Recognizing amino acid sidechains in a medium-resolution cryo-electron density map | JoVE Visualize

Area of Science:

Structural Biology
Biophysics
Computational Biology

Background:

Determining accurate atomic structures from cryo-electron microscopy (cryo-EM) maps, particularly those with resolutions worse than 3 Ångstroms, presents significant challenges.
Accurate assignment of amino acid sequences to protein backbone fragments is crucial for building reliable atomic models from cryo-EM data.

Purpose of the Study:

To develop and validate a novel computational method, EMSequenceFinder, for assigning amino acid residue sequences to protein backbone fragments within cryo-EM maps.
To enhance the process of atomic model building in cryo-EM, especially for datasets with lower resolution.

Main Methods:

EMSequenceFinder utilizes a Bayesian scoring function to rank the 20 standard amino acid types based on their fit to the cryo-EM density map, map resolution, and secondary structure propensity.
A convolutional neural network, trained on millions of residue densities from cryo-EM maps and atomic models, quantifies the fit to the density.
The method was benchmarked on over 58,000 ɑ-helix and β-strand fragments and tested on independent cryo-EM datasets.

Main Results:

EMSequenceFinder correctly identifies the amino acid sequence for 77.8% of benchmarked fragments as the top-scoring prediction.
On cryo-EM maps with resolutions ranging from 4 to 6 Ångstroms, EMSequenceFinder achieved 58% accuracy, outperforming existing methods like findMysequence (45%), ModelAngelo (27%), and sequence_from_map (12.9%).
Successful application was demonstrated by threading the SARS-CoV-2 Non-Structural Protein 2 sequence into cryo-EM maps ranging from 3.7 to 7.0 Ångstroms resolution.

Conclusions:

EMSequenceFinder provides a significant advancement in sequence assignment for cryo-EM data, facilitating more accurate protein atomic structure modeling.
The open-source availability of EMSequenceFinder within the Integrative Modeling Platform (IMP) is expected to benefit the structural biology community for integrative structure modeling.
This tool is valuable for integrating cryo-EM maps with other structural information, such as protein complex models and crosslinking data.