Overlap/comparison microarray and RNAseq

[MDonlin]

Hi all,

Not sure if this is the proper forum, but here goes. We did our first RNAseq analysis of a fungal genome. We had 2 conditions (stress and non-stressed) and 2 time points with three biological replicates of each condition and time point. One of the time points was also represented on a microarray experiment (Aglilent arrays) done with the same genome under the same conditions.

I used edgeR to determine the DE genes. There were ~1200 genes DE on the microarray and ~2000 DE for the same time point in the RNAseq experiment. The overlap of DE genes was 590/1200. However the direction of the change was often different between them.

My questions:
1) If anyone done a comparison between an Agilient microarray and RNAseq, what sort of concordance was observed?

2) The edgeR was done after RPKM normalization. How much does that affect the number of DE genes? I have the BAM files from the Bowtie alignment so I can redo the edgeR starting from non-normalized counts.

3) Is it even worth spending time trying to determine the level of concordance because these technologies are too different to compare?

I did see a recent paper in BMC Genomics comparing Affy exon arrays with RNAseq and they found a fairly high level of concordance.

Thanks in advance for any thoughts, comments or advice.

Regards,
Maureen

[Simon Anders]

1. Do NOT feed edgeR or DESeq with normalized data. Both tools need raw counts. Otherwise, you just get nonsense. (I guess we need to put this in flashing red letters somewhere, given the number of people who do it wrong.) So, please redo everything with raw counts. Use, e.g., htseq-count to get the raw counts.

2. When checking for concordance, it is not very helpful to look at number of hits, because genes which are borderline significant will let the concordance look worse than it is. The better idea is to make a scatter plot of log fold change according to microarrays vs log fold change according to RNA-Seq. Plot all genes and use colour to mark significant ones.

[stefanoberri]

3) Is it even worth spending time trying to determine the level of concordance because these technologies are too different to compare?

Well, technologies might be very different, but you are interested in the biological answer and that should be independent of the technology. If the two techniques give you very different results it would suggest there is something wrong with at least one of them.

I wouldn't be to much worried about the small overlap, because these sort of experiments (few replicates, multiple test correction) give you a lot of false negative. A study that used 72 replicated suggested that most genes are regulated. The problem is that the genes that you call differentially express should be concordant on the "direction" among the different technologies...

[qiudao]

Hi Stefanoberrl,
I totally agree with you. I believe that's a very important step to compare RNA-Seq and Microarray. Is there any work done in this regard? or any comment from anyone who has done this or know this? Thanks.

[wanfahmi]

1. Do NOT feed edgeR or DESeq with normalized data. Both tools need raw counts. Otherwise, you just get nonsense. (I guess we need to put this in flashing red letters somewhere, given the number of people who do it wrong.) So, please redo everything with raw counts. Use, e.g., htseq-count to get the raw counts.

2. When checking for concordance, it is not very helpful to look at number of hits, because genes which are borderline significant will let the concordance look worse than it is. The better idea is to make a scatter plot of log fold change according to microarrays vs log fold change according to RNA-Seq. Plot all genes and use colour to mark significant ones.

Hi Simon,

Already got the raw data count from RNA-seq which I used htseq-count. But, how to make a scatter plot to compare both RNA-seq and microarray data? Thank you

[Simon Anders]

What scatter plot are you talking about? This is a two year old thread! You can't seriously expect to get a useful answer if you post one-sentence questions without any context.

[wanfahmi]

What scatter plot are you talking about? This is a two year old thread! You can't seriously expect to get a useful answer if you post one-sentence questions without any context.

Hi Simon,

Sorry for the silly question. Alright let straight forward. I got 2 raw data from two different platform (microarray and rna-seq). As we all know that RNA-seq data perform better than microarray which gave more sensitivity and deep coverage. So, I want to create a scatter-plot to compare between RNA-seq and Microarray data. So the data should be log2-transformed in oder to get the data more easy to read? I would say that the y-axis (rna-seq) and x-axis (microarray) where the dot scattered around the straight line shows the agreement between 2 data. I have around 500 number of gene from each data. What would you suggest to create this scatterplot? Using R? Please help, thank you!

[willemate]

I have the same question. I have my DEG after rnaseq and microarray analysis (DESeq & Affy). I want to compare the 2 datasets in R (scatterplot, Venndiagramm). Any tips or scripts available?