Bioavailability & Bioequivalence: BA/BE Studies Summit

Biography

Biography: Urban Fagerholm

Abstract

Background: Prediction of in-vivo permeability (Pe), solubility, BCS-classing, food interactions, fraction absorbed (fa) and oral bioavailability(F) from in-vitro and animal data is a challenge, especially for compounds with low/moderate Pe, efflux and/or high lipophilicity/low solubility. For such compounds, in-vivo prediction from preclinical data is generally poor/uncertain and sometimes impossible. Thus, improvements are required. Methods: With extensive, diverse datasets (log P -9 to 9), new algorithms and various computational chemistry methods (including machine learning) we have developed and validated prospective in-silico prediction models (no retrospective data fitting) for the parameters described above. Results & Discussion: Models for fa and F (including compounds with low Pe, strong efflux, very low solubility, extensive gut-wall and hepatic extraction) showedQ^2 of 0.77 and 0.55 and median prediction errors of 1.1- and 1.4-fold, respectively. In direct comparison, the models outperformed lab methods. For the 100 compounds with lowest solubility (including albendazole, danazol, loperamide, lovastatin, ketoconazole and troglitazone), 74% correct in-vivo BCS-classing and 12% average absolute prediction error for fa was obtained. The mean prediction error for AUC-changes with food was 1.4-fold. Conclusion: The new in-silico models and algorithms enable improved and simplified prospective predictions of in-vivo Pe, fa, solubility, BCS-classing, food interactions and F. Benefits include reduced and defined uncertainty, reduced time and costs, and frontloaded and improved decision-making.