Improved prediction of protein-protein interactions using AlphaFold2

Area of Science:

• Structural biology
• Computational biology
• Bioinformatics

Background:

• Protein complex structure prediction is crucial for understanding biological function.
• Current computational methods lack accuracy for protein complex modeling.
• AlphaFold2 excels at predicting single protein chain structures.

Purpose of the Study:

• To evaluate AlphaFold2's performance in predicting heterodimeric protein complexes.
• To develop a method for assessing the quality and reliability of predicted protein complex structures.
• To accurately differentiate between interacting and non-interacting protein pairs.

Main Methods:

• Application of the AlphaFold2 protocol to heterodimeric protein complexes.
• Optimization of multiple sequence alignments for improved predictions.
• Development of a predictive function for DockQ scores based on predicted interfaces.

Main Results:

• AlphaFold2 generated models of acceptable quality (DockQ ≥ 0.23) for 63% of heterodimeric complexes.
• The developed scoring function accurately distinguishes acceptable from incorrect models.
• The method identified 51% of interacting protein pairs with a 1% false positive rate.

Conclusions:

• AlphaFold2, with optimized alignments, shows promise for heterodimeric complex structure prediction.
• The predictive DockQ scoring function offers a reliable way to assess model quality and identify interacting proteins.
• This approach advances the computational prediction of protein-protein interactions.