Refinement of AlphaFold2 Models against Experimental Cryo-EM Density Maps at 4-6Å Resolution.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Cryo-electron microscopy (cryo-EM) is crucial for determining molecular structures.
• Interpreting low-resolution cryo-EM maps (4-6 Å) remains challenging for accurate model building.
• Deep learning models like AlphaFold2 show promise in protein structure prediction.

Purpose of the Study:

• To evaluate the utility of AlphaFold2 for interpreting low-resolution experimental cryo-EM maps.
• To assess the strengths and limitations of combining AI-predicted models with experimental data.
• To investigate factors influencing successful refinement of AI models in cryo-EM maps.

Main Methods:

• Utilized experimental cryo-EM map/model pairs in the 4-6 Å resolution range.
• Assessed AlphaFold2 model accuracy using TM-scores.
• Evaluated the refinement performance of AlphaFold2 models within Phenix, focusing on experimental maps.

Main Results:

• AlphaFold2 models were highly accurate for most larger protein chains (TM-scores > 0.9).
• One small protein chain (115 residues) was poorly predicted (TM-score 0.52).
• Refinement success depended on AlphaFold2 prediction quality, cryo-EM data quality, and model-density alignment.

Conclusions:

• Deep learning approaches like AlphaFold2 are valuable tools for interpreting low-resolution cryo-EM data.
• Successful integration requires careful consideration of AI model accuracy and experimental map quality.
• Further development is needed to optimize AI model refinement in challenging cryo-EM datasets.