New AI Tool Pinpoints Genes, Drug Combos to Restore Health in Diseased Cells (hms.harvard.edu)

Researchers at Harvard Medical School unveiled PDGrapher, a free graph‑neural‑network tool that identifies the genes and drug combinations most likely to reverse diseased cells back to healthy states. Unlike conventional drug discovery that tests one protein or compound at a time, PDGrapher models multi‑gene and pathway interactions to propose single or combined targets that causally drive disease phenotypes. The approach aims to accelerate drug discovery for complex, multi‑pathway conditions (notably cancer and neurodegeneration) and could eventually support individualized combination therapies. The work is reported in Nature Biomedical Engineering and was supported by a mix of federal and private funding. Technically, PDGrapher builds a graph of genes, proteins and signaling links, trains on datasets of cells before-and-after treatment, and simulates perturbations (turning targets down or off) to predict which interventions restore healthy cellular behavior. The team tested it on 19 datasets spanning 11 cancer types: the model recovered known targets withheld during training and nominated plausible new targets such as KDR (VEGFR2) and TOP2A. Compared with similar tools, PDGrapher ranked correct targets up to 35% higher and ran up to 25× faster. The authors are now applying it to Parkinson’s, Alzheimer’s and rare neurodegenerative disorders to map actionable, causal routes to disease reversal.