Understanding `differential_expression()` and lvm-DE

Help scvi-tools
1

I are interested in looking code behind the differential_expression() function, in particular the lvm-DE method. From an issue in scVI-tools Github, it seems as though the lvm-DE method is now a part of the general differential_expression() function in ‘scVI-tools’ (when using weights=“importance”).

Exploring scVI-tools, it seems like the differential_expression() function code is tied to _de_core.py and _differential.py (same folder as _de_core.py; can’t add a third link). However, I can’t find the “weights” flag in either script (or a conditional looking for the word “importance”), which makes me think I’m looking in the wrong spot.

Any help to direct me would be most appreciated.

2

Hi, lvmDE is active when using mode=‘change’ (the default), while the old method requires mode=‘vanilla’. Weights=‘importance’ changes the manner samples are taken in get_normalized_expression.

1 Like
3

Thanks for your reply. I had a hard time understanding what combination of settings constituted the lvm-DE method. The guide really didn’t go over that.

de_results = model.differential_expression( groupby="subtype", idx1=group1, idx2=group2, weights="importance", batch_correction=False, delta=None, fdr_target=0.01 )

Can you help me understand where “differential_expression()” and “get_normalized_expression()” interact? Adding “weights=importance” to “differential_expression” clearly makes a big impact on the results, but I can’t seem to figure out how adding the “weights” flag to the former function calls the latter.

4

scvi-tools/scvi/model/base/_rnamixin.py at d094c9b3c14e8cb3ac3a309b9cf0160aff237393 · scverse/scvi-tools · GitHub defines a partial function of get_normalized_expression.

1 Like