Modeling CAPRI Targets of Round 55 by Combining AlphaFold and Docking.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Advancements in deep learning, particularly AlphaFold, have revolutionized protein structure prediction.
• Accurate modeling of protein tertiary and quaternary structures is crucial for understanding biological function.
• The Critical Assessment of PRedicted Interactions (CAPRI) benchmark assesses protein-protein docking methods.

Purpose of the Study:

• To evaluate a hybrid approach combining AlphaFold2 multimer predictions with traditional docking techniques for protein-protein docking.
• To assess the performance of this hybrid method in modeling oligomeric protein targets in the CAPRI Round 55.
• To explore strategies for enhancing protein-protein docking predictions using deep learning outputs.

Main Methods:

• Utilized the AlphaFold2 multimer pipeline for initial protein structure predictions.
• Developed a hybrid docking approach integrating AlphaFold2 predictions with traditional docking methods.
• Generated docking predictions by combining models from lower-oligomeric states (dimers, trimers) for higher-oligomeric targets (trimers, tetramers).
• Employed a template-based docking procedure on AlphaFold-predicted monomer structures.

Main Results:

• The hybrid approach demonstrated effectiveness in modeling oligomeric protein targets.
• Analysis included clustering of AlphaFold models and confidence assessment of residue-residue contacts.
• Correlation between AlphaFold prediction stability and the quality of submitted models was investigated.

Conclusions:

• The integration of deep learning predictions with docking techniques offers a powerful strategy for protein-protein docking.
• This hybrid approach enhances the accuracy and reliability of modeling complex protein assemblies.
• The findings contribute to the advancement of computational structural biology and drug discovery efforts.