RepurposeGNN

Inspiration

Neglected tropical diseases (NTDs) affect over 1 billion people worldwide, yet receive less than 1% of global pharmaceutical R&D investment . For diseases like Leishmaniasis, Chagas disease, and Schistosomiasis, the economics of de novo drug development are fundamentally broken — there is no commercial incentive to spend billions developing treatments for patients who cannot pay.

The scientific reality is equally frustrating. Effective treatments for many NTDs likely already exist among approved drugs, but manually screening millions of potential drug-disease combinations is computationally infeasible. Existing machine learning approaches for drug repurposing often ignore 3D molecular structure, fail to generalize across different organisms, or require proprietary datasets that academic researchers cannot access

What it does

RepurposeGNN is a heterogeneous graph neural network that predicts novel drug-disease associations by modeling biomedical knowledge as a graph where drugs, diseases, and proteins are nodes, and their known interactions are edges.

Given a disease of interest, the model outputs a ranked list of existing approved drugs most likely to have therapeutic effect, enabling rapid prioritization for experimental validation

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for RepurposeGNN

Built With

• graph
• machine
• ml
• python