Extended DeepILST for Various Thermodynamic States and Applications in Coarse-Graining

Area of Science:

• Computational chemistry
• Statistical mechanics
• Machine learning

Background:

• Molecular dynamics (MD) simulations require accurate interatomic potentials, which are challenging to predict.
• Existing methods for potential prediction lack transferability and are limited to specific thermodynamic states.

Purpose of the Study:

• To expand the training range of thermodynamic states for DeepILST.
• To assess DeepILST's performance in coarse-graining multiatom molecules.
• To identify molecular features influencing coarse-graining accuracy.

Main Methods:

• Deep inverse liquid-state theory (DeepILST) framework was trained on an expanded range of thermodynamic states.
• Performance evaluation of DeepILST for coarse-graining various multiatom molecules.
• Analysis of molecular characteristics impacting DeepILST's coarse-graining capabilities.

Main Results:

• Expanded thermodynamic state range improved DeepILST training.
• DeepILST demonstrated effective coarse-graining for diverse multiatom molecules.
• Key molecular characteristics influencing coarse-graining performance were identified.

Conclusions:

• DeepILST offers a robust, data-driven approach for predicting interatomic and coarse-grained potentials.
• The expanded framework enhances transferability across various molecular systems and thermodynamic conditions.
• This work advances the predictive power of simulations for complex molecular systems.