Prediction of anti-cancer activity of 1,8-naphthyridin derivatives by using of genetic algorithm-stepwise multiple linear regression

This paper compared the QSAR modeling of anti-cancer activity of compounds 1,4-Dihydro-4-oxo-1-(2-thiazolyl)-1,8-naphthyridines and its derivatives using stepwise multiple linear regression (S-MLR) and combined genetic algorithm-multiple linear regression methods (GA-MLR(.
A set of 100 compounds with certain anticancer activity were selected from literature. All molecules were “cleaned up” and the Allinger’s MM2 force field was used for energy minimization, the semi-empirical quantum method Austin method 1 (AM1) was used for geometry optimization using the Polak-Ribiere algorithm. A large number of theoretical descriptors for each molecule were calculated using Dragon software. In order to select the best set of descriptors for QSAR modeling, GA-MLR and Stepwise-MLR as two variable selection methods were used. First the random sampling of the training sets (80% of data) were randomly taken 20 times, and the remaining molecules (20 percent of the data) were used as prediction set for external validation. Among the random samples, one of the samples with high Q2CV, Q2cal, Q2test was selected as the best train and test set. Using this train set, QSAR modeling performed using GA-MLR and Stepwise-MLR methods.
QSAR models by GA-MLR modeling had larger validated squared correlation coefficient than the obtained models by S-MLR.
According to the results, it could be concluded that the activity of similar compounds will be predictable by the obtained model