Tool for preprocessing and creating datasets for drug sensitivity, and creating transferable representations across tasks. Contains utils, such as different strategies for featurization of drugs, cell-lines and proteins, and utilities to preprocess the most common public datasets for each tasks, and performing different cross-validation, cold-start settings.
Example of prediction tasks include:
- Anticancer sensitivity prediction: where cell-lines are represented by different sources of omics data, and drugs are featurized using different strategies.
- Protein-protein interaction prediction: where proteins are represented by their primary sequence, and are later featurized using embeddings obtained from pretrained language models.
- Drug-Target interaction prediction: Proteins are represented by their primary sequences, and drugs are represented by their molecular structure.
example.ipynb contains an ilustrative example of its usage.
ExDTI.ipynb contains an example for the prediction of binding affinities.
ExMultiTaskTransferDataset contains an example where a drug representation is created with the objective of transfer learning.
ExTabular.ipynb contains an example where the drugs and cell-lines are converted to tabular features and a baseline performance is obtained.