Molecular Determinants of Caco-2 Permeability Using 2D and 3D Descriptors for Drug Discovery

Abstract

Human intestinal absorption (HIA) is an important consideration in drug discovery, where permeability across intestinal epithelial cells can influence the bioavailability of orally administered compounds. This study investigated the molecular determinants of Caco-2 permeability and evaluated the complementary contribution of three-dimensional (3D) molecular information. Caco-2 permeability data from the ChEMBL database were classified into high and low permeability using a Papp threshold. A total of 212 molecular descriptors, including 2D RDKit and 3D WHIM descriptors, were generated and evaluated using ensemble classification methods. Ablation analysis showed that 2D descriptors alone produced comparable internal performance; however, independent validation accuracy decreased to 0.577–0.579 compared with 0.689–0.715 for the combined 2D and 3D descriptor models. Descriptor and SHAP analyses identified lipophilicity, polarity, hydrogen-bonding capacity, molecular shape, and spatial organization as important factors influencing permeability. These findings suggest that 2D descriptors capture much of the primary permeability-related information, while 3D WHIM descriptors provide complementary structural information that improves prediction of unseen compounds and supports early compound prioritization in drug discovery.