Identification of a preferred set of molecular descriptors for compound classification based on principal component analysis

An algorithm based on principal component analysis was investigated to clas sify molecules in a database consisting of 455 compounds with activities ag ainst seven different biological targets. Diversity profiles of these compo und sets were calculated and compared. To effectively classify compounds wi th similar biological activity, all possible combinations of 17 molecular d escriptors were tested by complete factorial analysis, and preferred descri ptor combinations were identified. High efficiency was achieved for a combi nation of a limited set of structural keys and two or three additional 2D d escriptors. The performance of the approach was compared to Jarvis-Patrick clustering.