Wilcoxon test, t-test and negative binomial test for DEG of scRNA-Seq data

Question

[CY]

I tried Wilcoxon test, t-test and negative binomial test across different amount of cells (biological replicates) (3, 6, 10, 50, 100, 500) derived from scRNA-Seq data. My intuition is that, since Negative Binomial distribution can better describe the expression distribution, it can detect more DEG than other two tests. However, it turned out Negative Binomial test always detect less DEG no matter the No. replicates is 6, 10, 50 or 100. Does this make sense? Can anyone share some opnion on this？ Thanks in advance.

[rpolicastro]

If you have biological replicates you should use a pseudobulk approach using DESeq2 or edgeR. See OSCA.

Also one method might be better than another method but produce less results. It depends how well the method is controlling for false negatives and false positives.