A quantitative structure–activity relationship study on CXL017 derivatives as effective drugs for cancer treatment

Abstract

AbstractA novel quantitative structure–activity relationship study was established on a set of ethyl‐2‐amino‐6‐(3.5‐dimethoxyphenyl)‐4‐(2‐ethoxy‐2‐oxoethyl)‐4H‐chromene‐3 carboxylate (CXL017) derivatives to predict their half‐maximal inhibitory concentration (IC50) values. A principal component analysis pattern was first used to divide the whole data set into training and test sets. Then, stepwise (SW) and genetic algorithm (GA) feature selection approaches were selected to choose the proper molecular descriptors which were then subjected to multiple linear regression (MLR). Accordingly, linear models consisting of seven and six molecular descriptors were constructed and their stability was further checked using leave‐one‐out (LOO) and leave group out (LGO) cross‐validation techniques. In addition, the 15 compounds of the test set were further directed to external statistical validation to assess the forecasting capability of the generated models. A simple comparison of the results obtained by the two methodologies showed that the GA‐MLR model had some superiorities versus the SW‐MLR modeling approach, for example, R2train = 0.832, R2test = 0.816, Q2LOO = 0.788 and Q2LGO = 0.796, Ftrain = 44.01, Ftest = 6.67, MSEtrain = 0.069, MSEtest = 0.053 for the computation of the relevant IC50 values. Based on the information derived from the generated models, some key features are identifiable to appraise the activity of the organic drugs that can be used to design new anticancer agents.