Evaluación del perfil neurotóxico de líquidos iónicos basado en técnicas de aprendizaje automático supervisado
Fecha
2015-05-20
Autores
Dominguez López, Rotceh
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas. Facultad de Matemática, Física y Computación. Departamento de Computación
Resumen
La enzima Acetilcolinesterasa (AChE) juega un papel imprescindible en la hidrólisis del neurotransmisor Acetilcolina, el cual es el responsable de la transmisión de los impulsos nerviosos. Desde la década de los 30, especialistas en las ciencias químicas han producido compuestos que son capaces de inhibir esta enzima y por tanto afectar el proceso de transmisión de los impulsos nerviosos, lo cual provoca consecuencias graves para el organismo afectado. En la actualidad se ha comenzado la producción de compuestos llamados “líquidos iónicos”; que poseen características físico-químicas atractivas para la producción de solventes que son utilizados en la sustitución de solventes moleculares tóxicos para el medio ambiente. Estudios han demostrado que algunos líquidos iónicos pueden inhibir el funcionamiento de la enzima AChE y provocar daños al sistema nervioso central, surgiendo la necesidad de evaluar el perfil neurotóxico de los líquidos iónicos utilizando la enzima AChE como indicador de neurotoxicidad. Los estudios sobre la cuantificación de la relación entre la estructura y la actividad (QSAR) surgen para realizar análisis basados en la cuantificación de dicha relación presente en compuestos químicos. En el presente trabajo fue usada la habilidad de los líquidos iónicos para inhibir la AChE y de esta manera evaluar el perfil neurotóxico de los líquidos iónicos y consecuentemente su perfil no neurotóxico. En este sentido estudios QSAR pueden proveer información útil sobre el perfil químico de los líquidos iónicos. En el desarrollo del trabajo se aplicaron multiclasificadores, como técnicas de aprendizaje automático supervisado, y como resultado se obtuvieron modelos capaces de predecir si un nuevo líquido iónico es capaz o no de inhibir la AChE. Los multiclasificadores Bagging, Boosting, Stacking y Vote fueron utilizados en la experimentación con el fin de identificar modelos QSAR predictivos. Fueron calculadas cinco medidas de diversidad para los clasificadores bases utilizados en los multiclasificadores Stacking y Vote. Finalmente se obtuvieron dos modelos que superaron el desempeño de los clasificadores individuales utilizados, razón por la que fueron seleccionados para solucionar el problema. El multiclasificador AdaBoostM1, que utiliza una red neuronal MultilayerPerceptron como clasificador base y el multiclasificador Stacking, que utiliza la combinación de clasificadores FDLA, Jrip, Kstar, NaiveBayes y SMO como clasificadores bases, fueron los multiclasificadores seleccionados.
Acetylcholinesterase (AChE) enzyme plays an important role in the hydrolysis of the neurotransmitter known as Acetylcholine, an important enzyme in transmission of nerve impulses. Since the early 30s, specialists in chemical sciences have produced compounds able to inhibit this enzyme and thus affect the process of transmission of nerve impulses, with serious consequences over the affected organism. Ionic liquids (IL) represent a chemical family with a high number of industrial applications due to their attractive physical and chemical properties, mainly for the production of safer solvents that can replace standard solvents usually toxic for humans and the environment. However, previous studies have shown that ILs can inhibit AChE, with the potential neurotoxic effects associated. In this work, we used the ability of IL to inhibit AChE to evaluate the neurotoxic profile of IL and consequently their safety profile. In this sense, quantitative structure-activity/property relationships (QSAR/QSPR) can provide useful information on the property/activity profile of chemicals based on the quantification of the relationship between chemical structure and biological activity. Specifically, ensembles classifiers were applied as part of the QSAR studies, obtaining models able to predict whether a new IL inhibit or not AChE and so, providing an in silico profiling of their neurotoxic potential. Bagging, Boosting, Stacking and Vote ensembles approaches were used to find predictive QSAR models. Five measures of diversity for base classifiers used in Stacking and Vote ensembles were calculated. Finally, two models outperformed the individual classifiers and therefore were suitable to solve the problem. The following two models were obtained: an ensemble AdaBoostM1 using a MultilayerPerceptron network as base classifier and an ensemble Stacking using the combination of the FDLA, Jrip, Kstar, SMO and NaiveBayes as base classifiers.
Acetylcholinesterase (AChE) enzyme plays an important role in the hydrolysis of the neurotransmitter known as Acetylcholine, an important enzyme in transmission of nerve impulses. Since the early 30s, specialists in chemical sciences have produced compounds able to inhibit this enzyme and thus affect the process of transmission of nerve impulses, with serious consequences over the affected organism. Ionic liquids (IL) represent a chemical family with a high number of industrial applications due to their attractive physical and chemical properties, mainly for the production of safer solvents that can replace standard solvents usually toxic for humans and the environment. However, previous studies have shown that ILs can inhibit AChE, with the potential neurotoxic effects associated. In this work, we used the ability of IL to inhibit AChE to evaluate the neurotoxic profile of IL and consequently their safety profile. In this sense, quantitative structure-activity/property relationships (QSAR/QSPR) can provide useful information on the property/activity profile of chemicals based on the quantification of the relationship between chemical structure and biological activity. Specifically, ensembles classifiers were applied as part of the QSAR studies, obtaining models able to predict whether a new IL inhibit or not AChE and so, providing an in silico profiling of their neurotoxic potential. Bagging, Boosting, Stacking and Vote ensembles approaches were used to find predictive QSAR models. Five measures of diversity for base classifiers used in Stacking and Vote ensembles were calculated. Finally, two models outperformed the individual classifiers and therefore were suitable to solve the problem. The following two models were obtained: an ensemble AdaBoostM1 using a MultilayerPerceptron network as base classifier and an ensemble Stacking using the combination of the FDLA, Jrip, Kstar, SMO and NaiveBayes as base classifiers.
Descripción
Palabras clave
Acetilcolinesterasa, Líquidos Iónicos, QSAR, Multiclasificadores, Medidas de Diversidad, Acetylcholinesterase, Ionic Liquids, QSAR, Ensembles, Diversity Measures