This project aims to develop a tool to dynamically design potential
drug molecules. The Flexible Topology method uses PyTorch to build a
ML model, which can be trainable or non-trainable. It will then
predict the structure and pose of a set of given ghost atoms to be a
potential ligand candidate for a protein. The output of the model is a
function whose gradient, with respect to positions, produces external
forces. These force will constally change the chemical type and
positions of ghost atoms and optimize them toward target drug-like
molecules.
We run molecular dynamics simulations using Openmm where the OpenMM Plugin MLForce is employed to apply the ML-based forces. For more details read the Flexible Topology: A Dynamic Model of a Continuous Chemical Space paper in JCTC.
To install this package do the folloeing commands
- git clone https://github.com/ADicksonLab/flexibletopology.git
- cd flexibletopology
- pip install -e .