2D QSAR Modeling and Preliminary Database Searching for Dopamine Transporter Inhibitors Using Genetic Algorithm Variable Selection of Molconn Z Descriptors

• Published in: Journal of medicinal chemistry 2000-11, Vol.43 (22), p.4151-4159
• Main Authors: Hoffman, Brian T, Kopajtic, Theresa, Katz, Jonathan L, Newman, Amy Hauck
• Format: Article
• Language: English
• Subjects: Algorithms; Binding, Competitive; Biological and medical sciences; Carrier Proteins - antagonists & inhibitors; Carrier Proteins - chemistry; Catecholaminergic system; Databases, Factual; Dopamine - metabolism; Dopamine Plasma Membrane Transport Proteins; Ligands; Medical sciences; Membrane Glycoproteins; Membrane Transport Proteins; Models, Molecular; Nerve Tissue Proteins; Neuropharmacology; Neurotransmitters. Neurotransmission. Receptors; Pharmacology. Drug treatments; Quantitative Structure-Activity Relationship; Radioligand Assay; Reproducibility of Results; Software
• ISSN: 0022-2623; 1520-4804
• DOI: 10.1021/jm990472s

In light of the chronic problem of abuse of the controlled substance cocaine, we have investigated novel approaches toward both understanding the activity of inhibitors of the dopamine transporter (DAT) and identifying novel inhibitors that may be of therapeutic potential. Our most recent studies toward these ends have made use of two-dimensional (2D) quantitative structure−activity relationship (QSAR) methods in order to develop predictive models that correlate structural features of DAT ligands to their biological activities. Specifically, we have adapted the method of genetic algorithms−partial least squares (GA-PLS) (Cho et al. J. Comput.-Aided Mol. Des., submitted) to the task of variable selection of the descriptors generated by the software Molconn Z. As the successor to the program Molconn X, which generated 462 descriptors, Molconn Z provides 749 chemical descriptors. By employing genetic algorithms in optimizing the inclusion of predictive descriptors, we have successfully developed a robust model of the DAT affinities of 70 structurally diverse DAT ligands. This model, with an exceptional q 2 value of 0.85, is nearly 25% more accurate in predictive value than a comparable model derived from Molconn X-derived descriptors (q 2 = 0.69). Utilizing activity-shuffling validation methods, we have demonstrated the robustness of both this DAT inhibitor model and our QSAR method. Moreover, we have extended this method to the analysis of dopamine D1 antagonist affinity and serotonin ligand activity, illustrating the significant improvement in q 2 for a variety of data sets. Finally, we have employed our method in performing a search of the National Cancer Institute database based upon activity predictions from our DAT model. We report the preliminary results of this search, which has yielded five compounds suitable for lead development as novel DAT inhibitors.