Extended Data Fig. 6: PDGrapher’s performance is influenced by network incompleteness.
From: Combinatorial prediction of therapeutic perturbations using causally inspired neural networks
Performance in ablation studies evaluated by percentage of accurately predicted samples (a) and nDCG (b) for the synthetic datasets with a [0, 0.1, … 0.6] fraction of random edges removed in the PPI. The number of remaining edges in the network upon random edge removal are [273,319; 242,912; 212,525; 182,177; 151,811; 121,472]. See the Method section for more details.
