Accurate and Efficient Multilevel Free Energy Simulations with Neural Network-Assisted Enhanced Sampling

Area of Science:

• Computational Chemistry
• Biophysics
• Artificial Intelligence

Background:

• Free energy differences (ΔF) are crucial for understanding chemical and biological processes.
• Directly calculating ΔF with high-level quantum mechanics is computationally expensive and incompatible with standard alchemical methods.
• Multilevel free energy simulations offer a solution by combining low-cost and high-cost methods, but face challenges due to poor configurational overlap.

Purpose of the Study:

• To develop an accurate and efficient method for estimating free energy differences (ΔF) using multilevel simulations.
• To overcome the challenge of poor configurational overlap between different levels of theory in free energy calculations.
• To leverage deep neural networks and enhanced sampling for improved accuracy and scalability in free energy simulations.

Main Methods:

• Utilized a deep neural network, specifically an adversarial autoencoder, to identify a low-dimensional latent space representing distinct distributions at different theory levels.
• Employed enhanced sampling techniques within this latent space to efficiently sample configurations crucial for free energy correction.
• Applied the data-driven approach to both gas and condensed phase systems.

Main Results:

• The developed deep learning model effectively addresses the configurational overlap challenge in multilevel free energy simulations.
• The method demonstrates high accuracy and efficiency in calculating free energy differences.
• The approach shows significant potential for scalability to more complex chemical and biological systems.

Conclusions:

• The integration of deep neural networks and enhanced sampling provides a powerful solution for accurate and efficient free energy calculations.
• This data-driven methodology overcomes key limitations of traditional multilevel free energy simulations.
• The approach holds promise for advancing computational studies in chemistry and biology.