Algoritmos de conglomerados combinatorios novedosos en aplicaciones Quimioinformáticas
Fecha
2010-06-30
Autores
Hernández Llanes, Jorge David
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
El trabajo aborda la temática de los algoritmos de conglomerados y su aplicación en problemas de la Quimioinformática. Como tarea primaria e íntimamente ligada a los propósitos del trabajo, se propone una metodología para la selección de descriptores moleculares linealmente relevantes que garanticen por el buen rendimiento de los algoritmos a comparar. Posteriormente se proponen doce algoritmos aglomerativos jerárquicos y combinatorios, siete de los cuales son clásicos en la literatura especializada y otros cinco son novedosos en esta área del conocimiento. El objetivo fundamental del trabajo es comparar los nuevos algoritmos con el de elección o de Ward. Los experimentos se realizan empleando ocho conjuntos de datos bien establecidos y validados en la literatura de la química medicinal que incluyen compuestos con usos en el tratamiento de problemas cardiovasculares, de enfermedades mentales, antiinflamatorios, entre otros. El tratamiento estadístico de los resultados le permite concluir al autor que este grupo de actividades farmacológicas es modelizable a través de unas pocas familias de descriptores moleculares 3D y que tres de los algoritmos propuestos se comportan superiormente a la mayoría de los algoritmos clásicos y similarmente (o ligeramente superior en el caso más optimista) al algoritmo de Unión Completa (o Vecino más Lejano) y al de Ward.
This work deals with the thematic of clustering algorithms and its application in Chemioinformatics’ problems. As primary task and intimately bound to this work’s purposes, it is proposed a methodology for the selection of molecular descriptors lineally relevants that guarantee a good behavior of the algorithms to compare. Later are proposed twelve aglomeratives hierarchical and combinatorial algorithms, seven of wich are classics in the specialized literature and the other five are novel in this knowledge area. The principal objective of this work is to compare the new algorithms with the selection’s and Ward’s. The experiments are made using eight data sets well established and validated in the medical chemistry literature that include compounds with usage in the treatment of cardiovascular diseases, mental diseases, antiinflammatories, among others. The statistical treatment of the results allows to the author to conclude that this group of pharmacological activities can be modeled through a few of families of 3D molecular decriptores and that three of the proposed algorithms behave superiorly to the most of the classics algorithms and similarly (or slightly higher in the most optimistic case) to the Complete Link (or Further Neighbor) and Ward’s algorithms.
This work deals with the thematic of clustering algorithms and its application in Chemioinformatics’ problems. As primary task and intimately bound to this work’s purposes, it is proposed a methodology for the selection of molecular descriptors lineally relevants that guarantee a good behavior of the algorithms to compare. Later are proposed twelve aglomeratives hierarchical and combinatorial algorithms, seven of wich are classics in the specialized literature and the other five are novel in this knowledge area. The principal objective of this work is to compare the new algorithms with the selection’s and Ward’s. The experiments are made using eight data sets well established and validated in the medical chemistry literature that include compounds with usage in the treatment of cardiovascular diseases, mental diseases, antiinflammatories, among others. The statistical treatment of the results allows to the author to conclude that this group of pharmacological activities can be modeled through a few of families of 3D molecular decriptores and that three of the proposed algorithms behave superiorly to the most of the classics algorithms and similarly (or slightly higher in the most optimistic case) to the Complete Link (or Further Neighbor) and Ward’s algorithms.
Descripción
Palabras clave
Algoritmos de Conglomerados, Algoritmo de Ward, Quimioinformática