Fig. 3: Robustness of RSS-NET to model mis-specification in enrichment analyses.
From: Modeling regulatory network topology improves genome-wide analyses of complex human traits
Here positive datasets were generated from M1 with θ > 0 and σ2 > 0 (Fig. 2c). Negative datasets were simulated from four scenarios where genetic associations were enriched in: a a random set of near-gene SNPs; b a random set of near-RE SNPs; c SNPs with MAF- and LD-dependent effects; d a random edge-altered network. By this design, RSS-NET was mis-specified in all four scenarios. Similar to positive datasets, the simulated false enrichments in all negative datasets manifested in both association proportion (more frequent) and magnitude (larger effect). RSS-E was excluded here because of its poor performance shown in Fig. 2c. The rest is the same as Fig. 2. Simulation details and additional results are provided in Supplementary Figs. 3–6.
