Skip to content

micrisor/NAT-ML

Repository files navigation

NAT-ML: ML prediction of response to neoadjuvant breast cancer therapy

We have designed an interactive dashboard to apply the fully-integrated NAT response model on new (or any desired) data. The prediction is updated in real time when you modify any of the features.

To launch the tool and use it interactively, please click here (and be patient, it may take a little while!):

This repository also contains all the code necessary to reproduce the results in the paper. In particular:

  • validation_online contains the code necessary to apply the integrated model on the validation dataset or any other dataset.
    • Run Predict from csv file.ipynb to apply the model on a dataset stored in a csv file. By default, the notebook applies the models on the validation dataset.
    • Run Predict from single case.ipynb to apply the model on a single case. This is the code-based equivalent to the Binder dashboard linked above.
  • training_code contains the code necessary to train the models from scratch.
    • Run schedule.py to train all the standard models on an HPC cluster.
    • Run schedule_manualimportance.py to train all the leave-one-out models on an HPC cluster.
  • trained_models contains the trained model files so you can apply them directly on the validation data or any other dataset.
    • Note that leave-one-out models (only necessary for feature importance figures) are too large to be stored on GitHub so you will need to re-create them locally as explained above.
  • validation_code contains the code necessary to apply all the trained models on the validation dataset in batch mode. The outputs are generated as text files, so this is less user-friendly than the validation_online option.
    • Run execute.py to apply all the models on the validation dataset.
  • graphs contains Jupyter Notebooks that generate the panels in Figure 4 and Extended Figure 9.
    • If you want to generate the feature importance panels, you will need to generate the leave-one-out models first as explained above.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published