Protein complex structure modeling by cross-modal alignment between cryo-EM maps and protein sequences.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Cryo-electron microscopy (cryo-EM) is a key technique for determining protein complex structures.
• Existing automated modeling methods often fail due to errors in chain separation without sequence guidance.
• Noise and cross-chain interactions in cryo-EM maps can lead to inaccurate protein structure models.

Purpose of the Study:

• To develop a fully automated method for protein complex structure modeling from cryo-EM data.
• To improve the accuracy and reliability of cryo-EM based protein structure determination.
• To provide a sequence-guided approach for robust cryo-EM model building.

Main Methods:

• EModelX utilizes multi-task deep learning to predict Cα atoms, backbone atoms, and amino acid types from cryo-EM maps.
• It employs cross-modal alignments between cryo-EM density maps and protein sequences for modeling.
• Sequence-guided Cα threading is used to fill gaps and refine the final structural models.

Main Results:

• EModelX achieved an average RMSD of 1.17 Å in recovering PDB-deposited structures, demonstrating near atomic-level precision.
• The method's final models reached an average TM-score of 0.808, outperforming state-of-the-art techniques.
• Combining EModelX with AlphaFold further improved the average TM-score to 0.911.

Conclusions:

• EModelX offers a robust, automated solution for cryo-EM protein complex structure modeling.
• The sequence-guided approach significantly enhances model accuracy and reliability.
• EModelX has the potential to refine existing Protein Data Bank (PDB) structures and advance structural biology research.