Averaging multiple causal graphs to find the right one

Finding causal structure from data is unstable: small changes in the data yield wildly different graphs. DAGgr sidesteps this by aggregating multiple candidate DAGs, weighting them by predictive accuracy on held-out data, then thresholding edge strengths to ensure the final graph remains acyclic. On synthetic networks and real protein-signaling data, this ensemble approach beats both individual models and standard bootstrap averaging.