Unable to reproduce the experiment results on the PoseBusters set

[myhDaniel]

Hello, thank you for your excellent work! I recently attempted to reproduce your results on the PoseBusters dataset but was unable to achieve the reported performance. I modified the following files to use the PoseBusters dataset (428 cases) as input. Specifically, for the ligand, I directly used the reference structure (ending with _ligand.sdf) provided in the PoseBusters set:

• SurfDock/comp_surface/prepare_target/computeTargetMesh_test_samples.py
• SurfDock/inference_utils/construct_csv_input.py

I also adjusted some file paths in SurfDock/bash_scripts/test_scripts/eval_samples.sh before running the evaluation. However, the results I obtained were quite abnormal. The percentage of top-1 RMSDs below 2 Å was only 39.02%, and the top-1 RMSDs for individual cases fluctuated significantly, ranging from 0.1943 to 76,562,226.4198.

Could you please inform me about any potential mistakes I might have made during this process? Alternatively, could you provide the code and scripts you used for this evaluation? I would greatly appreciate your guidance. Thank you!

[CAODH]

Hello, thank you for your excellent work! I recently attempted to reproduce your results on the PoseBusters dataset but was unable to achieve the reported performance. I modified the following files to use the PoseBusters dataset (428 cases) as input. Specifically, for the ligand, I directly used the reference structure (ending with _ligand.sdf) provided in the PoseBusters set:

• SurfDock/comp_surface/prepare_target/computeTargetMesh_test_samples.py
• SurfDock/inference_utils/construct_csv_input.py

I also adjusted some file paths in SurfDock/bash_scripts/test_scripts/eval_samples.sh before running the evaluation. However, the results I obtained were quite abnormal. The percentage of top-1 RMSDs below 2 Å was only 39.02%, and the top-1 RMSDs for individual cases fluctuated significantly, ranging from 0.1943 to 76,562,226.4198.

Could you please inform me about any potential mistakes I might have made during this process? Alternatively, could you provide the code and scripts you used for this evaluation? I would greatly appreciate your guidance. Thank you!

Sorry for the late reply, could you please provide the details of the test script? This way I can better reproduce your problem and give advice!