Developing a quantitative structure-property relationships (QSPR) model using Caco-2 cell bioavailability indicators...

The present study aimed to measure bioavailability (BA) indicators, including epithelial barrier function, apparent permeability (P_app) and efflux ratio, of 84 types of phytochemicals using Caco-2 cell and to develop predictive model systems using machine learning with a quantitative structure-property relationship (QSPR) model based on BA indicators and an Isomeric Simplified Molecular Input Line Entry System (SMILES). Analysis of phytochemicals was carried out with a validated HPLC analytical method.

With these BA indicators, Isomeric SMILES including information such as the stereochemistry, chemical structure and properties of phytochemicals was encoded to molecular descriptors using PaDEL-Descriptor and alvaDesc. The validity of the dataset was verified using principal component analysis, leverage plot and Williams plot. In the case of transepithelial electrical resistance (TEER), R² _Train is 0.86, root mean square error (RMSE)_Train is 55.25, R² _Test is 0.63 and RMSE_Test is 74.77, respectively. Regarding the P_app, the model demonstrated strong performance on the training set with RMSE_Train of 4.54 × 10^-6 and R² _Train of 0.95 with the test set results (RMSE_Test = 6.23 × 10^-6 and R² _Test = 0.91). For the efflux ratio, the modle explains 92% of the variance with RMSE_Train of 0.39, R² _Train of 0.92, R² _Test of 0.85 and RMSE_Test of 0.71.

The present study suggests that a prediction system for bioavailability, including TEER, P_app and efflux ratio, can be developed using a QSPR model, which could contribute to advancements in discover of functional ingredients and drugs. © 2025 The Author(s). Journal of the Science of Food and Agriculture published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.