Purpose: The purpose of this investigation was to develop aquantitative structure-bioavailability relationship (QSBR) model for drugdiscovery and development.
METHOD: A database of drugs with human oral bioavailability wasassembled in electronic form with structure in SMILES format. Usingthat database, a stepwise regression procedure was used to link oralbioavailability in humans and substructural fragments in drugs. Theregression model was compared with Lipinski's Rule of Five.
RESULTS: The human oral bioavailability database contains 591compounds. A regression model employing 85 descriptors was built topredict the human oral bioavailability of a compound based on itsmolecular structure. Compared to Lipinski's Rule of Five, the falsenegative predictions were reduced from 5;pc to 3;pc while the falsepositive predictions decreased from 78;pc to 53;pc. A set of substructuraldescriptors was identified to show which fragments tend toincrease/decrease human oral bioavailability.
CONCLUSIONS: A novel quantitative structure-bioavailabilityrelationship (QSBR) was developed. Despite a large degree of experimentalerror, the model was reasonably predictive and stood up tocross-validation. When compared to Lipinski's Rule of Five, the QSBRmodel was able to reduce false positive predictions.
