Métodos robustos de normalización de microarreglos de ADNc basados en la mediana

Abstract

This paper introduces two new methods to normalizing microarray expression data based on weighted median. The first approach exploits the fact that variations between replicated slides of the same experiment have impulsive characteristic and, therefore, they are better modeled by a Laplacian distribution leading to a least absolute deviation regression method for the estimation for the scaling parameter. The second approach adds robustness to a previously reported method derived using linear regression by replacing traditional correlation by a Median based correlation. The performances of the proposed methods are compared to those yielded by well-known methods reported in the literature using, as performance measure, the mean square error and the mean absolute error between the reference data set and the normalized set. Furthermore, variation of the normalized data is evaluated using boxplots.