QSRR Study of Psychiatric Drugs Using Classification and Regression Trees Combined with Adaptive Neuro‐Fuzzy Inference System
Mehdi Jalali‐Heravi, Parvis Shahbazikhah, Atieh Ghadiri‐Bidhendi · QSAR & Combinatorial Science · 2008
Abstract A new Quantitative Structure–Retention Relationship (QSRR) approach was carried out for prediction of gas–liquid retention times of 124 psychiatric drugs in whole blood on fused‐silica capillary column coated with crosslinked methylsilicone with nitrogen–phosphorus detection. After screening the descriptors, a total of 699 topological, geometric, and electronic descriptors (zero‐ to three‐dimensional) representing various structural characteristics were calculated for each molecule in the dataset. Combined method of Classification and Regression Tree (CART) as a feature selection method for the extraction of four relevant descriptors and Adaptive Neuro‐Fuzzy Inference System (ANFIS) as a modeling technique was used for the prediction of retention times of diverse set of psychiatric drugs. The Root Mean Square Errors (RMSEs) for the calibration and prediction sets are 0.457 and 0.514, respectively.