Building Robust Human Liver Microsomal Stability Prediction Models: Leveraging Inter-Species Correlation with Rat Data

Authors

• P Shah
• VB Siramshetty
• X Xu

Abstract

Hepatic stability plays a crucial role in drug discovery, impacting both oral bioavailability and compound elimination. The primary enzymatic system responsible for xenobiotic metabolism is the cytochrome P450 (CYP450) isoenzyme family, accounting for over 75% of known xenobiotic metabolism in humans. At NCATS, we utilize a single time-point rat liver microsomal (RLM) stability assay as a Tier I assessment, generating RLM stability data for every compound synthesized. Additionally, a high-throughput multi-time point human liver microsomal (HLM) stability assay serves as a Tier II assessment. Over recent years, we've screened >30,000 compounds from >300 projects for RLM stability and ~7,000 compounds from >100 projects for HLM stability. Our study aims to: evaluate the correlation between Tier I RLM data and Tier II HLM data; develop robust quantitative structure-activity relationship models for predicting HLM stability. Despite variations in assay and species, our analysis reveals that 81% of the compounds exhibited less than a two-fold difference, and ~ 90% of the compounds showed less than a three-fold difference. This vindicates our rationale for selecting RLM as the matrix for Tier I screening. Utilizing both traditional and advanced machine learning techniques, we identified XGBoost with RDKit descriptors as the best model for predicting HLM stability, achieving balanced accuracies >80%. Notably, incorporating RLM predictions as an additional descriptor for HLM stability prediction resulted in improved model performance. The best model along with a subset of our dataset will be made publicly available on the NCATS@ADME website (https://opendata.ncats.nih.gov/adme/) to benefit the greater drug discovery community.

Scientific Focus Area: Molecular Pharmacology

This page was last updated on Tuesday, August 6, 2024