Skip to main content

Table 3 Performance of the inference algorithms on the SysGenSIM networks

From: Reconstruction of large-scale regulatory networks based on perturbation graphs and transitive reduction: improved methods and their evaluation

  Noise configuration    
Inference algorithm 1 - LL 2 - LM 3 - LH 4 - ML 5 - MM 6 - MH 7 - HL 8 - HM 9 - HH
PGnew 0.7388 0.6835 0.5159 0.6735 0.6622 0.5850 0.5141 0.5218 0.4835
PGnew + TRANSWESDu,w, 0.7695 0.6906 0.5141 0.6921 0.6778 0.5921 0.5192 0.5269 0.4868
PGnew + TRANSWESDs,w, 0.7702 0.6910 0.5142 0.6923 0.6780 0.5922 0.5192 0.5269 0.4868
PGnew + TRANSWESDu,w,2 0.7335 0.5825 0.5041 0.6471 0.6410 0.5650 0.4963 0.5115 0.4737
PGnew + TRANSWESDs,w,2 0.7354 0.5929 0.5042 0.6478 0.6417 0.5653 0.4965 0.5116 0.4739
PGnew + LTRu,u 0.7561 0.6320 0.5114 0.6701 0.6583 0.5783 0.5093 0.5209 0.4816
PGnew + LTRs,u 0.7570 0.6390 0.5115 0.6705 0.6587 0.5784 0.5094 0.5210 0.4818
PGnew + LTRu,w 0.7737 0.7051 0.5285 0.6898 0.6751 0.5924 0.5196 0.5291 0.4880
PGnew + LTRS,W 0.7742 0.7057 0.5285 0.6900 0.6753 0.5925 0.5196 0.5291 0.4880
  1. Each score is the mean of the AUPR computed for the 10 networks with K 1.5 simulated according to the same noise configuration. Thresholds used by the inference algorithms are β = 2.0 and γ = 0.05 for generating PGnew, α = 0.95 for TRANSWESD·,w,, α = 1.50 for TRANSWESD·,w,2 and α = 0.15 for LTR·,w. Analogous performances for K 2 and K 2.5 are shown in Tables T1 and T2 in Additional file 1.