Human 450K array data and rnaseqv2 data

Question

[arpit.singh1203]

Hello All

I had planned to carry out 450K human methylation data analysis and gene expression analysis to corelate differentially expressed genes to differentially methylated regions for publicly available cancer data from TCGA.

I had used COHCAP to find DMRs in 75 samples (65 T + 10 N).[450 K human methylation data level 1 tcga]

Next I used edgeR to find differentially expressed genes in 68 samples (65T + 3N) (RNA-seq data was missing for 7 N samples) [rnaseqv2 data level3 tcga]

For a proof of concept, I extracted 458 genes from CpG_island_filtered-Avg-by_Island.xlsx (from COHCAP analysis) and only 36 downregulated genes from edgeR analysis.

Only 2 genes were common in both the lists. I am worried if I am missing some important steps as I had expected at least 20 - 30 genes that would be commonly related in both the COHCAP analysis(450K methylation array level 1) and edgeR analysis (rnaseqv2 level 3).

Any suggestions or improvements are welcome!

[Sean Davis]

Are the 65 tumor samples matched in the methylation and expression data? If so, why not simply perform a correlation instead of trying to compare gene sets?

[arpit.singh1203]

Yes the 65 samples (Tumor) are matched with unique tcga ID. I am afraid I don't understand what you mean by co-relating rather than comparing. Could you please discuss a little more on that?

[Sean Davis]

Compute the correlation between the expression and methylation for each gene directly using pearson correlation, for example.

[arpit.singh1203]

I ran DESeq2 on my data and got zero DE genes. I think the following reasons could cause this:

1. Data is taken from TCGA and only 3 samples are "Normal" while 65 are "Tumor".
2. Not having replicates for the data.

Do you think these could be the reasons? And is that why you suggested me to "compute correlation between the expression and methylation for each gene directly using pearson correlation"?

PS: All rows containing zeroes were removed.