A Shiny app that generates ROC and PR curves based on Drabme's ensemble-wise synergy output results.
The next figure shows how Drabme's synergy scores are translated to confusion matrix metrics that are used to construct the ROC and PR curves:
The following libraries must be installed before using the app:
You can test the app without cloning the repo: shiny::runGitHub("druglogics/druglogics-roc").
Otherwise, clone the repo and run: shiny::runApp()
The app generates by default a ROC (Receiver Operator Characteristic) and a PR (Precision-Recall) curve using the example test files. The user can also upload his own input files to generate a new ROC&PR curve. The input files must be:
- An observed synergies file to test against the synergies found from
Drabme's simulations. These are the true positive synergies. See example file. - An ensemble-wise synergy scores result file (most common use case is to run the Gitsbe and then Drabme using the
druglogics-synergymodule to get this output file). See respective example file.
In the case that two ensemble-wise synergy score files are used (usually the second represents the results of a Gitsbe analysis where the models are trained to a proliferation profile) we can merge the synergy scores to a combined-classifier output and produce thus new combined ROC and PR curves.
The app's interface allows the user to choose which output result will be used to generate the curves.
The options for the single output results are predictor1 and predictor2 and for the combined classifier output predictor1-predictor2 and predictor1+predictor2.