Abstract

Application of pattern recognition methods to determine analogy of novel molecular descriptors for drug development and synthesis

Author(s): Ronald Bartzatt

The design and synthesis of new pharmaceuticals is enhanced when molecular physicochemical properties are elucidated. Pattern recognitionmethods are ideally suited to identify the underlying relationships within a multivariate numericalmatrix ofmolecular descriptors. Fourteen descriptors of twelve barbiturate drugs are analyzed by non-metricmultidimensional scaling, discriminant analysis, cluster analysis, K-means cluster analysis, selforganizing tree algorithm(SOTA), and analysis of similarity (ANOSIM) to determine the detailed relationship of the novel descriptor LogKow(drug)/ LogKow(octanol) to formula weight, polarizability, polar surface area, molecular volume, and nine other descriptors. The barbiturate class of drugs are chosen for modeling purposes due to the vast number of existing barbiturate structures and the profound variation of medicinal activity effectuated byminor structural changes.Non-metricmultidimensional scaling clearly shows that the ratio LogKow(drug)/LogKow(octanol) is extremely similar to descriptors of polarizability, index of refraction, number of oxygens, nitrogens, number of hydroxyl (-OH) and amine groups (-NHn) in drug structures. Likewise hierarchical cluster analysis determined this same conclusion. In addition, the descriptor parachor is profoundly distinct from the remaining thirteen descriptors, but with formula weight and molar volume strongly similar to each other. Discriminant analysis determined that the novel descriptor LogKow(drug)/LogKow(octanol) ismost dissimilar to descriptors molar refractivity, molar volume, formula weight, polar surface area,molecular volume, parachor, andmolecular area.ANOSIMdetermined that this group of twelve barbiturates are moderately dissimilar from each other based upon the fourteen descriptors utilized for this study.