Overview

• Researchers detailed PDGrapher in Nature Biomedical Engineering and made the tool freely available for the research community.
• Using a causally inspired graph neural network that integrates protein–protein and gene regulatory networks, the system prioritizes single or paired targets predicted to reverse disease phenotypes.
• Across many held-out datasets spanning multiple cancers, the model outperformed competing approaches and recovered known drug targets excluded from training.
• The tool aligned with clinical and preclinical evidence by highlighting KDR (VEGFR2) in non-small cell lung cancer and prioritizing TOP2A in certain tumors.
• Beyond target ranking, the model traced how drugs such as vorinostat and sorafenib shift gene networks, and the team is now applying it to Parkinson’s, Alzheimer’s, and X-linked dystonia-parkinsonism in collaboration with clinical partners, with clinical use contingent on further validation.