ContiTE: continuous manifold MoE for few-shot cross-tissue mRNA translation efficiency prediction.

ContiTE achieves a 12.09% increase in Pearson correlation and 18.22% boost in R² for cross-tissue mRNA translation efficiency prediction using a continuous manifold MoE framework.

• Why it matters: Accurately predicting mRNA translation efficiency across different tissues is vital for understanding gene regulation, but concept shifts and limited data hinder existing models from capturing tissue-specific variations effectively.
• What they did: The authors developed ContiTE, a continuous manifold mixture-of-experts model that synthesizes domain-specific weights via a context-aware hyper-router, enabling efficient few-shot cross-tissue adaptation with rapid test-time tuning.
• The result: ContiTE outperforms state-of-the-art methods, reduces negative transfer, and requires minimal target data, offering a flexible, computationally efficient approach for tissue-specific gene expression prediction and potentially other sequence-learning tasks.