Biomolecular Interaction Prediction in the Pre- and Post-AlphaFold Era: The 8th CAPRI Evaluation

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• The Critical Assessment of PRediction of Interactions (CAPRI) is a community-wide experiment to assess protein structure prediction methods.
• Recent years have seen the emergence of AI-driven tools, notably AlphaFold, significantly impacting the field of protein structure prediction.

Purpose of the Study:

• To evaluate the performance of prediction methods, including AI tools and human predictors, during the 8th CAPRI Evaluation period (Rounds 47-55).
• To assess the challenges posed by difficult protein targets and interfaces in structure prediction.

Main Methods:

• Analysis of 11 targets and 21 interfaces from CAPRI Rounds 47-55, with a focus on difficult prediction categories.
• Retrospective analysis comparing the performance of AI prediction tools (e.g., AlphaFold) against human predictors.

Main Results:

• AI tools demonstrated strong performance across evaluated targets.
• Human predictors maintained an advantage over AI for difficult targets, particularly those involving antibody-protein and nucleic acid-protein interactions.
• The evaluation included a diverse set of challenging targets, highlighting the complexity of biomolecular structure prediction.

Conclusions:

• Despite advances in AI, human expertise remains crucial for predicting complex protein structures.
• Continued collaboration, experimental data provision, and advancements in AI, sampling, and scoring methods are essential for future progress in structural biology.