How to evaluate the similarity between two different samples by using RNA-Seq?

Question

[dz2353]

Hi, there!

I have two samples' RNA-Seq data, one is amniotic epithelial cells(AEC), and another is keratinocyte(KRT). I have done with the upstream analysis and get the original reads count matrix. After that, I did PCA analysis, differential gene expression analysis by using DESeq2. Actually, I want to find out the similarity between AEC and KRT at the gene level. But I do not know how to do that cause I do not think the un-differently expressed genes from the result of DESeq2 can represent the similarity. I only have two samples and each one has one replicate, so I can not do co-expression network analysis. Anyone can help me? Thanks in advance!

[Benn]

I only have two samples and each one has one replicate...

Do you mean you have n=1 for each group? It is impossible to do proper statistics with such a 'poor' design (no offense), please consider adding more biological replicates.

Similarity between samples can be evaluated with clustering, such as hierarchical clustering. But also correlation can be used as a measure for similarity.

[dz2353]

Sorry, I mean each group has two replicates. A_1 and A_2 in AEC group, K_1 and K_2 in KRT group.

[Benn]

Are these technical replicates? For sound statistics you need biological reps... try to calculate correlation between your samples, and make a heatmap such as here. Correlation of 1 means similar, correlation of 0 means not similar.

[dz2353]

Yes, they are biological replicates and I've done with correlation analysis. But actually what I want to do is to find out a gene list that shows the same expression level between two groups. Do you think the complementary set of differently expressed gene list is my target？Thanks for your reply!

[Benn]

Sounds like you are looking for equivalence test, haven't seen that before with RNA-seq data but if that's what you need maybe worth a try.

Answer 1

[Charles Warden]

PCA and a dendrogram with hierarchical clustering (with Pearson Dissimilarity and/or Euclidan Distance as the distance metric) are the main things I would use to assess replicates before differential expression.

Otherwise, I would create a heatmap of differential expressed genes. Even if gene list sizes are similar, you may visually see better consistency of replicates with one method versus another (and I would test DESeq2/edgeR/limma-voom for your n=4 comparison).

[dz2353]

Thanks a lot, Charles.