Fig. 2: Flexibility of RSS-NET to identify network-level enrichments from GWAS summary statistics.

We used a B cell-specific regulatory network and real genotypes of 348,965 genome-wide SNPs to simulate negative and positive individual-level data under two genetic architectures (“sparse” and “polygenic”). We simulated SNP effects (β) for negative datasets from the baseline model (M₀: θ = 0 and σ² = 0). We simulated β for positive datasets from the enrichment model (M₁: θ > 0 or σ² > 0) for the target network under three scenarios: a θ > 0, σ² = 0; b θ = 0, σ² > 0; c θ > 0, σ² > 0. Using the simulated individual-level data we computed single-SNP association statistics, on which we compared RSS-NET with RSS-E¹⁶, LDSC-baseline¹³, LDSC-baselineLD²⁷, and Pascal²⁶ using their default setups (Methods). Pascal includes two gene (“max”: maximum-of-χ²; “sum”: sum-of-χ²) and two pathway (“chi”: χ² approximation; “emp”: empirical sampling) scoring options. For each dataset, Pascal and LDSC methods produced P-values, whereas RSS-E and RSS-NET produced BFs; these statistics were used to rank the significance of enrichments. A false and true positive occurs if a method identifies enrichment of the target network in a negative and positive dataset respectively. Each panel displays the trade-off between false and true positives via receiver operating characteristics (ROC) curves for all methods in 200 negative and 200 positive datasets of a simulation scenario, and also reports the corresponding areas under ROC curves (AUROCs, higher value indicating better performance). Dashed diagonal lines denote random ROC curves (AUROC = 0.5). d RSS-NET, as well as other methods, does not perform well when the target network harbors weak genetic associations. Simulation details and additional results are provided in Supplementary Figs. 1, 2.