"Code and Data for the manuscript entitled Benchmarking Causal Discovery: Ground Truth and Interventional Validation"

"Most causal discovery benchmarks rely solely on consensus ground truth, leaving their ability to recover true causal mechanisms unclear. Using the Sachs protein signaling network, we benchmarked six causal discovery methods against both a gold-standard network and, critically, direct interventional data. Our results reveal a striking gap: only 16% of predicted edges survive interventional validation. DirectLiNGAM and ICALiNGAM offer the best balance of precision and recall, while deep learning methods produce substantially more false positives. We identify PKC\u2192Jnk as the most reliably inferred causal relationship. This study establishes a rigorous, intervention-based benchmark, provides practical guidance for method selection in small-sample biology, and demonstrates that agreement with domain knowledge alone is an insufficient measure of causal validity."