QSPR Models for Predicting Critical Micelle Concentration of Gemini Cationic Surfactants Combining Machine-Learning Methods and Molecular Descriptors

Abstract

A data set of 231 diverse gemini cationic surfactants has been developed to correlate the logarithm of critical micelle concentration (cmc) with the molecular structure using a quantitative structure-property relationship (QSPR) methods. The QSPR models were developed using the Online CHEmical Modeling environment (OCHEM). It provides several machine learning methods and molecular descriptors sets as a tool to build QSPR models. Molecular descriptors were calculated by eight different software packages including Dragon v6, OEstate and ALogPS, CDK, ISIDA Fragment, Chemaxon, Inductive Descriptor, SIRMS, and PyDescriptor. A total of 64 QSPR models were generated, and one consensus model developed by using a simple average of 13 top-ranked individual models. Based on the statistical coefficient of QSPR models, a consensus model was the best QSPR models. The model provided the highest R 2 = 0.95, q 2 = 0.95, RMSE = 0.16 and MAE = 0.11 for training set, and R 2 = 0.87, q 2 = 0.87, RMSE = 0.35 and MAE = 0.21 for test set. The model was freely available at https://ochem.eu/model/8425670 and can be used for estimation of cmc of new gemini cationic surfactants compound at the early steps of gemini cationic surfactants development.

Development of novel antipsychotic agents by inhibiting dopamine transporter –in silicoapproach

“…42,43 As the fitness evaluator, in the QSARINS program, GA is coupled with multiple linear regression (MLR). 44,45 For QSAR model development, the following parameters were tuned to the total number of features in the model (GA optimization included the number of variables) set to 4, the number of GA iterations (generations per size) set to 500, the number of models on which GA evolves (the population size) set to 10, and random mutations to generate a pool of variegated descriptors (mutation rate) set to 20% mutation rate.…”