Rafał A. Bachorz Poster

The Application of AI-driven Drug Discovery Technology for Molecular Optimization of Nuclear Receptor Ligands

Rafał A. Bachorz

Simulations Plus, Inc. (NASDAQ: SLP)

Nuclear receptors (NRs) are a superfamily of transcription factors whose activity is regulated upon the binding of a specific ligand. In the human genome, 48 genes encode NRs that are implicated in various physiological processes including development, differentiation, reproduction, and homeostasis. The dysregulation of NRs can contribute to many diseases including cancer, diabetes, infertility, and others. Because they are amenable to small molecule regulation, NRs are very promising therapeutic targets. The Retinoid Orphan Receptor (ROR), especially the RORγT isoform, is a high value target for its role in regulating Th17 cell-mediated auto-immune diseases. Despite promising preclinical results and clinical trials by several pharmaceutical and biotechnology companies, no RORyT inhibitors have been approved. Compounds have failed for a myriad of reasons, including unacceptable ADMET and pharmacokinetic (PK) properties.

In this work, we utilized the new AI-driven Drug Design (AIDD) module within the ADMET Predictor® platform to address this challenge and derived novel and promising RORyT inhibitors with suitable ADMET and PK properties. The AIDD module uses chemically intelligent SMIRKS transformations to generate new molecules based on seed compounds. The generative chemistry process is channeled towards the molecules of desired properties within a multicriteria optimization loop. The objectives, including potency and selectivity at the chosen target, synthetic feasibility, ADMET, and preclinical/clinical PK endpoints are considered simultaneously. Thus, the properties of optimized molecules are on a Pareto front and become excellent candidates for experimental verification.

We first built activity and selectivity QSAR models in ADMET Predictor® using in-house and literature-derived experimental data. We then generated molecules with AIDD using these QSAR models along with our ADMET Risk™ score, synthetic feasibility calculation, and high-throughput physiologically based PK endpoints (using the integrated GastroPlus® mechanistic models for rat/human) for simultaneous optimization. We selected several compounds among the Pareto-optimal solutions and carried out molecular docking calculations with a crystallographic RORγ structure. The results were checked against the binding energy level, and the qualitative aspects of ligand-target interactions. The most promising compounds were then selected for the experimental verification of the biological potential.