[Nicolas]

I think first of all, it means that microRNA and smallRNA cannot be studied by in canonical RNA-Seq experiment, aimed at studying "longer" RNA (mRNA, lincRNA).
For smallRNA, you do not need to assemble spliced transcripts, and therefore, you could use another type of analysis: something along the lines of:
1) remove the adapter
2) map with Bowtie/BWA
3) count the reads mapping to miRBase hairpins + your smallRNA genes of interest

[honey]

Small RNA

So that means Cufflinks cannot be used for micro RNA or small RNAs. Any one differ on this please.

[dietmar13]

steven, you are lucky,

both data are not from us, i am only playing around...

12 normal vs 12 colon cancer, paired:
sra:
SRP007584

[steven]

publicly available: GSE25070

http://www.ncbi.nlm.nih.gov/projects...i?acc=GSE25070

Thanks! Is the corresponding RNA-seq data available too by any chance?

[dietmar13]

@arvid

publicly available: GSE25070

http://www.ncbi.nlm.nih.gov/projects...i?acc=GSE25070

[arvid]

@dietmar

What method did you use to calculate expression on the microarrays and what kind of microarrays were they?

[dietmar13]

short genes

hello cole and epi,

i have made some comparisons of FPKM values (calculated from count data, not with cufflinks) with corresponding microarray data regarding gene length, and found some interesting details.

it is true that FPKM values from genes with length below 500 bp correlate much less with expression values derived from microarrays, but the differences between tissues (i.e. normal vs. cancer) from small genes correlated even better (NGS vs microarray) than the differences from larger genes (interestingly, the correlations go continuos down. see figure).

and in most study designs, the differences from two conditions are important, not the absolute expression values. therefore, i would not exclude small genes from statistical analysis!

in green and blue are the correlations of lg2_FPKM values from each 12 normal and 12 cancer tissues with corresponding lg2_microarray_expression values from 26 normal and 26 cancer tissues. in red the correlations of the differences (lg2_FPKM_NORMAL - lg2_FPKM_CANCER vs. lg2_microarray_NORMAL - lg2_microarray_CANCER) are shown. on the x-axis genes are grouped according gene length (and the number of genes in each bin are shown), e.g. 190 genes are below 500 bp length.

[epi]

Hi Cole, Thanks for your post. I keep reading your comments here which are useful for many including me. I asked a similar question, with a twist, here: http://seqanswers.com/forums/showthread.php?t=17992

Can you comment please. In short, it is about how to deal with larger(>300 bp) transcripts with high FPKMs.

This issue has been discussed elsewhere on this board. As Nicholas points out, RNA-Seq really isn't reliable for very short transcripts. The reason is that all the fragments that map to these transcripts come from the "tail" of the distribution of library fragment lengths. That is, fragments that map to microRNAs are much, much shorter than most fragments in the library - by design in the RNA-Seq protocol, which size selects away very short inserts. Thus, Cufflinks infers that even though relatively few fragments actually mapped to the microRNAs, there were probably TONS of individual microRNA molecules in the transcriptome before all of the various size selection parts of the protocol kicked in. Cufflinks accordingly increases the FPKM of these short transcripts to compensate for the bias against short fragments in the library.

This compensation was designed to improve accuracy for transcripts that are in the 500bp-1kb range - for longer transcripts, the "edge effects" due to library fragment size aren't much of an issue. However, I wouldn't trust FPKM values for transcripts shorter than your average fragment length. There's really just not enough data in most standard RNA-Seq libraries to say much about small RNA abundance.

I should also point out that other methods use this same bias correction technique (RSEM for example). As far as I'm aware, the "count-based" methods don't, but that doesn't mean they shouldn't. Most of those methods are strictly for differential analysis, where any edge effects are assumed to be affecting each condition the same way. That may or may not be the case in your data.

In any case, the quick answer to this problem is to simply remove or ignore transcripts shorter than around 300bp from your GTF. In a future version, we will be flagging these transcripts as too short for reliable quantification where appropriate.

