A new artificial intelligence tool called PDGrapher is redefining drug discovery by identifying the most effective single or combined targets to correct disease processes. The work, published Sept. 9 in Nature Biomedical Engineering and supported in part by federal funding, could accelerate drug design and lead to therapies for conditions that have long resisted traditional approaches.

Instead of focusing on one protein at a time, PDGrapher pinpoints targets most likely to reverse disease, significantly speeding up discovery. “Traditional drug discovery resembles tasting hundreds of prepared dishes to find one that happens to taste perfect,” said senior author Marinka Zitnik, associate professor of biomedical informatics at Harvard Medical School. “PDGrapher works like a master chef, knowing exactly how to combine ingredients to achieve the desired flavor.”

While single-target therapies like kinase inhibitors have been successful, many diseases are driven by multiple pathways. Breakthroughs such as immune checkpoint inhibitors and CAR T-cell therapies highlight the need to address disease processes at the cellular level. PDGrapher takes this broader view, finding compounds that restore cell health even without knowing precisely which molecules they act upon.

How PDGrapher works

PDGrapher is a graph neural network that maps the intricate relationships among genes, proteins, and signaling pathways. Instead of exhaustively testing massive drug libraries, it predicts the drug or drug combinations most likely to restore normal cell function. The model simulates how dialing down certain cellular drivers would affect diseased cells and then identifies promising therapeutic targets.

Advantages of the model

Trained on datasets of diseased and treated cells, PDGrapher learned which genes to target to shift cells toward health. It was tested on 19 datasets across 11 cancer types, using both genetic and drug experiments.

The model accurately predicted known drug targets (excluded during training) and identified new candidates supported by emerging evidence. For instance, it highlighted KDR (VEGFR2) in non-small cell lung cancer and TOP2A, already targeted by existing chemotherapies, as a potential way to limit metastasis.

Compared with other tools, PDGrapher showed up to 35% higher accuracy in ranking correct targets and delivered results up to 25 times faster.

Implications for medicine

By directly seeking targets that reverse disease traits, PDGrapher can streamline drug design and focus research on the most promising options. It is especially valuable for complex conditions like cancer, where single-target drugs often fail as tumors adapt. In the future, the model could be used to tailor personalized treatment strategies based on a patient’s cellular profile.

Beyond clinical applications, the tool also sheds light on the biological drivers of disease, offering insights into why certain therapies succeed. The research team is now applying PDGrapher to neurodegenerative diseases such as Parkinson’s and Alzheimer’s, and collaborating with Massachusetts General Hospital to uncover targets for X-linked Dystonia-Parkinsonism, a rare inherited disorder.

Source: https://hms.harvard.edu/news/new-ai-tool-pinpoints-genes-drug-combos-restore-health-diseased-cells