A collection of benchmark tasks and results for the LINCS dataset. See our BioKDD '18 Paper or Poster for more information.
To access the fully processed dataset or graphs (though fully derived from public data), including all cross validation folds, exactly as was used in this work, and/or our hyperparameter search results, contact mmd@mit.edu.
Hyperparameter Optimization Results.ipynb contains a record of all hyperparameter search results and optimal findings--if you want to run it, first you need to download the data file (contact mmd@mit.edu) and place it in the right location, then it should be completely reproducible.
distributions.py contain helper functions to build the distributions used for the random hyperparameter search.
gcn_distributions.py contains code used to build the random samples of the GCNN parameters.
sklearn_classifiers.py contains code used to build the random samples for sklearn classifiers.
The code used to run the GCNNs is available here, but the original source is accessible at https://github.com/mdeff/cnn_graph, via Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering.