How to compare RNA-seq with microarray

[tujchl]

HI,
I want to compare my RNA-seq with microarray which is found on GEO to indicate the highly correlation between them. I search many papers but found little information about details of comparing RNA-seq and microarray.
there is my workflow
1. mapping probe seq to refseq to cluster refseq to certain probeset. such as probeset A can be mapped to refseq A and B.
2. calculate reads mapping on these refseq and if one reads can be mapped to both A and B, it is counted only once.
3. calculate pearson corlation using microarray probeset intesity and RNA-seq reads counts.

can you tell me where can I improve? for the correlation between my RNA-seq and microarray is not very good (pearson corelation=0.589)

Thanks

[kopi-o]

Actually your correlation is not that bad, I would say. What do you get if you try Spearman correlation? That might be more appropriate here, as you probably have your RNA-seq and microarray expression values on different scales.

[tujchl]

Thank you for your replying, kopi-o
the spearman correlation is almost equally as pearson, about 0.57.
I use cor.test() in R to calculate spearman and pearson correlation.
I also consider my correlation is not so bad. but I just follow a papper. in this papper the correlation is more than 0.7 (see below). I didn`t calculate such high value even using the same data. So I think there could be something I missed.
I really need help, Thanks

[staylor]

HI,
I want to compare my RNA-seq with microarray which is found on GEO to indicate the highly correlation between them. I search many papers but found little information about details of comparing RNA-seq and microarray.
there is my workflow
1. mapping probe seq to refseq to cluster refseq to certain probeset. such as probeset A can be mapped to refseq A and B.
2. calculate reads mapping on these refseq and if one reads can be mapped to both A and B, it is counted only once.
3. calculate pearson corlation using microarray probeset intesity and RNA-seq reads counts.

can you tell me where can I improve? for the correlation between my RNA-seq and microarray is not very good (pearson corelation=0.589)

Thanks

Hi,

I am trying to do something similar.

I am using Normalised logged expression from the array vs log2(FPKM) from RNA-Seq on x and y respectively but as you would expect, the scales are very different. I calculate the Spearman correlation and get around 0.64.

I echo tujchl's comments and wonder if there is a better method I should be using!?

[Blahah404]

The microarray will detect many fewer genes, so perhaps the original authors took the correlation between the microarray data and a subset of the RNAseq data the same size as the microarray set.

[mbblack]

The microarray will detect many fewer genes, so perhaps the original authors took the correlation between the microarray data and a subset of the RNAseq data the same size as the microarray set.

Not necessarily - it will depend entirely on the read depth of the RNAseq dataset. Many times, in terms of genes actually detected, you will have far more genes with valid array signal than you will have with valid RNAseq counts.

In terms of equal coverage by genes detected, you may need far more sequence than most experiments come even close to collecting.

Equivalence to my mind actually has two meanings in this context. One is what proportion of the genome is detected by signal. In that regard, most genomic microarrays will detect a considerably greater proportion of genes than most RNAseq experiments will, unless the RNAseq experiment involved collecting extraordinary numbers of reads.

The second part of equivalence is statistical significance in genes detected. In that regard, even moderate read depth seems to be enough to either equal or exceed microarrays, for the genes detected in common.

[mbblack]

HI,
I want to compare my RNA-seq with microarray which is found on GEO to indicate the highly correlation between them. I search many papers but found little information about details of comparing RNA-seq and microarray.
there is my workflow
1. mapping probe seq to refseq to cluster refseq to certain probeset. such as probeset A can be mapped to refseq A and B.
2. calculate reads mapping on these refseq and if one reads can be mapped to both A and B, it is counted only once.
3. calculate pearson corlation using microarray probeset intesity and RNA-seq reads counts.

can you tell me where can I improve? for the correlation between my RNA-seq and microarray is not very good (pearson corelation=0.589)

Thanks

Are you aligning the two datasets by annotation in the same manner as the authors of the paper you mentioned? Are you including promiscuous probes in your microarray data?

[tujchl]

HI everyone,
I`m back. I think there are two ways which may work
1. align RNA-seq reads to mRNA seq (eg. refseq) then merge all the isoform to calculate gene expression and compare gene expression with microarray on gene level.

2. in one papper (http://lib.tkk.fi/Dipl/2012/urn100639.pdf) Karolis Uziela introduce a new method which may help

I try both there are indeed some improvement but I partially agree with Blahah404. I think there are some filterations

[Gonza]

Hi all,
I am trying to figure out how to compare arrays vs RNAseq. I am not really familiar with arrays, but I am a little bit concerned that we compare 'apples with pears'. I think RNA-seq and microarrays measure slightly different things - although both using mRNA.

I have calculated spearman correlation for a data frame with both array and RNAseq. Biologically, the tree makes a lot of sense. I just do not know how to 'justify' that i can do this.

Any thoughts are welcome.

### R script final step. all7 is my data.frame with 244 obs. of 20 variables ###

test = dist(cor(all7[,-1], method="spearman"))
a = hclust(test)
plot(a)

[NextGenSeq]

How old is the microarray data?
If they are the old Affy chips they are 3' biased.

We've found that comparing RNA-Seq results with the libraries made two different ways can give horrible correlation. We never use oligo-dT for RT priming if we can avoid it.