Modeling the mu-opiod receptor affinity of synthetic 8-aminocyclazocine analogues using TAE, PEST and PAD descriptors and machine-learning methods


Lingling Shen1, Curt M Breneman2, N Sukumar2, Mark P. Wentland2, and Mark J. Embrechts3. (1) Chemistry, Rensselear Polytechnic Institute, 110 8th St, Troy, NY 12180, (2) Department of Chemistry, Rensselaer Polytechnic Institute, Cogswell Laboratory, 110 8th Street, Troy, NY 12180-3590, (3) Decision Sciences and Engineering Systems, Rensselaer Polytechnic Institute, 110 8th St, Troy, NY 12180
Cyclazocine was studied in the early 1970's as an analgesic, however, clinical research ceased in this area because of the short duration of analgesic action as well as undesirable side-effects. Cyclazocine is currently being tested in humans to determine if it is a potential treatment for cocaine abuse. More recently, a series of 8-aminocyclazocine analogues was synthesized that could retard this metabolic inactivation with an associated increased duration of action. To predict the mu-opiod receptor affinity of these new cyclazocine analogues, three sets of descriptors have been used, including Transferable Atom Equivalents (TAEs), Property-Encoded Surface Translator (PEST) descriptors, and PEST Autocorrelation Descriptors (PAD), where the latter two types (PEST and PAD) incorporate hybrid shape/property information. Both partial least squares (PLS) regression and kernel partial least squares (KPLS) regression were used to develop predictive models, with feature selection being accomplished using a genetic algorithm approach. The best results were obtained using PEST descriptors with GA feature selection and a linear bootstrap-PLS model, where the q2 for the validation set was found to be 0.94 and the q2 for the blind test set was seen to be 0.97.