Descriptor selection evaluation of binary gravitational search algorithm in quantitative structure-activity relationship studies of benzyl phenyl ether diamidine's antiprotozoal activity and Chalcone's anticancer potency

In this work, the effect of various gravitational constant functions, G(t), were evaluated on the performance of the newly proposed binary gravitational search algorithm (BGSA) as a feature selection method for chemical systems. To fulfill this aim, linear, exponential, logarithmic and square root functions were studied. These functions were implemented in the binary gravitational search algorithm computer code sequentially and their performances were evaluated in a quantitative structure-activity relationship (QSAR) study for selection of the most informative descriptors. In the QSAR study, sixty cationic benzyl phenyl ether diamidine derivatives, which had been synthesized and evaluated against acute infection of Trypanosoma brucei rhodesiense (T.b. rhodesiense), were examined. The number and the kind of descriptors, which were selected by the BGSA, were highly dependent on the applied G(t) function. The results of internal and external validation tests indicate that the exponential function was superior to the other gravitational constant functions for applying in the binary gravitational search algorithm. A general model was established using seven descriptors for the ten training and validation sets. Regardless of subsetting, the selected descriptors and generated model can successfully describe experimental variation of antiprotozoal activity of benzyl phenyl ether diamidine derivatives. In addition, in another QSAR study, anticancer potency of a series of 87 Chalcone derivatives was satisfactorily modeled by using the BGSA-BRANN method. Comparison of BGSA results with those obtained by genetic algorithm (GA) indicates superiority of the BGSA.