Automated multiparameter optimization (MPO) at the point of initial drug design is a powerful emerging approach to improve and expedite drug development. We employed the AI-driven drug design (AIDD) platform to design novel RORγT ligands optimized using QSAR activity models, machine learning ADMET properties, 3D pharmacophore similarity, and synthetic difficulty predictions. We calculated several measures of novelty postdesign and then employed multicriteria decision analysis (MCDA) to select compounds for synthesis for this important drug target. We found that 19/27 (70%) of the selected compounds inhibited RORγT activity in a cell-based assay by at least 25% at 20 μM. The most potent compound had a measured IC₅₀ of 1.51 μM (the predicted IC₅₀ was 1.29 μM) and demonstrated activity in human T cells. The logP, thermodynamic solubility, liver microsome clearance, fraction unbound in plasma, and MDCK permeability of this compound were measured in vitro and shown to be close to or better, e.g., higher in vitro solubility, than the values predicted by ADMET Predictor. This compound has an indolizine scaffold that has not yet been reported in the context of the RORγT receptor, demonstrating the power of MPO in the earliest stages of drug design to create novel, active molecules already possessing ADMET properties suitable for advanced lead optimization.

By Rafał A. Bachorz, Joanna Pastwińska, Michael S. Lawless, David W. Miller, Anna Sałkowska, Kaja Karaś, Iwona Karwaciak, Jeremy O. Jones, Marcin Ratajewski