Skip to content

Latest commit

 

History

History
31 lines (22 loc) · 2.47 KB

USING-NEW-DOCKING-METHODS.md

File metadata and controls

31 lines (22 loc) · 2.47 KB
Raw

Setting up Open-ComBind to work with a new pose generator

Currently Open-ComBind is configured to work with GNINA and its outputs. However, Open-ComBind is relatively easy to reconfigure for a new docking tool. Using a new pose generator will require the user to recompute the "native" and "reference" statistics for the ligand poses generated by the new tool.

Note:

Open-ComBind is not currently set up to handle ligand poses that have a different protonation state or tautomeric state from the ground truth ligand. All RMSD calculations (including the RMSD of the MCSS) will result in an error if there is a difference between ligands. If you would like to use a docked pose generator that can alter the protonation of the ligand, some code reconfiguration will likely be required.

Calculating new statistics

In order to compute new statistics, the benchmark dataset will be used. The steps for recomputing the statistics are enumerated below.

  1. Generate dockings with the new pose generator

    • it is best to generate these poses as .sdf files (optionally, can gunzip these files as well and save as .sdf.gz)
    • Docking score should be saved to each pose in the sdf file as a property as an example, the GNINA scores (CNNscore) are saved as follows: 
      >  <CNNscore>  (1)
0.9719959497

  2. Run the featurization of the ligands, ensuring the docked files are sorted correctly by the new score.

    • featurization will now require setting the option --newscore with the name of your score in the sdf files (e.g. CNNscore for GNINA)
    • It is assumed that larger scores are better and the pose files will be resorted accordingly. If not, please use --no-reverse with the CLI or set reverse=False if using the Python API.

  3. Using the script open_combind/stats_data/features_to_stats.py calculate the statistics for the new pose generator.

  4. To calculate the new alpha for your scoring function, generate a plot with open_combind/stats_data/docked_to_alpha.py and use the (usually rounded) slope of the best fit line as alpha for pose prediction.

Open-ComBind docking ligands with a new pose generator

Likely you will want to use the entire Open-ComBind pipeline, so you will want to reconfigure Open-ComBind to use your new pose generator by default. This will require editing open_combind/features/dock.py to use your new pose generator rather than GNINA.

Pull requests to make Open-ComBind pose generator agnostic are welcome.