My understanding is that if we were to subset a set of genes and carry out DESeq2 just for those genes, it would generate inaccurate dispersion estimates since during estimation of gene-wise dispersion, the fitted curve (in the plot of dispersion vs mean of normalized counts) will lack the information from all other remaining genes in a given sample to correctly shrink gene-wise dispersion estimates. Consequently, this would increase the number of false positives and as a result the DESeq2 results for the subset of genes would be unreliable.

I just want to clarify whether my thinking is correct or downright wrong? If carrying out DESeq2 for a subset of genes of interest can't be done, is there any other reasoning behind it other than my explanation above (i.e, if my explanation is not the exact reason for that)?