the Ideal Correlation Index (IIC) and the Correlation Intensity Index (CII) as dual objective functions achieved the best predictive performance for split 4, with r m 2 = 0.7396, R 2 = 0.7757, and Q 2 = 0.7509 for validation set highlighting the effectiveness of multi-objective optimization strategies. • A global Quasi-SMILES–based QSAR model was developed to predict the aquatic ecotoxicity of 159 personal care products. • Quasi-SMILES descriptors with organism-specific conditions improved toxicity prediction vs conventional SMILES. • Monte Carlo optimization with dual objective functions (IIC and CII) enhanced model robustness and predictive accuracy. • The best model showed strong validation ( = 0.7396, = 0.7757, = 0.7509), supporting rapid ecotoxicity screening.