Prediction of retention indices of drugs based on immobilized artificial membrane chromatography using Projection Pursuit Regression and Local Lazy Regression

Abstract

The relationship between the logarithm of retention indices (log k IAM ) of 55 diverse drugs in immobilized artificial membrane (IAM) chromatography and molecular structure descriptors was established by linear and non-linear modeling methodsProjection Pursuit Regression (PPR) and lazy learning method-Local Lazy Regression (LLR). The descriptors calculated from the molecular structures by the software CODESSA and a widely accepted property parameter ClogP were used to represent the characteristics of the compounds. The best multi-linear regression (BMLR) method in the CODESSA was used to select the most important molecular descriptors from a large set of descriptors and to develop the linear and non-linear quantitative structure retention relationship (QSRR) models. By comparing these different methods, the LLR model gave the best predictive results for the drugs studied in the present work with the square of correlation coefficient (R 2 ) of 0.9540, 0.9305; root mean square error (RMSE) of 0.2418, 0.3949; for the training set and test set, respectively. It was proved that the LLR method was a promising method for QSRR modeling with good predictive capability for the retention indices of drugs in immobilized artificial membrane chromatography, and could be used in other similar chromatography research fields.

Structure–retention relationship study of arylpiperazines by linear multivariate modeling

“…In QSRR analysis, the retention ( e.g. the retention factor, $R_{\rm{M}}^0$ ) of solutes in specific chromatographic system is presented as a function of structural descriptors of the analytes 12. Most widely used linear modeling methods in QSRR are multiple linear regression (MLR), principal component regression (PCR) and partial least squares (PLS) 13, 14.…”