Back

Drug Repurposing through Meta-path Reconstruction using Relationship Embedding

Park, J.

2025-03-17 bioinformatics
10.1101/2025.03.15.641108 bioRxiv
Show abstract

De novo drug development are often costly, time-consuming and risky. Drug repurposing/repositioning, increasing usability of approved drugs, offers relatively high chance of success and efficiency based on the verified safety. Applying heterogeneous biological knowledge graph enabled compelling in silico drug repurposing, yet challenging to extract interaction between heterogeneous characteristics and utilizing multi-hop interactions. In this paper, I propose a PREDR model that predicts the drug-disease relationship by defining a drug-gene-disease meta-path from heterogeneous knowledge graphs and combining each heterogeneous relationship. The PREDR model reconstructs a meta-pathway based on relational information extracted by embedding each biological feature. The PREDR model outperformed compared to existing drug repurposing models in predicting drug-disease interaction, and demonstrated the effectiveness of meta-path reconstruction by showing higher performance than the result of learning and combining each heterogeneous relationships separately. The PREDR model can also explain the reaction mechanism of suggested drugs on the defined meta-pathway using heterogeneous interactions driven from reconstruction process. In predicting drug candidates for lymphoblastic leukemia disease conducted as a case study, the highest scored candidate is confirmed effectiveness of the disease in the literature, and predicted genes are verified to be targeted by both candidate and disease by various academic sources.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.