Molecular descriptors for effective classification of biologically active compounds based on principal component analysis identified by a genetic algorithm

We have evaluated combinations of 111 descriptors that were calculated from two-dimensional representations of molecules to classify 455 compounds bel onging to seven biological activity classes using a method-based on princip al component analysis. The analysis was facilitated by application of a gen etic algorithm. Using scoring functions that related the number of compound s in pure classes (i.e., compounds with the same biological activity), sing letons, and mixed classes, effective descriptor sets were identified. A com bination of only four molecular descriptors accounting for aromatic charact er, hydrogen bond accepters, estimated polar van der Waals surface area, an d a single structural key gave overall best results. At this performance le vel, similar to 91% of the compounds occurred in pure classes and mixed cla sses were absent. The results indicate that combinations of only a few crit ical descriptors are preferred to partition compounds according to their bi ological activity, at least in the test cases studied here.