Effect of Background Correction on Cancer Classification with Gene Expression Data

This paper empirically compares six background correction methods aimed at removing unspecific background noise of the overall signal level measured by a scanner across microarrays. Using three published cDNA microarray datasets we investigated the effect of background correction on cancer classification in terms of the predictive performance of two classifiers (k-NN and support vector machine with linear kernel) induced from microarray data where a particular background correction method is applied, individually and in combination with a single-bias or double-bias-removal normalization method.