Hello everybody,

I am using TCGA LUAD RNAseq data for DEG analysis. There are data of 522 patients, only 59 of them have paired (have normal and tumor sample) data without replicates. Can I compare all data of tumor samples (522) against all data of normal samples (59)? Is it possible in statistics? Or is it more logical to use paired samples? How should I design the DESeq2 analysis if it is possible to use unpaired data?

I input the data as summarized experiment and used design for paired samples as below:

> library("DESeq2")
> ddsSE <- DESeqDataSet(data, design = ~ patient + shortLetterCode)
> ddsSE$shortLetterCode <- relevel(ddsSE$shortLetterCode, ref = "NT")
> DE <- DESeq(ddsSE)
> DEresults <- results(DE)

shortLetterCode column have "NT" (Solid Tissue Normal) and "TP" (Primary Solid Tumor) sample types and I used relevel command to make "NT" as reference sample.

Should I remove "patient" from design part for unpaired samples?

Can we test for the tumor vs normal effect, controlling for patient effect by using unpaired samples?

Thanks.

If you have a mix of paired and unpaired samples, you should use limma-voom with the duplicateCorrelation() function. There is not a good way to control for the mix of paired and unpaired with fixed effects models.