Evaluating AlphaFold Tools and Related Scoring Functions for Protein-peptide Complex Prediction

Area of Science:

• Computational Biology
• Structural Biology
• Drug Discovery

Background:

• Accurate three-dimensional structures of protein-peptide complexes are vital for understanding biological mechanisms and developing peptide therapeutics.
• Protein-peptide docking computational methods are essential for predicting these complex structures.

Purpose of the Study:

• To evaluate the performance of AlphaFold-Multimer, ColabFold, and AlphaFold3 in predicting protein-peptide complex structures.
• To compare template-based (TB) and template-free (TF) prediction methods across these tools.
• To assess the effectiveness of various scoring functions in ranking predicted complex structures.

Main Methods:

• Comparative analysis of AlphaFold-Multimer, ColabFold, and AlphaFold3 using both TB and TF approaches.
• Evaluation of prediction accuracy for top-ranked models and the overall prediction pool.
• Assessment of scoring functions, including the integrated AlphaFold scorer, FoldX-Stability, and HADDOCK-mdscore.

Main Results:

• AlphaFold-Multimer demonstrates strong performance in TB predictions and moderate capability in TF scenarios.
• ColabFold shows adaptability in both TB and TF prediction settings.
• AlphaFold3 produces high-quality structures but with lower medium accuracy compared to AlphaFold-Multimer's large model pool.
• The integrated AlphaFold scoring function performed best, with FoldX-Stability and HADDOCK-mdscore offering complementary ranking insights.

Conclusions:

• The study highlights the strengths and weaknesses of AlphaFold-Multimer, ColabFold, and AlphaFold3 for protein-peptide structure prediction.
• Combining multiple scoring functions or using consensus approaches may enhance the accuracy of AlphaFold-based predictions.