Fig. 6

Benchmark of the proposed testing method on the E. coli and A. thaliana datasets. Boxplots compare the distributions of precision between hard-thresholding (green) and hard-thresholding followed by the removal of non significant edges as predicted by the testing procedure (purple). The 20 replicates for each configuration provide an estimation of the precision dispersion caused by randomness in GENIE3 and testing by permutations. For each organism, we investigate two appropriate connectivity densities, and three adjusted p-value thresholds (FDR). On the right of the boxplots, the number of edges kept in the final network are displayed. P-values significance of non parametric mean comparisons between the strategies are encoded as follows : 0≤*** <0.001≤** <0.01≤* <0.05≤. <0.1. The results demonstrate that the proposed testing strategy offers a robust gain in precision when using a stringent adjusted p-value threshold for edges removal. a Results for the GRN inferred on E. coli genes, validated on the regulonDB database. b Results for the GRN inferred on A. thaliana heat-responsive genes, validated on the connecTF database. Additional metrics about the number of genes, interactions to test, and computation time on DIANE’s interface are shown