Ariel Larey, Elay Dahan, Amit Bleiweiss, Raizy Kellerman, Guy Leib, Omri Nayshool, Dan Ofer, Tal Zinger, Dan Dominissini, Gideon Rechavi, Nicole Bussola, Simon Lee, Shane O'Connell, Dung Hoang, Marissa Wirth, Alexander W. Charney, Nati Daniel, Yoli Shavit
JEPA-DNA is a new framework for genomic foundation models that improves understanding of genomic sequences by integrating high-level functional embeddings with traditional generative objectives.
Current genomic models often focus on small details and miss the bigger picture of how genes work together. JEPA-DNA is a new approach that changes this by teaching models to understand not just individual parts of DNA, but also their overall function and how they interact. This makes the models better at predicting and understanding biological processes. Tests show that JEPA-DNA outperforms other models in both supervised learning and tasks where it has to make predictions without prior examples.
