Insider Brief
- Harvard Medical School researchers have developed PDGrapher, an AI tool that identifies drug targets capable of reversing disease states in cells, potentially transforming drug discovery.
- Tested on 19 datasets across 11 cancer types, PDGrapher outperformed other AI models—ranking correct targets up to 35% higher and delivering results up to 25 times faster—while surfacing both known and novel treatment candidates.
- Backed by federal grants and industry partnerships, the model is already being applied to cancer, Parkinson’s, Alzheimer’s, and rare neurodegenerative disorders, with long-term potential for personalized combination therapies.
Researchers at Harvard Medical School have built an artificial intelligence tool that can pinpoint drug targets to reverse disease in cells, offering a faster and more accurate path for drug discovery.
The model, called PDGrapher, was unveiled in Nature Biomedical Engineering and, according to Harvard, was backed by a mix of federal funding and private research support, including NIH, NSF, ARPA-H, the Chan Zuckerberg Initiative, the Gates Foundation, and partnerships with major pharmaceutical companies such as AstraZeneca, Roche, Sanofi, and Pfizer.
The researchers designed PDGrapher as a graph neural network that maps the relationships between genes, proteins, and signaling pathways. Instead of looking at one protein target at a time, the system identifies multiple genetic drivers of disease and predicts how turning them off or modifying them would shift a diseased cell back toward healthy function, researchers said.
“Traditional drug discovery resembles tasting hundreds of prepared dishes to find one that happens to taste perfect,” noted study senior author Marinka Zitnik, associate professor of biomedical informatics in the Blavatnik Institute at HMS. “PDGrapher works like a master chef who understands what they want the dish to be and exactly how to combine ingredients to achieve the desired flavor.”
In tests across 19 datasets spanning 11 cancer types, PDGrapher accurately highlighted drug targets already known to work while also surfacing new candidates supported by emerging evidence. For example, it identified KDR (VEGFR2) as a promising target in non-small cell lung cancer and flagged TOP2A, an enzyme already linked to chemotherapy, as a potential therapy for metastases.
The tool consistently outperformed comparable AI systems, ranking correct therapeutic targets up to 35% higher and producing results up to 25 times faster.
The new approach could help drugmakers bypass the trial-and-error bottleneck in drug discovery, reseachers pointed out. Traditional methods focus on one protein at a time, which works for some drugs, such as kinase inhibitors, but falters when diseases are driven by complex interactions across multiple pathways. By contrast, PDGrapher takes aim at the combinations most likely to restore normal cell function, a step that researchers said could speed up development timelines and expand treatment options for difficult diseases.
The researchers noted particular promise for cancers and neurodegenerative diseases, where single-target therapies often fail. The team is already applying PDGrapher to Parkinson’s, Alzheimer’s, and is working with the Center for XDP at Massachusetts General Hospital on applying it to X-linked Dystonia-Parkinsonism, a rare inherited disorder. In the long term, they envision tailoring the tool to individual patient profiles, allowing clinicians to design personalized drug combinations.
“Our ultimate goal is to create a clear road map of possible ways to reverse disease at the cellular level,” said Zitnik.
