PPI-Diff: De Novo Generation of Peptide Binders via Resolution-Aware Geometric Diffusion

Area of Science:

• Computational Biology
• Drug Discovery
• Protein Engineering

Background:

• Peptide binders are crucial for targeting protein-protein interaction (PPI) interfaces, offering advantages over small molecules.
• Current diffusion models like RFdiffusion face limitations in de novo binder design due to rigid-body assumptions and data heterogeneity.
• Existing methods struggle with induced-fit dynamics, experimental resolution variations, and synergistic sequence-structure generation.

Purpose of the Study:

• To develop a novel generative framework, PPI-Diff, for de novo peptide binder design targeting challenging PPIs.
• To address the adaptability bottlenecks of current diffusion models in accurately capturing peptide binding dynamics.
• To enhance the synergy between structure generation and physicochemical properties for improved binder efficacy.

Main Methods:

• Introduced a resolution-aware constraint mechanism to mitigate noise from low-resolution experimental data.
• Implemented an internal-coordinate-driven manifold diffusion model for flexible peptide conformation generation.
• Integrated a geometry-semantic synergistic modeling approach using protein language model embeddings (ESM-2).

Main Results:

• PPI-Diff demonstrates superior performance compared to baseline models on a non-homologous test set.
• Generated binders exhibit enhanced interface contact density and improved stereochemical validity.
• The framework achieves greater sequence novelty in designed peptide binders.

Conclusions:

• PPI-Diff offers a robust and adaptable framework for de novo peptide binder design against undruggable PPI targets.
• The proposed mechanisms effectively address limitations in current diffusion-based protein generation models.
• This advancement holds significant potential for accelerating drug discovery for complex diseases.