HydraProt: A New Deep Learning Tool for Fast and Accurate Prediction of Water Molecule Positions for Protein...

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Water molecules are crucial for protein structural stability and molecular interactions.
• Precisely predicting water molecule positions around proteins is a significant computational challenge.

Purpose of the Study:

• To introduce HydraProt, a novel deep learning methodology for predicting the precise positions of water molecule oxygen atoms around protein structures.
• To develop a method that surpasses existing state-of-the-art approaches in accuracy and efficiency.

Main Methods:

• Utilized two interconnected deep learning architectures: a 3D U-net for coarse water position sampling and a Multi-Layer Perceptron (MLP) for refining predictions.
• Employed a voxel-based protein representation and Euclidean space embedding for water-protein relationship assessment.
• Incorporated a postprocessing step for further refinement of MLP predictions.

Main Results:

• HydraProt demonstrates superior precision and recall compared to existing methods for predicting water molecule positions.
• The method was validated on extensive datasets of protein structures, confirming its robustness.
• Achieved rapid inference runtime, making it suitable for large-scale studies.

Conclusions:

• HydraProt offers a highly accurate and efficient solution for predicting water molecule hydration shells around proteins.
• The method has significant implications for advancing protein structure studies and accelerating drug discovery pipelines.
• Open-source availability of models, data, and code facilitates reproducibility and further research.