When Alphafold2 predictions go wrong for protein-protein complexes, is there something to be learnt?

Area of Science:

• Structural biology
• Computational biology
• Bioinformatics

Background:

• Protein-protein interactions are crucial for cellular functions.
• Accurate prediction of these complexes is essential for understanding biological processes.
• AlphaFold2 has shown remarkable success in predicting single protein structures.

Purpose of the Study:

• To investigate the reasons behind failed protein-protein complex predictions using AlphaFold2.
• To determine if prediction errors are linked to data quality or inherent model limitations.

Main Methods:

• Analysis of specific cases where AlphaFold2 failed to accurately predict protein-protein complex structures.
• Comparison of AlphaFold2 predictions with existing experimental data and annotations in the Protein Data Bank (PDB).

Main Results:

• Failed AlphaFold2 predictions for protein-protein complexes were found to correlate with erroneous annotations in the PDB.
• In some instances, the predicted binding sites were accurate despite overall complex prediction failure, suggesting issues with data rather than the model's core prediction capability.

Conclusions:

• The study highlights the importance of high-quality data in the PDB for reliable protein complex structure prediction.
• AlphaFold2's performance in complex prediction may be significantly influenced by the accuracy of the input structural data.
• Further refinement of AlphaFold2 or data curation efforts may be needed to improve complex prediction accuracy.