READ: A Retrieval-Alignment Diffusion Framework for Structure-based Drug Design

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Structure-based drug design (SBDD) is crucial for pharmaceutical research, focusing on protein-ligand interactions.
• Current SBDD models often overlook shared binding patterns, limiting their ability to capture molecular recognition principles.
• Limited high-quality experimental data hinders the generalization and applicability of existing SBDD models.

Purpose of the Study:

• To introduce READ, a retrieval-alignment molecular generation framework for SBDD.
• To address the limitations of isolated optimization and one-to-one matching in current SBDD approaches.
• To improve the capture of fundamental principles governing molecular recognition and binding specificity.

Main Methods:

• READ conditions molecular generation on small molecules targeting homologous proteins.
• Retrieved ligands are aligned with a diffusion model across multiple representational spaces.
• Ligands provide conditional guidance throughout the generative process.

Main Results:

• READ demonstrates consistently strong performance against state-of-the-art SBDD methods under a standardized docking-based evaluation protocol.
• The framework introduces a novel retrieval-alignment paradigm for structure-based molecular generation.
• READ offers a practical framework for early-stage computational hit generation.

Conclusions:

• READ effectively addresses challenges in SBDD, including fragmented perspectives and data limitations.
• The retrieval-alignment paradigm represents a significant advancement in structure-based molecular generation.
• This work provides a practical framework for computational hit generation, with experimental validation as future work.