How to compare the correlation between the RNAseq data from human and mouse tumor?

Question

[ifudontmind_plzz]

Hello,

I would like to compare the correlation between the gene expression pattern of mouse and human tumor type. We have generated a mouse tumor model (squamous cell carinoma) and have performed RNA sequencing (single end, ffpe) on few tumor samples. There are no biological replicates and controls. We only have mouse tumor samples sequenced. I would like to see if our mouse tumor model show more correlation towards human squamous cell carinoma when compared to other human tumor model. Human Rnaseq data (raw read counts) is available in TCGA database.

Answer 1

[TriS]

the first thing that comes to my mind is to rank your genes by expression and compare the rankings (Spearman correlation). however, TCGA data contain hundreds of patients which makes your comparison more complicate since there is a high heterogeneity in the RNASeq data. using the sheer mean across the samples can give you an idea (maybe)...perhaps a better approach would be to cluster the samples based on gene expression and then compare each single cluster with the mouse data that you have. by then integrating clinical data, this will tell you which clinical phenotype your data approach. limitation is that with only one biological replicate you can't be sure about the replicability of the phenotype/transcriptome that you observed in-vivo

[ifudontmind_plzz]

@tris, How should I rank the gene by expression? The mouse data should be normalised first.

[TriS]

you can rank them from the highest/lowest expressed. what would you normalize your mouse data to? do you have normal tissue data? or did you mean to transform the data? if you compare the ranks, transformation won't really change the ranks

[ifudontmind_plzz]

I dont have normal tissue data for mouse. All i have is raw read counts for few mouse tumors. If im not mistaken, the raw read counts does not directly correlate with the level of gene expression. it also depends up on several other factors like gene length etc.

For human, we can get tumor normal matched samples from TCGA.

[TriS]

FPKM is already normalized per gene length. read here and here

yes, you can get normal and tumor from TCGA but you need to look at what you are comparing. if you only have tumor from mouse I would use the tumor from TCGA for the correlation and then go back and look at the differential expression tumor/normal in TCGA

[ifudontmind_plzz]

tris, I apologize if I am asking a vague question.

How should i get the FPKM for tcga data. For mouse data do i have to compute the FPKM, if so could you suggest a method ?