Tissue-Specific And Housekeeping Gene Ratio

Question

[manjovial]

I work both on RNAseq data ( 10 tissues) and Microarray data ( 70 tissues) . The problem is that I find microarray data to be false negative for many of the housekeeping genes. The ratio of tissue-specific to housekeeping genes from RNAseq data comes around 1:10 whereas by microarray it comes 1:2 . This is a huge difference and I don't know which one is more reliable.What is the ratio of tissue-specific to housekeeping genes in the cell? Which dataset I should consider for my study?

[Michael]

which organism? what kind of analysis did you do with the data? absolute quantification (or expressed not expressed) or differential expression (DE). You cannot compare results of absolute quantification and DE results directly, further it might me difficult to compare p-values from microarray DE with RNA-seq DE. Also, I don't think it is surprising to see RNA-seq is more sensitive than microarrays in general.

[manjovial]

Its human's gene expression data. I am looking for absolute present and absent (which is decided by taking a cut-off value) in both the cases. I am studying certain properties of tissue-specific genes and housekeeping genes and want to show that my results are gene expression data independent.

[Mikael Huss]

What do you mean by "false negative", exactly? That the expression for the supposed housekeeping genes varies too much (according to some criterion) across the tissues, or?

[manjovial]

False negative here means genes which are really housekeeping (by prior knowledge or swissprot) are not showing signal intensities above decided cut-off value in all the tissues .

Answer 1

[Michael]

If it is human you could compare your results with the ENCODE data or in general with data from ArrayExpress and run the same calculations. Assuming you have chosen a reasonable cutoff for your detection level (signal detectable with high probability above noise level, but arbitrary cut-offs are problematic, and MAs are not good to measure absolute expression), it doesn't look surprising to me that RNA-seq should have 5x higher dynamic range; house-keeping doesn't really mean they are all highly expressed, does it? There are also pitfalls with the array designs (e.g. using suboptimal reporter sequences on the array, which platform is that anyway? Affymetrix?)

Also, I would run a (rank) correlation analysis of the expression values for different groupings of genes: all, somewhat arbitrary 'housekeeping', highly/ lowly expressed genes.

Maybe, if you give provide more details about your experiment, others can say more.