Errors in CGAP xProfiler and cDNA DGED: the importance of library parsing and gene selection algorithms

摘要

Background The Cancer Genome Anatomy Project (CGAP) xProfiler and cDNA Digital Gene Expression Displayer (DGED) have been made available to the scientific community over a decade ago and since then were used widely to find genes which are differentially expressed between cancer and normal tissues. The tissue types are usually chosen according to the ontology hierarchy developed by NCBI. The xProfiler uses an internally available flat file database to determine the presence or absence of genes in the chosen libraries, while cDNA DGED uses the publicly available UniGene Expression and Gene relational databases to count the sequences found for each gene in the presented libraries. Results We discovered that the CGAP approach often includes libraries from dependent or irrelevant tissues (one third of libraries were incorrect on average, with some tissue searches no correct libraries being selected at all). We also discovered that the CGAP approach reported genes from outside the selected libraries and may omit genes found within the libraries. Other errors include the incorrect estimation of the significance values and inaccurate settings for the library size cut-off values. We advocated a revised approach to finding libraries associated with tissues. In doing so, libraries from dependent or irrelevant tissues do not get included in the final library pool. We also revised the method for determining the presence or absence of a gene by searching the UniGene relational database, revised calculation of statistical significance and sorted the library cut-off filter. Conclusion Our results justify re-evaluation of all previously reported results where NCBI CGAP expression data and tools were used.