Construction and Validation of HQSAR models for cell permeability, human intestinal absorption and oral bioavailability

Early stages of drug development benefits from pharmacokinetic studies, making research more efficient and less expensive.¹ Accordingly, HQSAR models were generated (Sybyl X 2.1) using three known datasets with experimentally determined properties for Caco-2 cellular permeability (P, 1272 compounds),2 human intestinal absorption (HIA, 506 compounds)3 and oral bioavailability (F, 261 compounds).4 Datasets were curated and had their molecular structures determined, followed by descriptors calculation, such as logP, molecular weight, H-bond donor and acceptors, PSA, number of rotatable bonds, fingerprints and others. Molecules with drug-like properties (that fit at least 3 of Lipinski’s rules) were then separated into two sets for model generation: training and test (4:1) through HCA method. For each subset, 1140 models were exhaustively constructed, by using different parameter combinations and were considered internally robust when calculated Q2 > 0.6 (according to OECD). Selected models, from all datasets and curation (complete and drug-like datasets), will undergo subsequent external validation, however Table 1 results shows that both P and HIA models were acceptable (Q2 >0.6) and preliminary results indicate that HIA model has acceptable predictive power (Q2F1, Q2F2, Q2F3 and CCC coefficients are all above 0.6). Onward, physicochemical interpretation of all six models will be carried and generated models will be employed in a consensus model for general permeability.