data of SCT and RNA is kinda the same thing, it is believed SCT does a "better" job. You can try SCTfor DE but my experience was that the results are subtle. Since satijalab has said in many issues that data of RNA is recommended you think you better use it.

@duocang Does that mean that even when you run SCTransform for normalization initially, you would run NormalizeData and ScaleData subsequently on the RNA assay for DE analysis? While I understand that @satijalab has recommended the usage of RNA assay in many issues, I was just puzzled by the fact that we would ultimately need to run all these functions (SCTransform, NormalizeData and ScaleData) in a single pipeline - even though SCTransform is meant to replace the other 2 functions.

@satijalab once answered the same issue that you @eiden309 asked. here
In the above-mentioned link, @satijalab was up to use obj[["SCT"]]@scale.data for DE, but obj[["RNA"]]@data is recommended.

Thanks for asking. We had anticipated extending Seurat to actively support DE using the pearson residuals of sctransform, but have decided not to do so. In some cases, Pearson residuals may not be directly comparable across different datasets, particularly if there are batch effects that are unrelated to sequencing depth. While it is possible to correct these differences using the SCTransform-based integration workflow for the purposes of visualization/clustering/etc., we do not recommend running differential expression directly on Pearson residuals. Instead, we recommend running DE on the standard RNA assay.

I feel puzzled too and I am not yet confident to say which option is better or more normal/standard to go.

The easiest way is to stick to NormalizeData and ScaleData which is still used in recent vignettes (V4) so I think it is not bad to use them.

What I am doing is to rerun NormalizeData and ScaleData even my data has been processed by SCTransform. At leaset RNA and SCT do not effct each other.

I see, thanks for your input @duocang! Hopefully, @satijalab can make it clearer in their future vignettes as this issue has been raised many times

Hello and sorry for reopening this two-month old thread, but I am struggling to understand which assay I should for plotting my DE results. After one-and-a-half years of working with Seurat, we can all agree on that differential gene expression analysis is done on the RNA assay. However, how should I interpret my findings when a gene that is downregulated in the RNA assay (and therefore downregulated in my DE results) is in fact upregulated in the SCT data slot, and as a result "brighter" on my heatmap?
To give some numbers, in the [["RNA"]]@data slot, a gene has an average expression level of 3.79 for control and 1.48 for disease (thus downregulated in disease), as calculated by AverageExpression(). However, in the [["SCT"]]@scale.data slot, which I have used to build my heatmap, the gene has an expression value of -9.19e-17 in control and -2.71e-20 for disease. The SCT-value for disease is greater than the value for control on the number line, contradicting the [["RNA"]]@data results and my DE results. How should I interpret this? Should I perhaps use another slot for my heatmap?

Dear @Susuqu,
Glad to see you're interested in the question. My research is on hold for the moment while I'm continuing on my MSc degree so I it's been a while since I looked at my data. The only hypothesis I can offer is that this is the result of the SCTransform algorithm correcting the data?
I would still use the RNA assay though, since it feels more meaningful to draw conclusions based on actual data than "virtual" data. In general I have used the SCT slot for heatmaps and as a bridge to data integration, and I have continued to use RNA for all DE analysis.