Machine learning approach for accurate backmapping of coarse-grained models to all-atom models

Area of Science:

• Computational chemistry and molecular modeling.
• Application of artificial intelligence in structural biology.

Background:

• Coarse-grained (CG) models simplify molecular simulations but lose atomic detail.
• Backmapping reconstructs detailed all-atom (AA) models from CG representations.
• Existing backmapping methods have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for backmapping CG models to AA models.
• To compare the performance of different ML regression techniques for this task.
• To assess the potential of ML for improving backmapping accuracy.

Main Methods:

• Development of four ML regression models: artificial neural network (ANN), k-nearest neighbors (KNN), Gaussian process regression (GPR), and random forest (RF).
• Training and validation of ML models using established molecular structures.
• Quantitative comparison of ML model predictions against existing backmapping approaches.

Main Results:

• The developed ML models demonstrated superior predictive performance compared to conventional backmapping methods for selected structures.
• Artificial neural network, k-nearest neighbors, Gaussian process regression, and random forest models showed varying degrees of accuracy, with overall improved results.
• The ML approach offers a promising alternative for accurate and efficient molecular model reconstruction.

Conclusions:

• Machine learning regression models provide a powerful tool for accurate backmapping of coarse-grained to all-atom molecular models.
• These ML models represent a significant advancement over current backmapping techniques.
• The findings suggest broad applicability of ML in computational structural biology and molecular dynamics.