Aiming at the visualization of interactions between treatment ands other experimental factors on HeatMaps.

FERNÁNDEZ, ELMER ANDRÉS; SALVATIERRA, EDGARDO; BIZAMA, CAROLINA; GIDEKEL, MANUEL; PODHAJCER, OSVALDO LUIS; BALZARINI, MÓNICA

Quilmes, Argentina

Congreso; ongreso de Bioinformatica y Biologia Computacional, Quilmes Mayo 12-14,2010; 2010

Sociedad Argentina de Bioinformatica y Biologia Computacional

p { margin-bottom: 0.08in; } Background One of the most used visualization methods for microarray gene expression analysis is the HeatMap, that simultaneously represents hierarchical clusters of genes and arrays/samples in a bi-dimensional image where expression values are represented by a color scale. The usual way for representing the expression levels of many expressed genes, to remove undesired technical effects, is to use the within and between arrays/samples normalized expression values as input of the HeatMap algorithm. Nowadays, more complex experiments are carried out and many different sources of systematic variation, related to experimental factor such us gender or sex of sampled experimental units, could be present. These factors could mask important biological effects when expression levels are displayed through HeatMaps. In order to improve the visualization of interactions between treatment and other experimental factor effects, we propose the use of residuals from an additive reduced model over those genes where the treatment effect is present, that is differentially expressed genes. The methodology is shown in two cancer studies where we explore the interaction effects between treatment (tumor and adjacent tissue) and sex of patient from who tissue samples were drawn. For each gene the following model is proposed where is the overall gene expression mean, T, S and TS stands for treatment, sex and their interaction effect, is random term for the patient effect, and the the last term is a random error. Here we are interested in those genes were . Classically, the heatmap is done by using Y as input values. We propose to use the residuals from the additive model, that is, where and their residuals are , as input for the HeatMap. In our case we expect that upregulated (downregulated) genes in tumor (adjacent) samples from females (males) are downregulated (upregulated) in males (females). This effect is easily seen with the proposed approach (right panel in the figure ) but not using the Y data (left panel in the figure).