QSAR Models for Predicting Oral Bioavailability and Volume of Distribution and Their Application in Mapping the TK Space of Endocrine Disruptors

Toxicokinetic (TK) properties are essential in the framework of chemical risk assessment and drug discovery. Specifically, a TK profile provides information about the fate of chemicals in the human body. In this context, Quantitative Structure-Activity Relationship (QSAR) models are convenient computational tools for predicting TK properties. Here, we developed QSAR models to predict two TK properties: oral bioavailability and volume of distribution at steady state (VD_ss). We collected and curated two large sets of 1712 and 1591 chemicals for oral bioavailability and VD_ss, respectively, and compared regression and classification (binary and multiclass) models with the application of several machine learning algorithms. The best predictive performance of the models for regression (R) prediction was characterized by a Q²_F3 of 0.34 with the R-CatBoost model for oral bioavailability and a geometric mean fold error (GMFE) of 2.35 with the R-RF model for VD_ss. The models were then applied to a list of potential endocrine-disrupting chemicals (EDCs), highlighting chemicals with a high probability of posing a risk to human health due to their TK profiles. Based on the results obtained, insights into the structural determinants of TK properties for EDCs are further discussed.