GrandFEP

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)

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)

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)

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

3. Full Documentation

huichenggong.github.io/GrandFEP

4. Contact

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