Análisis multivariado en la modelación antimalárica de sustancias orgánicas
Fecha
2019
Autores
Delgado González, Osvaldo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central ``Marta Abreu`` de Las Villas. Facultad de Ingeniería Química y Farmacia. Departamento de Licenciatura Química
Resumen
La malaria es una enfermedad causante de más de medio millón de muertes anuales; los fármacos utilizados convencionalmente presentan fenómenos de resistencia. Surge así, la necesidad de desarrollar nuevos antimaláricos. Los estudios in silico han demostrado ser capaces de disminuir los altos costos de los procesos de síntesis y bioensayos, así como el tiempo empleado en la identificación de compuestos efectivos contra diversas dianas terapéuticas. En la presente investigación se realizan estudios de regresión y clasificación sobre un conjunto químico (Malaria Box); donde se emplean técnicas de Inteligencia Artificial y programas hechos en casa. Se modeló la actividad antimalárica (expresada como IC50) de 317 entidades químicas, frente al parásito Plasmodium falciparum. Las estructuras químicas fueron codificadas mediante los descriptores moleculares 0D-2D implementados en el software Dragon7. Se empleó los programas hechos en casa con la metodología implementada en el software Weka para realizar la selección de atributos. Se obtuvieron 6 modelos de clasificación y 1 de regresión empleando Redes Neuronales Artificiales y Máquinas de Vectores de Soporte respectivamente. Los modelos de clasificación alcanzaron porcentajes predictivos entre 66 % - 81 %. El modelo de regresión es capaz de predecir con un coeficiente de determinación de 0.55, más del 50 % de la serie de prueba, después de extraídos compuestos outliers. Los modelos obtenidos permiten la
realización de posteriores estudios de cribado virtual para la identificación y propuesta de potenciales antimaláricos.
Malaria is a disease that causes more than half a million deaths annually. Conventionally used drugs exhibit resistance phenomena. That suggests, the need to develop new antimalarials. In silico studies have shown to be able to reduce the high costs of the synthesis and bioassay processes, as well as the time spent in the identification of effective compounds against different therapeutic targets. In the present investigation, regression and classification studies are carried out on a chemical group (Malaria Box); where Artificial Intelligence techniques and homemade programs are used. The antimalarial activity (expressed as IC50), of 317 chemical entities, was modeled against the Plasmodium falciparum parasite. The chemical structures were encoded using the 0D-2D molecular descriptors implemented in the Dragon7 software. Homemade programs were used with the methodology implemented in Weka software to perform the selection of attributes. Was obtained 6 classification models and 1 regression model using Artificial Neural Networks and Support Vectors Machines respectively. The classification models reached predictive percentages between 66 % - 81 %. The regression model is able to predict with a coefficient of determination of 0.55, more than 50 % of the test series, after extracted outliers. The models obtained allow the realization of subsequent virtual screening studies for the identification of new antimalarials.
Malaria is a disease that causes more than half a million deaths annually. Conventionally used drugs exhibit resistance phenomena. That suggests, the need to develop new antimalarials. In silico studies have shown to be able to reduce the high costs of the synthesis and bioassay processes, as well as the time spent in the identification of effective compounds against different therapeutic targets. In the present investigation, regression and classification studies are carried out on a chemical group (Malaria Box); where Artificial Intelligence techniques and homemade programs are used. The antimalarial activity (expressed as IC50), of 317 chemical entities, was modeled against the Plasmodium falciparum parasite. The chemical structures were encoded using the 0D-2D molecular descriptors implemented in the Dragon7 software. Homemade programs were used with the methodology implemented in Weka software to perform the selection of attributes. Was obtained 6 classification models and 1 regression model using Artificial Neural Networks and Support Vectors Machines respectively. The classification models reached predictive percentages between 66 % - 81 %. The regression model is able to predict with a coefficient of determination of 0.55, more than 50 % of the test series, after extracted outliers. The models obtained allow the realization of subsequent virtual screening studies for the identification of new antimalarials.
Descripción
Palabras clave
Sustancias Orgánicas, Epidemiología, Quimiometría, Drug Therapy, Epidemiology, Malaria