Peptide-protein docking: from physics-based models to generative intelligence.

Area of Science:

• Biochemistry and Structural Biology
• Computational Biology
• Drug Discovery

Background:

• Peptide-protein interactions (PepPIs) are crucial for cellular processes and therapeutic development.
• Experimental determination of peptide-protein complex structures is difficult and expensive.
• Computational prediction of these structures is vital for understanding binding and guiding drug design.

Purpose of the Study:

• To review the shift from conventional to deep learning-based computational methods for peptide-protein complex structure prediction.
• To categorize modern deep learning approaches in peptide docking.
• To identify current challenges and future directions in the field.

Main Methods:

• Review of existing literature on computational peptide-protein interaction prediction.
• Categorization of deep learning methods into three modules: binding region prediction, AlphaFold-based protocols, and deep generative models.
• Analysis of the strengths and limitations of current computational approaches.

Main Results:

• Deep learning-based pipelines are emerging as a powerful alternative to traditional search-and-score methods.
• Modern methods significantly improve accuracy and applicability in peptide-protein docking.
• Key challenges include limited training data and difficulties with flexible, disordered, or modified peptides.

Conclusions:

• Deep learning has substantially advanced peptide-protein docking accuracy and utility.
• Future research should focus on integrating biophysical constraints, improving datasets, and developing large-scale generative models.
• These advancements aim to create robust, design-ready computational tools for peptide docking.