Add a way to create a DeseqDataSet object from normalized count matrix / preprocessed adata object

[VictorCurean]

Currently, the only way to initialize a DeseqDataSet object is to have a non-negative integers in adata.X. The AnnData object might have been normalized/preprocessed before-hand, so this limits the integration with other tools.

Describe the solution you'd like

1. Solution:
   Instead of throwing an Exception, a Warning might be better in the test_valid_counts function.

2. Solution

Add a constructor flag that ignores the test.

I am not sure which solution might work better, I can't think of a case where a Warning wouldn't do the job, but I can take care of this one if there is one preferred fix for it.

[vcabeli]

Hello @VictorCurean !

Just to make sure I understand the solution you are suggesting, you would like PyDESeq2 to throw a warning and perform a conversion of adata.X to integer ? (integers are necessary for the DESeq2 method)

The current way to use PyDESeq2+Anndata is to make sure adata.X contains the raw counts and adata.obs the metadata. If your adata object is normalized/preprocessed before pydeseq2, make sure you keep the raw counts somewhere (e.g. adata.layers['raw_counts'], or copy the raw object if you subset samples / genes during processing adata_raw = adata.copy()) and then set adata.X = adata.layers['raw_counts'] before calling PyDESeq2.