Once CrossNorm (#122) and a single cell method (#124) are tested, how well do they do with our prediction task?
We should test first in the setting with the easiest prediction task (BRCA subtype). If they perform poorly, we can be more confident it is the normalization method not performing well, rather than the difficulty of prediction task.
and I'll need to read some other places carefully to ensure compatibility
One idea is to add options to the NormalizationWrapper() functions such that we can optionally include either method, like what was done for add.untransformed = TRUE and add.qn.z = TRUE.
A big time saver will be not testing in the PLIER context. If they work well at ML tasks, we can later test with PLIER. The update required in the PLIER script will involve adding an additional item to a vector (trivial).
Once we see that these methods are useful/comparable to our existing suite of methods, we can think about how to include them in the main and supplementary figure plots...
Once CrossNorm (#122) and a single cell method (#124) are tested, how well do they do with our prediction task?
We should test first in the setting with the easiest prediction task (BRCA subtype). If they perform poorly, we can be more confident it is the normalization method not performing well, rather than the difficulty of prediction task.
This will involve changes to:
One idea is to add options to the
NormalizationWrapper()functions such that we can optionally include either method, like what was done foradd.untransformed = TRUEandadd.qn.z = TRUE.A big time saver will be not testing in the PLIER context. If they work well at ML tasks, we can later test with PLIER. The update required in the PLIER script will involve adding an additional item to a vector (trivial).
Once we see that these methods are useful/comparable to our existing suite of methods, we can think about how to include them in the main and supplementary figure plots...