Deep-Learning-Assisted Enhanced Sampling for Exploring Molecular Conformational Changes.

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Biophysics

Background:

• Exploring molecular conformational changes is crucial for understanding biological processes.
• Traditional methods often require prior knowledge or extensive computational resources.
• Identifying stable molecular states is key to drug design and function prediction.

Purpose of the Study:

• To develop a novel, knowledge-independent strategy for exploring molecular conformational dynamics.
• To efficiently identify stable states and reconstruct free-energy landscapes of molecular systems.
• To demonstrate the computational efficiency and broad applicability of the proposed method.

Main Methods:

• Utilizing deep learning to extract collective variables (dCVs) from enhanced-sampling simulations.
• Integrating Gaussian-accelerated molecular dynamics (MD) for ergodic sampling with dCV-steered simulations.
• Applying the strategy to toy models and complex biomolecules like chignolin and villin.

Main Results:

• Successfully captured conformational changes of molecular objects without a priori knowledge.
• Achieved blind folding of fast-folding proteins (chignolin, villin) within nanosecond timescales.
• Reconstructed accurate free-energy landscapes for reversible protein folding processes.

Conclusions:

• The novel deep learning-based strategy offers remarkable computational efficiency for conformational analysis.
• This approach significantly advances the ability to study complex molecular dynamics and folding.
• The method holds great promise for applications in chemistry, biology, and pharmaceutical research.