Title:	TOWARDS AN EFFICIENT AND ACCURATE EVALUATION OF THE P53 FUNCTIONAL STATE: A HORIZONTAL FILTERING APPROACH
Author(s):	Felipe Viegas, Thiago Salles, Rafael Sachetto, Leonardo Rocha
ISBN:	978-989-8533-06-7
Editors:	Hans Weghorn, Leonardo Azevedo and Pedro Isaías
Year:	2011
Edition:	Single
Keywords:	Protein, p53, automatic classification, feature selection
Type:	Full Paper
First Page:	195
Last Page:	202
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	The p53 protein plays an important role in the study of human cancer, since it is responsible for controlling the life cycle of cells. In fact, it is estimated that mutations in the p53 are related to the occurrence of more than 50% of all cancers in the human beings. Hence, the study of such mutations is truly beneficial to counter-attack them. However, in vitro evaluations are expensive and time consuming, motivating us to adopt computational models to reduce such demand. These computational models automatically determine which mutants lost their cancer suppression function. In this work, we propose a methodology based on a two level horizontal filtering aimed at reducing the computational demand of these models, as well as the number of mutations to be evaluated in vitro. With this methodology, one is able to significantly reduce the attribute space to characterize the p53, by filtering out the attributes with not enough discriminative information regarding the p53 functional state. As our experimental evaluation shows, the horizontal filtering approach is able to reduce the attribute space to less than 2% of all attributes (identified as the most important by our approach), without sacrificing the automatic classification effectiveness. In fact, such strategy improved the classification effectiveness with gains up to 13% and 11% in terms of Accuracy and Macro F1. These results are promising due to, at least, two aspects. First, the aggressive reduction in the attribute space significantly reduces the number of possible combinations of mutations to be evaluated in vitro, focusing only on mutations related to the attributes identified by the filtering procedure as the most discriminative ones (ultimately reducing the experimental costs associated with the p53 in vitro analysis). Second, such reduction in the attribute space is a key aspect when learning effective classifiers, since it reduces the data dimensionality and filter out noisier information (by removing less discriminative attributes).
	Go Back