gennadyFauna

I believe the standard way to do this is to pass in `tf_names='all'` or `None`.

@BorisMuzellec, thank you for looking into it. On further investigation: - The procedure fails even on the much smaller dataset `ad=ad[ad.obs.dataset.isin(['litvinukova'])]`, i.e., restricted to only the seven donors in the...

@BorisMuzellec, this is correct. MoM estimates, size factors, mu_hat. are identical, but genewise dispersions are different. I have narrowed this down to **two** possible causes for discrepancies. **1. The dispersion...

I did a deeper dive into the code to try to isolate the issue. At least part of the inconsistency is that the AnnData.X matrix I initially pass into DESeq2...

@nickodell sure, here you go: https://github.com/Fauna-Bio/GG_2025/blob/pydeseq2_bug/pydeseq2/5_differential_expression_L-BFGS-B.ipynb The h5ad files (same directory) are obtained by applying `ad_f=ad_f[ad_f.obs.dataset.isin(['litvinukova'])].copy()` after file import in the other notebook and running the respective pipelines (and stripping...

Powell might be more reliable (i.e., match the grid search result and the visual optimum) for cases with flat likelihoods that nevertheless have a real minimum: This seems to hold...

I am still working out some tests because I only recently realized this may be a SciPy issue, but could this be related to https://github.com/owkin/PyDESeq2/issues/391#issuecomment-3086465534? Given (nearly) identical input functions...

Thank you for taking a look and clarifying the nomenclature! I have been trying to rule out a few explanations, and the issue I saw may be combination of obscure...