[Cole Trapnell]

This issue has been discussed elsewhere on this board. As Nicholas points out, RNA-Seq really isn't reliable for very short transcripts. The reason is that all the fragments that map to these transcripts come from the "tail" of the distribution of library fragment lengths. That is, fragments that map to microRNAs are much, much shorter than most fragments in the library - by design in the RNA-Seq protocol, which size selects away very short inserts. Thus, Cufflinks infers that even though relatively few fragments actually mapped to the microRNAs, there were probably TONS of individual microRNA molecules in the transcriptome before all of the various size selection parts of the protocol kicked in. Cufflinks accordingly increases the FPKM of these short transcripts to compensate for the bias against short fragments in the library.

This compensation was designed to improve accuracy for transcripts that are in the 500bp-1kb range - for longer transcripts, the "edge effects" due to library fragment size aren't much of an issue. However, I wouldn't trust FPKM values for transcripts shorter than your average fragment length. There's really just not enough data in most standard RNA-Seq libraries to say much about small RNA abundance.

I should also point out that other methods use this same bias correction technique (RSEM for example). As far as I'm aware, the "count-based" methods don't, but that doesn't mean they shouldn't. Most of those methods are strictly for differential analysis, where any edge effects are assumed to be affecting each condition the same way. That may or may not be the case in your data.

In any case, the quick answer to this problem is to simply remove or ignore transcripts shorter than around 300bp from your GTF. In a future version, we will be flagging these transcripts as too short for reliable quantification where appropriate.

[Xiaobin]

These genes don't seem to be that short. There must be other reasons.
I suggest you try count method first. Cufflinks is just too complex to be understood.

[honey]

High RPKM

That does not make sense to me. Unless it is an option in either Cufflinks or Cuffdiff, but I have never saw a log relationship between Cufflinks and Cuffdiff outputs.

Honey, how did you run Cufflinks? RABT mode or simple "quantification" mode? How long are the genes with super-high RPKM?


It seems to me that Cufflinks has a tendency to report super-high RPKM for very short transcripts (such as microRNA). I now routinely filter out the transcripts shorter than the expected fragment size (from the GTF annotation file). I think there is a good rationale to filter them out, because they can not be accurately captured by the RNA-Seq protocol....

In RABT mode, Cufflinks also reports a large number of short transcripts with crazy high values. A solution could be to re-quantify the discovered transcripts with something like BEDtools or HTSeq-count...

I used simple quantification

So you mean probably count method is better?

[honey]

very high RPKM

Here are three examples with genomic coordinates

CGA - 87795222 to 87804824
KISS1 – 204159469- 204165619
TFP12- 93515745- 93520065

Should I then go back to count method?

Thanks for all your help.

[Xiaobin]

Are those genes very short?
As cufflinks will remove the fragment length from gene length in calculating FPKM, sometimes it will give this kind of results.

[Nicolas]

I think that these values should be taken to the log(10).. this is not documented, but my suspicion.

log(10) values from cufflinks roughly equals FPKM values from cuffdiff..

That does not make sense to me. Unless it is an option in either Cufflinks or Cuffdiff, but I have never saw a log relationship between Cufflinks and Cuffdiff outputs.

Honey, how did you run Cufflinks? RABT mode or simple "quantification" mode? How long are the genes with super-high RPKM?

It seems to me that Cufflinks has a tendency to report super-high RPKM for very short transcripts (such as microRNA). I now routinely filter out the transcripts shorter than the expected fragment size (from the GTF annotation file). I think there is a good rationale to filter them out, because they can not be accurately captured by the RNA-Seq protocol....

In RABT mode, Cufflinks also reports a large number of short transcripts with crazy high values. A solution could be to re-quantify the discovered transcripts with something like BEDtools or HTSeq-count...

[honey]

very large RPKM

It is human genome so it is not small.
The egnes which have very high RPKM values are relavnt to biology of the tissue samples, but my problem is how to provide a scientific rational that our results are not nonspecific.
Thanks for the input