Fuzzy rough set as feature selection for QSAR modeling of 2,4,5-trisubstituted imidazoles, nontoxic modulators of P-glycoprotein mediated multidrug resistance

In cancer chemotherapy, multidrug resistance (MDR) is a major clinical problem which occurs by an influential mechanism and which leads to the failure of cancer chemotherapy and/or a relapse of the cancer. In this study, Fuzzy Rough Set and Genetic Algorithms were compared as variable selection techniques, while both linear and nonlinear 2D QSAR models were constructed for predicting the multidrug resistance modulating potency (expressed as ED50 values) of 2,4,5-trisubstituted imidazoles, as potent and nontoxic modulators of P-glycoprotein mediated multidrug resistance. The variables to select are the proper molecular descriptors. The (linear) Multiple Linear Regression (MLR) and the nonlinear Radial Basis Function Neural Network (RBFNN) techniques were used to search for a relation between the selected descriptors and the corresponding activity. A cross-validation approach and a test set were used as internal and external model validation, respectively. The results indicate that Fuzzy Rough Set can be used as a descriptor selection technique because the obtained models have a similar predictive property compared to those where Genetic Algorithm as a common feature selection method, was applied.