Fig. 5

Performance evaluation of five adjustment methods and the raw data applied to six representative tissues. ROC curves and their corresponding AUC values are presented. ROC curves [30] are graphical representations of both specificity and sensitivity that take into account both the gene-gene co-expression of the adjusted dataset against the gold standard, a-priori knowledge of true and false gene-gene associations derived from the GIANT project [18]. (a) Performance evaluation for the Adipose Subcutaneous tissue dataset. Performance was evaluated using 2975 gold standard edges (1796 and 1179 true and false edges respectively) for this tissue. (b) Performance evaluation for the Skin - Not Sun Exposed (Suprapubic) dataset. Performance was evaluated using 2986 gold standard edges (1820 true and 1166 false edges). (c) Plot summarizing the AUC values for six tissue datasets (x-axis) and five adjustment methods and raw data (see Methods section). It can be seen that LR (linear regression-based adjustment for known confounders) and ComBat [4] outperforms the other adjustment methods