11. Robustness to realistic data problems
11. Robustness to realistic data problems
Two realistic data problems occur in the run itself: 14/182 responses are missing from IID+TRRUST, and directed TRRUST edges could be wrong. Dropping all 531 subnetwork TRRUST edges moves ranks by rho = 0.945 (top-20 16/20) — regulatory direction matters, but the ranking survives its removal. The TRRUST directions here are annotations, not random: this ablation is a genuine test of their contribution. Sign of regulation is still ignored throughout.
Assay supplement ()
Sema5a 0% (Gö-6976) to Col6a1 94.7%.
Nanomolar potencies from the dissemination screen.
IC50 against dissemination.
All 18 genes with a dissemination readout.
dissemination % with NetPert ranks, strongest 9 of 18.
dissemination % with NetPert ranks, weakest 9 of 18.
inhibition ordered with NetPert ranks annotated.