Superwater as a generative AI framework to predict water molecule positions on protein structures

Area of Science:

• Structural Biology
• Computational Biology
• Artificial Intelligence

Background:

• Water molecules are crucial for protein structural stability and interactions.
• Accurate prediction of water positions is vital for understanding protein functions and for applications in protein engineering and drug discovery.

Purpose of the Study:

• To introduce SuperWater, a novel generative AI framework for predicting water molecule placements around proteins.
• To achieve state-of-the-art accuracy in predicting water molecule positions.

Main Methods:

• Utilizes a score-based diffusion model integrated with equivariant graph neural networks.
• A generative AI framework designed for precise water molecule localization around protein structures.

Main Results:

• SuperWater achieves state-of-the-art performance in crystal water coverage and prediction precision.
• Predicted water molecule positions are highly accurate, with localization within 0.3 ± 0.06 Å of experimental data.
• Demonstrated capabilities in protein hydration, protein-ligand binding, and protein-protein binding site analysis.

Conclusions:

• SuperWater offers a significant advancement in predicting water molecule positions around proteins.
• The framework has broad applicability in structural biology, binding site prediction, docking, and water-mediated drug design.