Right now, LPO just splits randomly. How about we enforce that in each training split, every cell line and every drug occur so that we really optimize for missing value imputation. Because currently, it could happen by chance that we have an unseen drug/cell line in the validation or test dataset.
Then we could also enforce for cross-study prediction that we only predict seen cell lines/drugs.
Right now, LPO just splits randomly. How about we enforce that in each training split, every cell line and every drug occur so that we really optimize for missing value imputation. Because currently, it could happen by chance that we have an unseen drug/cell line in the validation or test dataset.
Then we could also enforce for cross-study prediction that we only predict seen cell lines/drugs.