Reproduce “JEPA-DNA: Grounding Genomic Foundation Models through Joint-Embedding Predictive Architectures” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ariel Larey, Elay Dahan, Amit Bleiweiss, Raizy Kellerman, Guy Leib, Omri Nayshool, Dan Ofer, Tal Zinger and 10 more

JEPA-DNA adapts pretrained genomic models by adding a task that predicts a whole-sequence representation from masked DNA while retaining native token prediction. In frozen linear-probe tests across three backbones, it improves 6–8 of 9 supervised tasks per backbone and reaches state-of-the-art results on 7 of 9 tasks in the comparison. Gains vary, including regressions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.