Challenges with Relation Mismatches Between BioREx and PubTator3 in Biomedical Knowledge Graphs

Hi everyone 👋! My name is Leonora Vesterbacka and I am the Founding Data Scientist at Redpine, a research heavy startup based out of Stockholm, unlocking proprietary data for AI agents. One of our research initiatives is constructing a massive KG for biomedical data, building on top of all the excellent initiatives such at PubTator3 etc. To this end, we are reconstructing our own entity and relationship extraction pipeline, with the goal to reproduce the PT3 entities and relationships. Now my question is to those of you in here with experience with PT3 and biomedical KGs: Has anyone run into relation mismatches running BioREx locally vs. what PubTator3 API returns? When we feed BioREx the entities PT3 itself extracted and run it locally, we're both losing relations PT3 has published and picking up extra ones PT3 doesn't have. Curious if anyone here has hit this same lose-some/gain-some pattern running BioREx locally? Would love to chat with someone who has experienced this before. Also we are hiring! Please send me a DM if this sounds like the role for you, you would be joining the data science and engineering team at one of Swedens fastest growing startups redpine.ai/careers?ashby_jid=…