Experimental and background bias on pathway analysis of proteomic experiments. A simulation study,

FRESNO, CRISTOBAL; LLERA, ANDREA SABINA; GIROTTI, MARÍA ROMINA; BALZARINI, MÓNICA; PRADA, FEDERICO; FERNÁNDEZ, ELMER ANDRÉS

Quilmes, Argentina

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

Asociación Argentina de Bioinformatica y Biologia Computacional

BackgroundA common challenge when using high-throughput proteomic data is the use of bioinformatic toolswhich were mainly developed for genomic experiments[1]. Tools for the discovery of enrichedpathways, in a particular biological condition, need a reference gene set (RGS) to contrast with theinput data. In proteomic studies this reference varies with the technology used for analysis; however, acommonly used approach is to use the complete set of genes of a genome as a universal reference. Wehave explored the impact of this approach in proteomic-based pathway enrichment analysis, and weshow here the effects of using different RGSs.Materials and methodsUsing DIGE we have identified a list of differentially expressed proteins (DEP) in two humanmelanoma cell-based experiments in which the protumoral protein SPARC, normally secreted by tumorcells, was knocked out by RNA interference. Gene enrichment analysis was done using the DAVID[2]under different RGSs: DAVID´s human reference and two assay-based references. We selected one ofthe KEGG[3] pathways affected by SPARC downregulation, to devise the effect of the size of RGS inthis path. Two biases usually seen in proteomic experiments were tested: 1) "Background Bias" (BB)refers to the size change of the used contrast. 2) "Experimental Bias" (EB), appears when theexperiment only allow the expression of a specific set of proteins (i.e secreted , membrane orintracellular ones) eliminating the chance to detect all the proteins in the path. Numerical simulation ofthe BB effect was done modifying the number of genes in the RGS (keeping constant the genes in thepath) and EB was simulated modifying the number of genes in the path not in the DEP list and keepingalmost constant the RGS size. The effect on these two biases in the enrichment results was evaluatedwith the chi squared statistic.ResultsThe enrichment of pathways strongly varies when the RGS is changed. The pathway enrichmentincreases in a linear fashion when the number of genes in the reference increases (i.e. BB). The slope ofthis enrichment tendency augments or diminishes when the number of genes in the path but not in theDEP decreases or increases, respectively (i.e. EB).ConclusionIn proteomic studies both BB and EB could impact the analysis of emerging pathways. BB couldproduce false positive candidate pathways due to an increase of the number of genes outside the path.EB could produce false negative candidate pathways by means of an artificial increase of the pathlength.