QSAR analysis of cyclooxygenase inhibitor using particle swarm optimization and multiple linear regression.

Quantitative structure-activity relationship (QSAR) models of inhibiting action of some diarylimidazole derivatives on cylcooxygenase (COX) enzyme were constructed using modified particle swarm optimization (PSO) method. As a comparison to this method, the genetic algorithm (GA) was also tested. It has been demonstrated that the modified PSO is a useful tool for variable selection comparable to GA and even superior to GA. QSAR models are constructed separately for COX-2 inhibitory activity and selectivity of COX-2 inhibition over COX-1. The spatial descriptors play a key role in the compounds' activity and selectivity to COX-2, especially Jurs descriptors. Polar interactions are the principal binding strength between compounds and COX-2 enzyme. In addition, the aqueous desolvation free energy (FH2O) value of substituent will affect the COX-2 inhibitory activity, while the charge distribution can affect the selectivity to COX-2. Copyright 2004 Elsevier B.V.