Additional file 3.
The enrichment performance of the human HECS database for select HECS criteria and enrichment scores. We evaluated (1) does the precise cutoff for defining a HECS gene affect the enrichment performance and (2) for each cutoff, what values of the enrichment score seems to best minimize the false positive rate (FPR) without impacting the true positive rate (TPR). We reconstructed the HECS database by defining the HECS assignment threshold as (A) 5, (B) 10, and (C) 15 times the median. Then, from the Human U133A/GNF1H Gene Atlas dataset, we took the top 10% of the most highly expressed genes for each cell type. From this 10%, we randomly sampled between 500 to 4000 genes 3 times to create 252 gene lists. Using the same procedures described in the CTen implementation, these lists were analyzed for cell type enrichment for each HECS database constructed. The ROC curve illustrates the that sensitivity (TPR) and the FPR are not greatly affected by the HECS assignment threshold selected. Furthermore, on each figure, we show the performance expected for selected values of the enrichment score. We see that selecting enrichment scores of 2 or higher results in a reasonably low FPR but this can be significantly improved by demanding enrichments scores of ~20 before the TPR is affected.
Format: PDF Size: 41KB Download file
This file can be viewed with: Adobe Acrobat Reader
Shoemaker et al. BMC Genomics 2012 13:460 doi:10.1186/1471-2164-13-460