Deep GIST: Deep Learning Models for Predicting the Distribution of Hydration Thermodynamics around Proteins

Area of Science:

• Computational Chemistry and Molecular Modeling
• Biophysics and Structural Biology

Background:

• Hydration thermodynamic quantities are crucial for understanding protein function via free-energy calculations.
• Grid Inhomogeneous Solvation Theory (GIST) computes hydration energy and entropy distributions from molecular dynamics (MD) simulations but is computationally intensive.

Purpose of the Study:

• To develop a computationally efficient deep learning (DL) approach for predicting hydration thermodynamic quantities.
• To provide an accurate and rapid alternative to GIST for analyzing hydration free energy distributions.

Main Methods:

• Developed and trained deep learning models to predict spatial distributions of hydration energy (ΔE_W(r)), hydration entropy (TΔS_W(r)), and hydration free energy (ΔG_W(r)).
• Validated DL model predictions against GIST results for thermodynamic quantities and the free energy change of water replacement (ΔG_W,replace) in protein ligand-binding sites.
• Compared DL model predictions with experimental data for a representative protein-ligand complex.

Main Results:

• DL models predicted ΔG_W(r) distributions with a coefficient of determination (R²) of 0.76-0.84 compared to GIST.
• The DL model accurately predicted the free energy change for water replacement (ΔG_W,replace) with a correlation coefficient of 0.78 against GIST.
• Predicted hydration free energy values correlated with experimental observations of water molecule retention or displacement upon ligand binding.

Conclusions:

• The developed 'Deep GIST' deep learning models offer an efficient and accurate method for predicting hydration thermodynamics.
• This approach enables the inclusion of protein conformational fluctuations, which is challenging with conventional GIST.
• Deep GIST provides a valuable tool for studying protein-ligand interactions and protein function at the molecular level.