JEPA-DNA framework boosts genomic foundation models with latent semantic grounding

Researchers have developed JEPA-DNA, a new framework for training genomic foundation models. This approach integrates Joint-Embedding Predictive Architectures (JEPA) with existing generative objectives to improve how these models learn biological sequences. By focusing on predicting functional representations in a latent space, JEPA-DNA enhances semantic understanding beyond simple token reconstruction. The framework has demonstrated state-of-the-art performance across 17 genomic benchmark tasks, showing consistent gains in both linear probing and zero-shot evaluations. AI

IMPACT Enhances semantic understanding in genomic models, potentially improving drug discovery and biological research.

RANK_REASON The cluster contains a research paper detailing a new framework for training AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Ariel Larey
• Genomic Foundation Models
• JEPA-DNA

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →