A diffusion-based framework for designing molecules in flexible protein pockets.

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Designing molecules for flexible protein pockets is a major challenge in structure-based drug discovery.
• Proteins undergo conformational changes upon ligand binding, complicating molecular design.
• Current deep learning methods often treat protein pockets as rigid, failing to capture dynamic interactions.

Purpose of the Study:

• To present YuelDesign, a novel diffusion-based framework for designing drug molecules targeting flexible protein pockets.
• To enable joint modeling of protein pocket structures and ligand conformations in protein-ligand complexes.
• To address the limitations of existing methods in capturing protein-ligand dynamics.

Main Methods:

• YuelDesign employs a diffusion-based framework utilizing E3former for rotational and translational equivariance.
• Dual diffusion processes are incorporated: an elucidated diffusion model (EDM) for coordinates and a discrete denoising diffusion probabilistic model (D3PM) for ligand atom types.
• The framework iteratively refines both the geometry and chemical identity of generated molecules.

Main Results:

• YuelDesign generates molecules with favorable drug-likeness and low synthetic complexity.
• The generated molecules exhibit diverse chemical functional groups.
• Docking energies of YuelDesign-generated molecules are comparable to those of native ligands.

Conclusions:

• YuelDesign offers a versatile framework for designing drugs targeting flexible protein pockets.
• The joint modeling of pocket and ligand dynamics enhances drug design capabilities.
• This approach holds significant promise for advancing drug discovery applications.