Toward Modeling Protein Multimers by Combining AlphaFold 3 Predictions with Secondary Structures from...

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• AlphaFold 3 (AF3) demonstrates enhanced accuracy in predicting protein multimer structures.
• Fitting predicted models to medium-resolution cryo-electron microscopy (cryo-EM) maps (5-10 Å) presents challenges due to limited high-resolution features.

Purpose of the Study:

• To evaluate the accuracy of AlphaFold 3 multimer models when fitted to medium-resolution cryo-EM maps.
• To assess the utility of cross-correlation (CC) scores and secondary structure analysis for distinguishing model quality.

Main Methods:

• A case study using four AF3 multimer models and corresponding 7-8 Å resolution cryo-EM maps.
• Comparison of AF3 models with deposited atomic structures using TM-scores.
• Analysis of cross-correlation (CC) scores between models and cryo-EM maps.
• Secondary structure segmentation using the DeepSSETracer tool.

Main Results:

• AF3 multimer models showed partial correctness, with accurate domains, secondary structures, and individual chains (16/17 chains had TM-scores > 0.5).
• Some models exhibited inaccuracies in the relative positioning of chains or domains.
• Cross-correlation (CC) scores correlated with TM-scores, indicating their sensitivity in distinguishing model quality when regions are correctly masked.
• Major secondary structures (α-helices, β-sheets) were detectable in medium-resolution cryo-EM maps.

Conclusions:

• AF3-predicted multimer models hold potential for integration with medium-resolution cryo-EM data.
• Combining CC scores and secondary structure similarity can improve the assessment of AF3 model accuracy.
• Further development is needed to refine model fitting and address domain/chain positioning errors in cryo-EM map interpretation.