Modelo de multiclasificador basado en selección de rasgos para predecir la afinidad de antagonistas por receptores de adenosina A2B
Fecha
2011-06-20
Autores
Franco Montero, Pedro Enrique
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Los receptores de adenosina de tipo A están ampliamente distribuidos a través del cuerpo humano y por lo tanto envueltos en numerosos procesos fisiológicos y fitopatológicos. Dentro de estos el subtipo A2B ha motivado interés como una potencial diana farmacológica para la investigación y desarrollo de nuevos medicamentos. Precisamente, con este trabajo se pretende predecir la afinidad de antagonistas por receptores de adenosina de tipo A2B mediante el uso de técnicas de Inteligencia Artificial en específico de clasificación. Para esto se diseñó e implementó un modelo multiclasificador basado en selección de rasgos, el cual combina varios modelos de clasificación, los cuales son entrenados con rasgos diferentes. Para la combinación de las salidas de los clasificadores de base se utilizan tres funciones matemáticas: promedio de probabilidades, voto mayoritario y máxima probabilidad. Finalmente, este multiclasificador fue adicionado al Weka como un clasificador más.
Se utilizaron métodos de filtrado y envoltura para seleccionar subconjuntos de rasgos. La selección de los clasificadores de base se hace de acuerdo a un grupo de medidas de diversidad. Para el cálculo de estas medidas se diseñó e implementó una herramienta que permite combinar los resultados de varios clasificadores individuales y proponer combinaciones de los mismos. Se entrenaron y se evaluaron multiclasificadores con el modelo propuesto utilizando validación cruzada con 10 subconjuntos y una base como validación externa. Además se realizó una comparación del multiclasificador propuesto con varios clasificadores individuales así como con otros multiclasificadores clásicos, obteniendo los mejores resultados con el multiclasificador propuesto.
Adenosine receptors type A are widely distributed throughout the human body and therefore involved in many physiological and phytopathological processes. Among these, subtype A2B has motivated interest as a potential drug target for new drugs research and development. Precisely, the aim of this work is the prediction of the affinity A2B adenosine receptor antagonists using specifically Artificial Intelligence classification techniques. In order to accomplish this, a multiclassifier model which combines several classification models was designed and implemented based on features selection which is trained with different features. Three mathematical functions (probabilities average, majority vote and maximum probability) are used to combine the outputs of base classifiers. Finally, this multiclassifier was added to Weka as another classifier. Filtering and packaging methods were used to select features subsets. The selection of base classifiers is carried out according to a group of diversity measures. A tool that allows to combine the results of some individual classifiers and to propose combinations for them was designed and implemented in order to calculate the before-mentioned measures. Multiclassifiers were trained and evaluated with the proposed model using cross-validation 10-fold and a base as external validation. In addition, a comparison on the proposed multiclassifier with several individual classifiers was carried out as well as with other classic multiclassifiers, obtaining the best results with the proposed multiclassifier.
Adenosine receptors type A are widely distributed throughout the human body and therefore involved in many physiological and phytopathological processes. Among these, subtype A2B has motivated interest as a potential drug target for new drugs research and development. Precisely, the aim of this work is the prediction of the affinity A2B adenosine receptor antagonists using specifically Artificial Intelligence classification techniques. In order to accomplish this, a multiclassifier model which combines several classification models was designed and implemented based on features selection which is trained with different features. Three mathematical functions (probabilities average, majority vote and maximum probability) are used to combine the outputs of base classifiers. Finally, this multiclassifier was added to Weka as another classifier. Filtering and packaging methods were used to select features subsets. The selection of base classifiers is carried out according to a group of diversity measures. A tool that allows to combine the results of some individual classifiers and to propose combinations for them was designed and implemented in order to calculate the before-mentioned measures. Multiclassifiers were trained and evaluated with the proposed model using cross-validation 10-fold and a base as external validation. In addition, a comparison on the proposed multiclassifier with several individual classifiers was carried out as well as with other classic multiclassifiers, obtaining the best results with the proposed multiclassifier.
Descripción
Palabras clave
Modelo de Multiclasificador, Selección de Rasgos, Predicción, Afinidad de Antagonistas, Receptores de Adenosina A2B, Inteligencia Artificial