Curating a Synergy Model

Curating a synergy model is similar to curating a Drug Response Prediction model with a few key differences:

Interface changes

  • Data loading doesn’t require the creation of data loading objects, instead functions can be called to load the appropriate dataset

  • Response dataframes can be loaded in two ways: 1) by loading the individual train/val/test response dataframes, or 2) by loading a single response

dataframe with a column indicating which split the row belongs to - Split files can be a single file name, or a list of files. - Commonly use preprocessing selections and transformation can be indicated using a parameter and will be automatically performed with the data loading functions

Application-specific differences

  • Uses the synergy benchmark dataset (details can be found here).

  • Two drug IDs and one cell ID

  • y_col_name should be one of loewe, bliss, zip, hsa, smean, or css.

Core improvelib imports import improvelib.utils as frm

Synergy imports from improvelib.applications.synergy.config import SynergyPreprocessConfig import improvelib.applications.synergy.synergy_utils as syn

Create a config file <model>_params.ini