DMFF (Differentiable Molecular Force Field) is a Jax-based python package that provides a full differentiable implementation of molecular force field models.
GNU Lesser General Public License v3.0
151
stars
43
forks
source link
[Feature Request] Workflow to fit relative protein-ligand binding free energy data #90
A workflow specially designed for fine-tuing drug-like force field parameters (typically non-bonded ones) to fit against experimental relative protein-ligand binding free energy data.
Motivation
Accurate prediction of protein-ligand binding free energy is one of the most important tasks of drug-like force field development. Relative protein-ligand binding free energy calculations based on free energy perturbation (FEP) theory are proved to have the ability to reach chemical accuracy. However, the direct fitting against FEP experimental data are rarely reported due to :
difficulty to calculate derivatives of the objective function with respect to force field parameters
complexity of the workflow to perform free energy calculations
Therefore, I would suggest DMFF supports a general workflow (or framework) to fit against experimental FEP data, which will be revolutionary in this realm.
Suggested Solutions
The workflow should take the following data as inputs:
force field parameters (organized in xml, either atom-type based or SMIRKS-based) as a staring point
protein-ligand binding structures or MD-simulated trajectories
calculated data based on the initial force field parameters
experimental data
and then
compute gradients of relative free energy against parameters by taking advantages of the free energy is a state function and the ensemble relationship:
$$\frac{\partial G}{\partial \theta}=-\frac{1}{Z\beta}\frac{\partial Z}{\partial\theta}=-\frac{1}{Z\beta}\int-\beta e^{-\beta U}\frac{\partial U}{\partial \theta}=\left\langle\frac{\partial U}{\partial\theta}\right\rangle$$
Summary
A workflow specially designed for fine-tuing drug-like force field parameters (typically non-bonded ones) to fit against experimental relative protein-ligand binding free energy data.
Motivation
Accurate prediction of protein-ligand binding free energy is one of the most important tasks of drug-like force field development. Relative protein-ligand binding free energy calculations based on free energy perturbation (FEP) theory are proved to have the ability to reach chemical accuracy. However, the direct fitting against FEP experimental data are rarely reported due to :
Therefore, I would suggest DMFF supports a general workflow (or framework) to fit against experimental FEP data, which will be revolutionary in this realm.
Suggested Solutions
The workflow should take the following data as inputs:
and then
$$\frac{\partial\Delta G}{\partial\theta}=\left\langle\frac{\partial U}{\partial\theta}\right\rangle{\lambda=1}-\left\langle\frac{\partial U}{\partial\theta}\right\rangle{\lambda=0}$$
Further Information, Files, and Links
No response