Quantitative Structure - Activity Relationships Study of Carbonic Anhydrase Inhibitors Using Logistic Regression Model

Abstract

Binary Logistic Regression (BLR) has been developed as non-linear models to establish quantitative structure- activity relationships (QSAR) between structural descriptors and biochemical activity of carbonic anhydrase inhibitors. Using a training set consisted of 21 compounds with known ki values, the model was trained and tested to solve two-class problems as active or inactive on the basis of the predicted value for IC50. Many quantitative descriptors were generated to express the physicochemical properties of 21 compounds with optimized structures. After filtration of these descriptors, 39 of descriptors for carbonic anhydrase (CA, EC 4.2.1.1) isozyme IX (CAIX) and 45 for isozymeXII (CAXII) remained and were selected for QSAR study. Logistic regression was then used to non-linearly select the most important descriptors and to develop a model for prediction of IC50. To evaluate the performance of the established models, Jjackknife and self consistency tests were performed during implementation of the two model-building methods. The applied indices including accuracy, sensitivity, and specificity were 85%, 82% and 100% for CAIX and also 71%, 68% and 80% for CAXII respectively. The primary advantage of such an approach is the reduction of redundant variables and the consequent improvement in the efficiency of modeling.