GP-SAMPL7

Stacking Gaussian Processes to Improve pKa Predictions in the SAMPL7 Challenge

Authors

• Robert M. Raddi
  • Department of Chemistry, Temple University
• Vincent A. Voelz
  • Department of Chemistry, Temple University

Prediction of relative free energies and macroscopic pKas for SAMPL6 and SAMPL7 small molecules using a standard Gaussian process regression as well as a deep Gaussian process regression.

Below is a description of what this repository contains:

• GPR/: the code to perform standard and deep GPR, processing, analysis, etc.
• scripts_and_notebooks/: scripts that call on GPR/ e.g., script to get features, analysis notebooks
  • scripts_and_notebooks/compile_results.ipynb: master notebook for creating figures and tables
  • scripts_and_notebooks/database_info.ipynb: notebook for analyzing the database
  • scripts_and_notebooks/features.py: script for computing descriptors
  • scripts_and_notebooks/standardGPR.py: script for running sklearn.GaussianProcessRegressor
  • scripts_and_notebooks/runme_deepGP.py: script for running deep Gaussian process regression using deepGPy
• Structures/: input smiles strings, input microtransitions
• Submissions/: SAMPL7 submission (only for the standard GP model)
• predictions/: directories separating results & prediction files for SAMPL6 and SAMPL7. Each consisting of relative free energies and macroscopic pKa values for each small molecule.
  • predictions/SAMPL6_deepGP/: free energies and macro-pKa predictions
  • predictions/SAMPL6_stdGP/: free energies and macro-pKa predictions
  • predictions/SAMPL7_deepGP/: free energies and macro-pKa predictions
  • predictions/SAMPL7_stdGP/: free energies and macro-pKa predictions
  • predictions/RandomForestSAMPL6/: free energies and macro-pKa predictions
  • predictions/RandomForestSAMPL6_without_filter/: free energies and macro-pKa predictions
  • predictions/RandomForestSAMPL7/: free energies and macro-pKa predictions
  • predictions/RandomForestSAMPL7_without_filter/: free energies and macro-pKa predictions
• pKaDatabase/: curated database from various sources.
  • pKaDatabase/pKaDatabase.pkl: pickle file of database (loads with pandas) NOTE: feature calculations already performed and stored inside pd.read_pickle('pKaDatabase.pkl')
  • pKaDatabase/Sulfonamides.pkl: pickle file of only sulfonamides database (loads with pandas)
• tables/: LaTeX tables for free energies and macroscopic pKas
• figures/: macroscopic pKa comparison between std GP and deep GP