Anncolvar

Anncolvar approximates complex collective variables (CVs) in molecular simulations using feed-forward artificial neural networks to enable monitoring of trajectories, computation of free-energy profiles, and application of biases in enhanced sampling methods such as metadynamics.

Key Features:

• Neural Network-Based Approximation: Employs a feed-forward artificial neural network to approximate CVs, avoiding explicit calculation of complex variables at every simulation step.
• Integration with Enhanced Sampling: Generates PLUMED-compatible inputs to allow monitoring and biasing of ANN-approximated CVs in enhanced sampling methods.
• Demonstrated Efficiency and Accuracy: Shown on systems including cyclooctane derivatives and Trp-cage miniproteins to accurately approximate CVs such as Isomap coordinates and molecular surface area while improving computational efficiency.

Scientific Applications:

• Monitoring Simulation States: Provides reliable CV approximations for tracking progression and state changes in molecular dynamics trajectories.
• Free Energy Calculations: Supplies CV estimates used to compute free-energy landscapes and profiles.
• Acceleration of Rare Events: Supports methods like metadynamics by enabling application of bias potentials or forces to ANN-approximated CVs to accelerate processes such as ligand binding and unbinding.

Methodology:

Train a feed-forward artificial neural network on a dataset of molecular configurations and use the trained model to predict CV values for new configurations encountered during simulations, producing PLUMED-compatible inputs for monitoring and biasing.