Predicting structures of large protein assemblies using combinatorial assembly algorithm and AlphaFold2.

Area of Science:

• Structural biology
• Computational biology
• Biochemistry

Background:

• Accurate protein structure prediction is crucial for understanding biological function.
• Current deep learning models like AlphaFold2 excel at monomeric protein prediction but struggle with large, multi-subunit complexes.
• Predicting the structure of large protein complexes remains a significant challenge due to their scale and intricate subunit interactions.

Approach:

• CombFold is a novel combinatorial and hierarchical assembly algorithm designed for predicting the structures of large protein complexes.
• It leverages pairwise subunit interactions predicted by AlphaFold2 to assemble complex structures.
• The algorithm incorporates distance restraints from crosslinking mass spectrometry and efficiently enumerates potential complex stoichiometries.

Key Points:

• CombFold achieved high accuracy (TM-score > 0.7) for 72% of large, asymmetric protein complexes in the Top-10 predictions across two independent datasets.
• Predicted complex structures exhibited 20% greater structural coverage compared to entries in the Protein Data Bank (PDB).
• The method successfully generated high-confidence predictions for complexes with known stoichiometry but unknown structures from the Complex Portal.

Conclusions:

• CombFold offers a significant advancement in predicting the structures of large protein complexes.
• Its accuracy and ability to integrate experimental data enhance its utility for structural biology research.
• This tool holds promise for expanding the structural coverage of protein assemblies beyond individual proteins.