Improving visualization of interaction effects of expressed genes in microarray studies

FERNÁNDEZ E.A.; SALVATIERRA E.; BIZAMA C.; BENAVENTE F.; GIDENKEL M.; PODHAJCER O.; BALZARINI M.

Florianópolis, Brasil.

Congreso; XXVth International Biometric Conference (IBC-Floripa-2010); 2010

Visualization techniques such us heatmaps and biplots are used to explore microarray gene expression data. In both cases a simultaneous representation of genes and arrays/samples in a bi-dimensional plot is achieved. The traditional values for gene expression level analyses are the array normalized log intensities. We use both visulization techniques over residual data as a tool to explore interactions between treatments affecting gene expression and other experimental factors such sex or age of the sampled units which may change the microarray output. The residuals from a reduced (without treatment-by-factor interaction) additive main effect model fitted over differentially expressed genes were used as input values for the visualization techniques. The methodological approach is shown in a cancer study to explore the interaction effects between treatments (tumour and adjacent tissue) and sex of patients from who tissue samples were drawn. The following gene by gene additive model was proposed: Y_pst=(μ+γ_p )+T_t+S_s+TS_ts+ϵ_pst where μ is the overall gene expression mean, T,S and TS stands for treatment, sex and interaction effects, γ_p~N(0,〖σ^2〗_p) is a random term representing patient effects, and the last term is a random experimental error ϵ_pst~N(0,σ^2). Thereafter, the residuals from the reduced model, that is, Y_pst=(μ+γ_p )+T_t+S_s+〖ϵ´〗_pst where 〖ϵ´〗_pst~N(TS_(st,) 〖σ^γ〗_(ϵ^´ )) obtained as r_pst=Y_pst-{(μ ̂+γ ̂_p )+T ̂_t+S ̂_s } were used to feed heatmap and biplot procedures. The expected upregulated (downregulated) genes in tumour (adjacent) samples from females (males) which are downregulated (upregulated) in males (females) were easily observed. A crossvalidation study showed that by means of this approach the interaction effects are clearly visualized.