Ayush, a freshman studying Chemistry and Computer Science at Duke University, built disease-specific hypergraphs to identify drug repurposing suggestions.
The drug development pipeline for a new compound can last 10-20 years and cost over $10 billion. Drug repurposing offers a more time- and cost-effective alternative. Spending a summer at the Broad is one of the best decisions I have ever made. I learned how to turn excitement into passion, a skill I hope will guide me for the rest of my career. Computational approaches based on network graph representations, comprising a mixture of disease nodes and their interactions, have recently yielded new drug repurposing hypotheses, including suitable candidates for COVID-19. However, these interactomes remain aggregate in nature/by design and often lack tissue and disease specificity. The dilution of information may affect the relevance of node embeddings and the resulting drug-disease and drug-drug similarity scores. Building upon these hypotheses, we propose constructing and learning upon disease-specific hypergraphs in which hyperedges encode biological pathways of various lengths. We use a modified node2vec style algorithm to generate pathway embeddings. We evaluate our hypergraph’s ability to find repurposing targets for Alzheimer’s disease (AD) and compare our suggestions to a state-of-the-art knowledge graph, the multiscale interactome. Using our method, we successfully identified seven promising repurposing suggestions for AD that were ranked as unlikely repurposing targets by the multiscale interactome. Additionally, our drug repositioning suggestions are accompanied by pathway explanations. Further, making our approach disease-specific ensures our embeddings are not biased by irrelevant nodes and paths. In the future, we plan on scaling this method to 800+ diseases, potentially combining them to match diseases presented in a patient to formulate the best repurposing profile on a case-by-case basis.
Project: Formulating new drug repurposing hypotheses using disease-specific hypergraphs
Mentors: Marie-Laure Charpignon
PI: Philippakis Lab, Eric and Wendy Schmidt Center