🚀 New Preprint Alert!

Mélanie Courtot Director Genome Informatics & Principal Investigator

August 29, 2025

I’m excited to share that our latest work, and FIRST PREPRINT FROM THE COURTOT LAB - “LLM-based data extraction for a large cancer registry, the Ontario Hereditary Cancer Research Network,” has just been published on medRxiv.

In this study, we leverage large language models (LLMs) to automate data extraction workflows for the Ontario Hereditary Cancer Research Network—pushing forward the boundaries of scalable, accurate, and efficient health informatics in oncogenomics.

A huge thank-you to my co-authors from the Courtot lab: Andrés Felipe Melani De La Hoz , alumni Pratham Hemlani and Jochen Weile . Our amazing Ontario Hereditary Cancer Research Network (OHCRN) collaborators: Elif Tuzlali, Harriet Feilotter, Lauren Hughes, Raymond Kim and Jordan Lerner-Ellis. With help from Brandon Chan from Genome Informatics, and collabroating faculties Lincoln Stein and Benjamin Haibe-Kains. Finally thanks to our genomic labs partners: Kathy Chun, Daria Grafodatskaya, Sarah Ridd, Laila Schenkel, Amanda Smith, Andrea Vaags and Hong Wang, and many others who made this work possible!

🔗 Read the paper here.

Looking forward to feedback and next steps toward publication!

The expected time savings afforded by LEI was quantified by measuring the time needed to
perform the Report, Molecular Test and Variant extraction tasks across the nine molecular reports
by LEI, as well as by a human curator performing the task manually.LEI was capable of performing the extraction ~2 times faster than the manual process. The mean extraction time by LEI, including the time needed to review results, was 7.59 minutes, while the mean manual extraction time by the human curator was 14.88 minutes. — The expected time savings afforded by LEI was quantified by measuring the time needed to perform the Report, Molecular Test and Variant extraction tasks across the nine molecular reports by LEI, as well as by a human curator performing the task manually.LEI was capable of performing the extraction ~2 times faster than the manual process. The mean extraction time by LEI, including the time needed to review results, was 7.59 minutes, while the mean manual extraction time by the human curator was 14.88 minutes.