A comprehensive study on anthracycline derivatives was done. A quantitative structure–activity relationship (QSAR) study on the half-maximal inhibitory concentration ( IC50) of these analogs was developed. These antitumor compounds are used as topoisomerase II enzyme inhibitors. Genetic algorithm (GA) was applied for feature extraction. Multiple linear regression (MLR) was established based on the GA. High stability and robustness of the model were evaluated by leave-one-out cross-validation (LOO-CV), Y-randomization, and external test set (R2 = 0.879, Q2 LOO = 0.857, RMSE = 0.148, R2 max = 0.224, F = 45.4, PRESS = 0.541). This model was generalized to 29 analogs with the quantitative or qualitative in vitro observations for their pIC50 to be calculated. Good agreement between experimental observations and calculated pIC50, indicated that the model was reliable. This result also showed that probably the drug structural-specific is preferred to the cancer cell line-specific in such analogs. Furthermore, the developed model was generalized to 49 other analogs to select potent drug candidates. To do so, four criteria were used simultaneously including (i) effective inhibitory range, (ii) leverage value of structural similarity, (iii) GATS8v value as an important descriptor, and (iv) substituent effect. This approach resulted in the discrimination of 11 candidates.
