Compiling the features. A. We sketch the method with a simple example pathway consisting of six reactions R1-R6 and six metabolites M1-M6. B. To apply the pattern analysis method (wavelet transforms), the pathway needed to be represented on a two dimensional lattice grid. Reactions were optimally arranged to preserve next nearest neighborhoods while minimizing the distances of neighboring reactions. Metabolites didn't need to be displayed in this representation but rather used to determine the neighborhoods (e.g. R1 is a neighbor of R2 because it produces M1 which is needed as a substrate for R2 or vice versa). C. Gene expression data was mapped onto the corresponding enzymatic reactions. In this example genes of enzymes for reactions R1, R4, R5 were high expressed and of enzymes for reactions R2 and R6 low expressed. D. Combined gene expression features were assembled by Haar wavelet transforms which basically calculated additive and subtractive combinations of 2 × 2 pixels of the grid (pixels without reactions were filled with zeros). The figure shows all four possible arrangements of 2 × 2 pixels for which the wavelet transforms were calculated. The same procedure was done for all tumor samples. The feature which best separated the tumor entities (favorable from unfavorable) was selected for the significance of this pathway.
Schramm et al. BMC Medical Genomics 2010 3:39 doi:10.1186/1755-8794-3-39