Unveiling molecular signatures for precision drug design: machine learning insights from trypanothione reductase, PKC-θ, and CB1.

Traditional drug discovery methods like high-throughput screening and molecular docking are slow and costly. This study introduces a machine learning framework to predict bioactivity (pIC₅₀) and identify key molecular properties and structural features for targeting Trypanothione reductase (TR), Protein kinase C theta (PKC-θ), and Cannabinoid receptor 1 (CB1) using data from the ChEMBL database. Molecular fingerprints, generated via PaDEL-Descriptor and RDKit, encoded structural features as binary vectors. Three models-Random Forest (RF), Gradient Boosting (GB), and a stacking ensemble with Ridge Regression-predicted pIC₅₀, with the ensemble achieving the lowest RMSE. Results highlight heteroatom-containing rings for TR, multiple ring systems for PKC-θ, and aromatic rings for CB1 as critical for high bioactivity. This adaptable framework accelerates drug design by pinpointing optimizable structures, enhancing efficiency in therapeutic development.