Prediction of caco-2 cell permeability using bilinear indices and multiple linear regression

The qualitative relationship between in vitro Caco-2 cellular transport and in vivo drug permeability allow using Caco-2 cell assay for intestinal absorption studies. In this work, atom-based bilinear indices and multiple linear regression (MLR) are applied to obtain models useful for the prediction of Caco-2 cell absorption. Making use of a previously reported database, we obtain four statistically significant MLR models, the best models shown R²=0.72 (s=0.435) for nonstochastic indices and R²=0.66 (s=0.464) for stochastic indices. No significant difference was found when comparing to previous reported studies. The models were internally validated using leave-one-out cross-validation, bootstrapping, as well as Y-scrambling experiments. Additionally, we performed an external validation using a test set, which yields significant values of R²_ext of 0.70 and 0.72 for stochastic models, showing a better predictive power. Furthermore, we define a domain of applicability for our models. These results suggest that our approach could offer an appropriate tool as an alternative to predict the absorption in Caco-2 cells in a short time and decrease experimental costs.