Predicting human oral bioavailability of a compound: development of a novel quantitative structure-bioavailability relationship.

Abstract

The purpose of this investigation was to develop a quantitative structure-bioavailability relationship (QSBR) model for drug discovery and development. A database of drugs with human oral bioavailability was assembled in electronic form with structure in SMILES format. Using that database, a stepwise regression procedure was used to link oral bioavailability in humans and substructural fragments in drugs. The regression model was compared with Lipinski's Rule of Five. The human oral bioavailability database contains 591 compounds. A regression model employing 85 descriptors was built to predict the human oral bioavailability of a compound based on its molecular structure. Compared to Lipinski's Rule of Five, the false negative predictions were reduced from 5% to 3% while the false positive predictions decreased from 78% to 53%. A set of substructural descriptors was identified to show which fragments tend to increase/decrease human oral bioavailability. A novel quantitative structure-bioavailability relationship (QSBR) was developed. Despite a large degree of experimental error, the model was reasonably predictive and stood up to cross-validation. When compared to Lipinski's Rule of Five, the QSBR model was able to reduce false positive predictions.