Novel feature extraction technique for fuzzy relational clustering of a flexible dopamine reuptake inhibitor

This paper describes a novel clustering methodology for classifying over 700 conformations of a flexible analogue of GBR 12909, a dopamine reuptake inhibitor that has completed phase I clinical trials as a treatment for cocaine abuse. The major aspect of the clustering methodology includes an efficient data-conditioning scheme where a systematic feature extraction procedure based on the structural properties of the molecule was used to reduce the associated feature space. This allowed region-specific clustering that focused on individual pharmacophore elements of the molecule. For clustering of the reduced feature set, the fuzzy clustering partitional method was utilized. Due to the relational nature of the feature data, fuzzy relational clustering was employed, and it successfully detected natural groups defined by rotational minima around N(sp³)-C(sp³), O(sp³)-C(sp³), and C(sp³)-C(sp²) bonds. The proposed clustering methodology also employed several cluster validity measures, which corroborated the partitions produced by the clustering technique and agreed with the results of hierarchical clustering using the XCluster program. Representative structures which exhibited a reasonable spread of energies and showed good spatial coverage of the conformational space were identified for use as putative bioactive conformations in a future Comparative Molecular Field Analysis of GBR 12909 analogues. The clustering methodology developed here is capable of handling other computational chemistry problems, and the feature extraction technique can be easily generalized to other molecules.