Artificial neural networks analysis used to evaluate the molecular interactions between selected drugs and human α1-acid glycoprotein

Quantitative structure-retention relationships (QSRR) were proposed for α1-acid glycoprotein (AGP) column using physicochemical molecular descriptors of the selected drugs and interacting with that column. The set of 52 structurally diverse drug compounds, with experimentally derived logarithms of retention factors (log k) values was considered. Thirty-six physicochemical property descriptors were calculated by standard molecular modeling and used to establish QSRR and predict the retention data by artificial neural network (ANN). The QSRR indicated that heat of formation (HF), Moriguchi n-octanol-water partition coefficient (M log P) and the energy of the highest occupied molecular orbital (HOMO) are the most important for interactions between drugs and AGP. The proposed ANN model based on selected molecular descriptors showed a high degree of correlation between log k observed and computed. The final model possessed a 36-5-1 architecture and correlation coefficients for learning, validating and testing sets equaled 0.975, 0.950 and 0.972, respectively.