GrandFEP

GrandFEP is a Python library for relative binding free energy (RBFE) calculations that explicitly enhances water sampling. It combines Grand Canonical Monte Carlo (GCMC), Water-Swap Monte Carlo (Water MC), Replica Exchange Solute Tempering (REST2), and Terminal-Flip Monte Carlo (TFMC) on top of OpenMM, enabling water molecules to be inserted and deleted during the simulation so that water occupancy differences between ligand pairs are captured correctly.

Key features:

• Alchemical water swap moves (WaterMC) for moving water between the active site and bulk using nonequilibrium candidate Monte Carlo (NCMC)
• Alchemical water insertion/deletion (GC ensemble)
• REST2 enhanced sampling for ligand and protein degrees of freedom
• Terminal-Flip MC for enhanced sampling of terminal groups dihedrals

<img src="docs/picture/interpolated_AB.gif" alt="AB\_water" style="width: 50%;" />

1. Quick Installation

1.1 Prepare Env

mamba env create -f env.yml # edit cuda and MPI according to your cluster
mamba activate grandfep_env
pip install .

1.2 Later on the cluster

source /home/NAME/SOFTWARE/miniforge3/bin/activate grandfep_env
module add openmpi4/gcc/4.1.5 # only as an example
which mpirun                  # check if the correct mpirun is used

2. GrandFEP Sampling Performance

1) Overall performance (weighted RMSE, 95% CI)

<p align="center"> <img src="docs/picture/Water_Set_FEP+_OpenFE.png" width="750" alt="Weighted RMSE with 95% CI on the water set for GrandFEP (GCMC/WaterMC) vs FEP+ and OpenFE." /> </p>

What this shows: aggregated error across the full water set (lower is better).

• GrandFEP (GCMC): 0.94 kcal/mol
• GrandFEP (WaterMC): 1.00 kcal/mol
• FEP+: 0.86 kcal/mol
• OpenFE: 1.60 kcal/mol

---

2) Per-target predictions (8 systems)

<p align="center"> <img src="docs/picture/All_8_system_deltaG_GCMC+WaterMC.png" width="900" alt="Scatter plots of predicted vs experimental ΔG across 8 targets, comparing GCMC and WaterMC." /> </p>

How to read: each panel is one target; diagonal is perfect agreement; shaded band indicates 1 kcal/mol error region.

---

3) Accuracy and correlation by target (RMSE and R²)

<p align="center"> <img src="docs/picture/RMSE_R2_1x16x15ns.png" width="900" alt="Bar charts of RMSE and R² by target for GCMC, WaterMC, FEP+, and OpenFE." /> </p>

What this shows: target-by-target breakdown of error (RMSE) and correlation (R²), including bootstrapped 95% CI.

3. Full Documentation

https://degrootlab.github.io/GrandFEP/

4. Contact

Chenggong Hui
chenggong.hui@mpinat.mpg.de
huicgx@126.com