Novel Molecular Descriptors for ADME Prediction in Beyond Rule of Five Chemical Space

Beyond Rule of Five (bRo5) molecules, including proteolysis-targeting chimeras (PROTACs), macrocycles, and cyclic peptides, present challenges for conventional ADME prediction because their size, flexibility, polarity, and potential for intramolecular hydrogen bonding (IMHB) can influence solubility, membrane permeability, and downstream pharmacokinetics (PK). Here, we introduce molecular descriptors designed to encode structural features relevant to bRo5 chemical space, including long-range IMHB potential and macrocyclic topology. We also developed predictive models for three experimentally derived physicochemical properties relevant to bRo5 behavior: EPSA, ChromLogD, and ChameLogK. These descriptors and predicted properties were incorporated into machine-learning models for multiple ADME endpoints, including solubility, permeability, plasma protein binding, and intrinsic clearance. Models incorporating expanded bRo5 chemical space showed improved performance for bRo5 compounds relative to previous models. The effects of the updated ADME models were evaluated in high-throughput physiologically based pharmacokinetic (HTPK) simulations. For two PROTAC case studies not included in model training, use of the updated models produced PK predictions that were generally closer to reported experimental observations. Together, these results demonstrate the utility of computationally efficient descriptors of long-range IMHB potential and related structural features for extending ADME modeling into bRo5 chemical space and illustrate how improved representation of these properties can propagate to predictions of in vivo PK.