DMFF

DMFF implements differentiable molecular force fields to enable parameter optimization and differentiable evaluation of energies, forces, ensemble averages, free energies, and virial tensors using automatic differentiation within a Jax-based Python package.

Key Features:

• Differentiable Implementation: Provides a fully differentiable implementation of molecular force field models for computing derivatives with respect to model parameters.
• Automatic Differentiation: Utilizes automatic differentiation to compute parameter derivatives required for optimization and analysis.
• Top-Down and Bottom-Up Development: Supports both top-down and bottom-up approaches to force field development and refinement.
• Thermodynamic Quantities Evaluation: Enables differentiable evaluation of energies, forces, ensemble averages, and free energies.
• Forces and Virial Tensor Computation: Computes forces and virial tensors to support validation of advanced force field models in molecular dynamics contexts.
• Automation of Parameterization: Automates aspects of force field parameterization through differentiable computations to streamline model development.

Scientific Applications:

• Computational chemistry: Improves force field parameterization and predictive accuracy for molecular simulations.
• Molecular biology: Enables more accurate simulation-based studies of biomolecular systems through refined force fields.
• Drug discovery: Enhances the reliability of molecular simulations used for ligand binding and lead optimization.
• Materials science: Supports parameter development for force fields applied to materials modeling and property prediction.
• Biophysical studies: Facilitates investigation of thermodynamic and structural properties via differentiable evaluation of ensemble quantities.

Methodology:

Implemented as a Jax-based Python package that uses automatic differentiation to compute derivatives with respect to force field parameters and to enable differentiable evaluation of energies, forces, ensemble averages, free energies, and computation of forces and virial tensors while supporting both top-down and bottom-up development approaches.