The Human ADME Score: Quantifying the ADME Performance of Oral Therapeutics

Abstract Quantifying ADME-related drug-likeness is challenging as drug discovery expands into larger, more complex molecules. We developed the human ADME Score to assess oral drug-likeness through functional ADME properties, positioned between molecular descriptors and clinical pharmacokinetic outcomes. The continuous score (0−1) integrates experimental or machine-learning-predicted solubility, passive permeability, free hepatocyte intrinsic clearance, P-gp efflux, and plasma fraction unbound using Softmin-5, balancing minimum and average values. Datasets of 32,307 nonclinical and 588 clinical compounds with over 123 thousand ADME measurements were used for the construction and evaluation. Scores were higher in clinical than nonclinical compounds and varied across projects, target classes, and chemotypes. The score achieved the highest observed correlation and discrimination among evaluated comparators, supporting compound comparison and prioritization. Endpoint scores highlight potential ADME liabilities. Current application is best supported for molecules ≤1000 Da, reflecting chemical representation and assay validity.