adamklie / drpredict Goto Github PK

• Utilizes MultiomicsDataset
• Splits into train, val, test
• Consider making sure drugs are separated into train, val and test set (don't want it to memorize drugs
• Others?

• Load from file
• Preprocess as a whole, not per sample
• Make flexible for any omics type oriented a certain way
• Put into dictionary depending on what omics types are available (include binary labels)
• Allow for choice of label

FOR FUTURE!!!
Opportunity to utilize AnnData loader perhaps

• Collapse into 4 or 2 groups
• Perform statistical tests to quantify stratification

• Test AE build
• Make sure training and val (no test needed)
• Get working when passing in a single `omics (i.e. make sure compatible with MultiomicsDataModule)
• Get metric tracking working
• Test model training on single omics types
• Hyperparam optimize on encoding of mutations
• Hyperparam optimize on encoding of expression
• Hyperparam optimize on encoding of cnv
• Hyperparam optimize on encoding drug response

https://pytorch-lightning.readthedocs.io/en/latest/api/pytorch_lightning.loops.dataloader.PredictionLoop.html?highlight=write_prediction#pytorch_lightning.loops.dataloader.PredictionLoop