Hi, Dr Love. I found after lfcShrink, the stat value is throwed out . I am just curious about it, becasue someone recommend using stat value as GSEA statistic. Is it means that after lfcShrink, the stat value is no more accurate, so it no needs to be matained?

here is just the result in DESeq2 manual

dds <- DESeq(dds)
res <- results(dds)
res

## log2 fold change (MLE): condition treated vs untreated 
## Wald test p-value: condition treated vs untreated 
## DataFrame with 9921 rows and 6 columns
##               baseMean log2FoldChange     lfcSE       stat    pvalue      padj
##              <numeric>      <numeric> <numeric>  <numeric> <numeric> <numeric>
## FBgn0000008   95.14429     0.00227644  0.223729   0.010175 0.9918817  0.997211
## FBgn0000014    1.05652    -0.49512039  2.143186  -0.231021 0.8172987        NA
## FBgn0000017 4352.55357    -0.23991894  0.126337  -1.899041 0.0575591  0.288002
## FBgn0000018  418.61048    -0.10467391  0.148489  -0.704927 0.4808558  0.826834
## FBgn0000024    6.40620     0.21084779  0.689588   0.305759 0.7597879  0.943501
## ...                ...            ...       ...        ...       ...       ...
## FBgn0261570 3208.38861      0.2955329  0.127350  2.3206264  0.020307  0.144240
## FBgn0261572    6.19719     -0.9588230  0.775315 -1.2366888  0.216203  0.607848
## FBgn0261573 2240.97951      0.0127194  0.113300  0.1122634  0.910615  0.982657
## FBgn0261574 4857.68037      0.0153924  0.192567  0.0799327  0.936291  0.988179
## FBgn0261575   10.68252      0.1635705  0.930911  0.1757102  0.860522  0.967928

resLFC <- lfcShrink(dds, coef="condition_treated_vs_untreated", type="apeglm")
resLFC

## log2 fold change (MAP): condition treated vs untreated 
## Wald test p-value: condition treated vs untreated 
## DataFrame with 9921 rows and 5 columns
##               baseMean log2FoldChange     lfcSE    pvalue      padj
##              <numeric>      <numeric> <numeric> <numeric> <numeric>
## FBgn0000008   95.14429     0.00119920  0.151897 0.9918817  0.997211
## FBgn0000014    1.05652    -0.00473412  0.205468 0.8172987        NA
## FBgn0000017 4352.55357    -0.18989990  0.120377 0.0575591  0.288002
## FBgn0000018  418.61048    -0.06995753  0.123901 0.4808558  0.826834
## FBgn0000024    6.40620     0.01752715  0.198633 0.7597879  0.943501
## ...                ...            ...       ...       ...       ...
## FBgn0261570 3208.38861     0.24110290 0.1244469  0.020307  0.144240
## FBgn0261572    6.19719    -0.06576173 0.2141351  0.216203  0.607848
## FBgn0261573 2240.97951     0.01000619 0.0993764  0.910615  0.982657
## FBgn0261574 4857.68037     0.00843552 0.1408267  0.936291  0.988179
## FBgn0261575   10.68252     0.00809101 0.2014704  0.860522  0.967928

Best wishes Guandong Shang

After lfcShrink, we have a posterior estimate and posterior SD for the LFC, but we no longer use that to compute a Wald statistic (stat). One reason that people use a Wald statistic with methods like GSEA is because the MLE of the LFC has high variance, but this is not the case after Bayesian shrinkage. I think you can just use the posterior LFC.