TCGA microarray comparison with RNA-seq

Question

[Vasei]

I downloaded microaaray and rna-seq data for breast cancer of the TCGA project from the firehose portal. When I computed the correlation between expression of each gene in different patients, correlation was high, as I expected. Correlation mean was 0.65 and median was 0.75. Intrestingely there exist genes with 0.-24 correlation between what is reported by microarray and what is reported by rna-seq.

But when I look at correlation between these two methods at one or some samples(patients), I observe near zero correlation. And when I plot the rna-seq vs microarray for these samples, I get a shape nearly like a rectangle!

Am I doing something wrong? I downloaded normalized data for both technologies, but I used log(rna-seq) in my analysis.

Edit: Agilent G450A_07 arrays vs Illumina Hiseq (RSEM normalized)

Answer 1

[Vasei]

I found the answer to my own question! The microarray platform that I mentioned is a two channel microarray and the answer is stated in Large Scale Comparison of Gene Expression Levels by Microarrays and RNAseq Using TCGA Data:

The low Spearman correlation between the two-channel microarray data and RNAseq data is a result of the normalization difference between ratio and non-ratio representations of the data. For two-channel microarray, the intensity of a gene is represented by the ratio of case vs control, which makes two-channel microarray data only comparable within a platform. Since RNAseq is a direct count measurement of gene transcript abundance, a low correlation in this case does not necessarily indicate low concordance between two-channel microarray data and RNAseq data. Poor agreement between the Agilent two-channel array data and the Affymetrix one-channel array data was also observed due to the same reason.

However I just wondered that if one channel is used for the tumor(or normal) sample, what does the other channel shows! After some googling, I think they used Universal Human Reference RNA.