Modeling pH-Dependent Protein Dynamics by Integrating Coarse-Grained Molecular Simulation and the Deep Neural Network...

Area of Science:

• Computational Biology
• Biophysics
• Structural Biology

Background:

• Protein pKa values are crucial for understanding protein behavior, influencing folding, dynamics, and interactions.
• Current pKa prediction methods require detailed all-atom structures, limiting their use with widely adopted coarse-grained models.
• Estimating pKa in coarse-grained models is challenging due to the lack of explicit ionizable groups.

Purpose of the Study:

• To develop a deep-learning-based pKa predictor for coarse-grained protein structures.
• To enable accurate pKa prediction within coarse-grained simulation frameworks.
• To improve the modeling of electrostatic interactions in coarse-grained biomolecular simulations.

Main Methods:

• Introduction of DeepCGpKa, a novel deep-learning algorithm for pKa prediction.
• Application of DeepCGpKa to coarse-grained protein structures.
• Benchmarking against state-of-the-art all-atom pKa prediction methods.
• Integration with coarse-grained molecular dynamics simulations.

Main Results:

• DeepCGpKa achieves accuracy comparable to existing all-atom prediction methods.
• The predictor demonstrates robust performance on partially unfolded protein structures.
• Coupling DeepCGpKa with simulations effectively captures pH-dependent protein conformational changes.

Conclusions:

• DeepCGpKa offers a powerful tool for pKa prediction in coarse-grained protein models.
• This approach enhances the treatment of electrostatic interactions in coarse-grained simulations.
• The integration of data-driven prediction and physics-based simulation provides a practical solution for coarse-grained biomolecular modeling.