Recognizing amino acid sidechains in a medium resolution cryo-electron density map.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Accurate atomic model building into cryo-electron microscopy (cryo-EM) maps is challenging, particularly at resolutions worse than 3 Å.
• Determining the amino acid sequence of protein backbone fragments within a cryo-EM map is a critical step for model building.

Purpose of the Study:

• To develop and validate a computational method, EMSequenceFinder, for assigning amino acid sequences to protein backbone fragments in cryo-EM maps.
• To improve the accuracy and efficiency of atomic model building in cryo-EM structural studies.

Main Methods:

• EMSequenceFinder utilizes a Bayesian scoring function to rank amino acid residue types based on map density fit, resolution, and secondary structure propensity.
• A convolutional neural network was trained on millions of residue densities from cryo-EM maps and atomic models in the Electron Microscopy Data Bank (EMDB).
• The method was benchmarked on extensive datasets of known protein fragments and applied to SARS-CoV-2 cryo-EM maps.

Main Results:

• EMSequenceFinder correctly identified the amino acid sequence as the top prediction for 77.8% of tested α-helix and β-strand fragments.
• On cryo-EM maps with resolutions from 4 to 6 Å, EMSequenceFinder achieved 63.5% accuracy, outperforming three state-of-the-art methods.
• Successful application was demonstrated by threading the SARS-CoV-2 NSP2 sequence into multiple cryo-EM maps across a range of resolutions.

Conclusions:

• EMSequenceFinder provides a robust and accurate method for sequence assignment in cryo-EM density maps, significantly aiding atomic model building.
• The tool enhances integrative structure modeling by combining cryo-EM data with other biological information.
• EMSequenceFinder is available as an open-source tool within the Integrative Modeling Platform (IMP).