Automated model building and protein identification in cryo-EM maps.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Interpreting electron cryo-microscopy (cryo-EM) maps requires significant expertise and manual effort.
• Accurate atomic model building is crucial for understanding protein function and biological mechanisms.

Approach:

• Developed ModelAngelo, a machine-learning approach utilizing graph neural networks.
• Integrated cryo-EM map data with protein sequence and structural information.
• Applied ModelAngelo to both protein and nucleotide components within cryo-EM maps.

Key Points:

• ModelAngelo builds atomic models for proteins with quality comparable to human experts.
• Achieved similar accuracy to human experts for nucleotide backbone construction.
• Outperformed human experts in identifying proteins with unknown sequences using predicted amino acid probabilities and hidden Markov model searches.

Conclusions:

• ModelAngelo automates and enhances atomic model building in cryo-EM.
• The tool reduces bottlenecks and increases objectivity in cryo-EM structure determination.
• Facilitates faster and more reliable structural biology research.