Fine-tuning AlphaFold with limited cryo-EM observations.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryogenic electron microscopy single-particle analysis (cryo-EM SPA) is vital for determining macromolecular structures.
• Current cryo-EM SPA methods face limitations due to insufficient particle numbers and missing views, hindering accurate structure determination.

Purpose of the Study:

• To introduce CoCoFold, a novel framework designed to overcome data limitations in cryo-EM SPA.
• To directly guide atomic model prediction using raw cryo-EM particle images integrated with AlphaFold.

Main Methods:

• CoCoFold fine-tunes AlphaFold by integrating raw cryo-EM particle images for direct atomic model guidance.
• A memory-efficient tuning strategy employs a fused attention mechanism within AlphaFold's structure module.
• A differentiable network enables end-to-end refinement of predicted structures against experimental cryo-EM data.

Main Results:

• CoCoFold demonstrates superior performance in benchmark experiments with limited particle numbers and missing views.
• The framework consistently outperforms state-of-the-art methods across various evaluation metrics.
• CoCoFold effectively refines atomic models by leveraging experimental cryo-EM observations.

Conclusions:

• CoCoFold significantly enhances the accuracy of atomic model prediction in cryo-EM SPA, particularly under data-scarce conditions.
• The integration of raw cryo-EM data directly into prediction frameworks offers a powerful approach for structural biology.