Differential gene expression shows very low raw_means

Help scvi-tools
1

Dear all,
I used the differential gene expression function, however I only get a very little amount of DEGs and they have a very low raw_mean1 and 2. If I use MAST for example I get much more DEGs

de_results_HFD_L1 = model_scvi.differential_expression(
subset,
groupby=“Sample”,
group1=“A”,
group2=“B”,
mode=“change”,
filter_outlier_cells=True,
batch_correction=False,
delta=0.15,
fdr_target=0.05,
all_stats=True,

    )
2

Hi, most benchmarks show that MAST inflates the number of detected genes. However, we changed our default DE function in our current main branch to filter more of these lowly expressed genes. We will release it with scVI-tools 1.3. Pseudobulk methods (especially DESeq2) are current top-performers in DE benchmarks and I would consider using those.
ScVI-tools DE serves its purposes if the data is not very trustworthy (large sample variation or very small dataset), see our scVI-hub preprint for a case of this.

3

Thank you very much. Would it help to filter the sc data in a certain way bevor training the model for DEG analysis.