[GaMD-discuss] Inquiry regarding 2D PMF reweighting issues with GaMD simulations

[廖兵 <202...@zc...>]

Dear Prof. Miao and the GaMD Team,
I hope this email finds you well.
I am currently conducting Gaussian accelerated Molecular Dynamics (GaMD) simulations on the CB2 receptor system. I am writing to seek your advice regarding some difficulties I have encountered during the reweighting process using PyReweighting.
1. Simulation Context:
I have performed GaMD simulations and have collected data from 10 log files (gamd_GaMD_1.log through gamd_GaMD_10.log) along with their corresponding trajectories.
2. The Workflow:
I followed the standard tutorial to generate the weights and calculate the Free Energy Landscape (FEL).
Step A: Weight Generation
I concatenated the log files and extracted the weights using the following command (ensuring temperature $T=300K$):
Bash

cat gamd_GaMD_{1..10}.log | grep -v "#" | awk '{

    kBT = 0.001987 * 300;

    dV = $7 + $8;

    print dV/kBT " " $2 " " dV
}' > weights.dat



I confirmed that weights.dat has the correct line count matching my combined trajectory frames.
Step B: Reaction Coordinates
I calculated two reaction coordinates (RCs) and saved them into input_2d.dat:
RC1 (X): Protein Backbone RMSD (Residues 0-298).
RC2 (Y): Distance between TM3 and TM6 (Residues 110 and 219).
Step C: PyReweighting Execution
I used PyReweighting-2D.py with the Cumulant Expansion method (amdweight_CE) as follows:
Bash

python PyReweighting-2D.py \

  -cutoff 10 \

  -input input_2d.dat \

  -weight weights.dat \

  -Xdim 0.0 6.0 -discX 0.3 \

  -Ydim 10.0 18.0 -discY 0.3 \

  -Emax 20 \

  -job amdweight_CE \

  > log_reweight_final.txt 2>&1



3. The Issue:
After reweighting, I plotted the FEL using the generated pmf-2D-Final.xvg file (using Python/Matplotlib with tricontourf). However, the resulting landscape appears discontinuous and lacks a clear, defined energy basin.
I am trying to compare my results with a reference study (Nature, Fig 7k, https://www.nature.com/articles/s12276-024-01296-x#Fig7), but my FEL shows significant discrepancies and lacks the expected minimum energy wells.
4. Attachments & Code:
I have attached the plotting script I used and a screenshot of the "strange" FEL I obtained.
Plotting Script: (See below)
Resulting Figure: Final_Landscape_Styled.png (Attached)
Log Files: log_reweight.txt (Attached)
My Plotting Code Snippet:
Python

# (I am using tricontourf to visualize the xvg file)
# ... [Variables setup] ...
cf = ax.tricontourf(x, y, z_relative, levels=levels_range, cmap='jet', extend='max')
# ...



Could you please provide some insights on the following:
Is my procedure for concatenating weights from multiple log files correct for use with PyReweighting?
Does the discontinuity in the FEL suggest an issue with the sampling convergence, or is it likely a parameter issue (e.g., bin size discX/Y=0.3 or cutoff)?
Are there specific checks I should perform to verify if the reweighting factors are correctly mapping to the trajectory frames?
Any guidance or suggestions would be greatly appreciated. Thank you very much for your time and for developing this excellent tool.
Best regards,
LiaoBing
Zhejiang Chinese Medical University







 廖兵
202...@zc...