Search
Question: Using non-default thresholds into results call vs subset only (post-results call)
1
2.5 years ago by
rodrigoguarischi10 wrote:

Hi Michael

I would like to focus only on genes with fold-change greater than 2 and I also would like to decrease FDR a bit, to 5%. I initially tried:

> res <- results( dds );
> subset( res , padj <= 0.05 & abs(log2FoldChange) >= 1 );

but than I tried adding alpha and lfcThreshold into to results function:

> res <- results( object = dds, alpha = 0.05, lfcThreshold = 1 );
> subset( res , padj <= 0.05 & abs(log2FoldChange) >= 1 )

I was expecting both to give the same result but I was shocked when I figure out that my gene list decreased from 696 to 153 genes.  Is it because fold-change filter is integrate into p-values calculation? I guess you recommend using the second option once this thresholds will also be incorporated in the independent filtering step, right?

Thanks

modified 2.5 years ago by Michael Love19k • written 2.5 years ago by rodrigoguarischi10
3
2.5 years ago by
Michael Love19k
United States
Michael Love19k wrote:

"Is it because fold-change filter is integrate into p-values calculation?"

Yes. For more details, see ?results, vignette("DESeq2") or the DESeq2 paper.

We recommend using lfcThreshold instead of abs(res\$log2FoldChange) > x.