Check coverage from RNA-seq data???

[ritzriya]

Dear All,

I have read in a lot of forums as well as discussions posted on the web, which say that the way coverage is calculated for genome, is not the way we can get for transcriptome data as the transcripts level will change from sample to sample. I agree to this.

All I want to knw is If my client says they have given us 20x coverage in RNA-seq data for a particular organism, then how can I reconfirm that the coverage is really 20x? Is there anyway to do this?

Any insight on this will be help me a lot...

Thanks in advance..

[ritzriya]

No answers yet!! 99 viewers but no solution to this problem?

Please help..

[Jeremy]

Is it normalised RNA?

[ritzriya]

Dear Jeremy,

Well, I am not sure of that as the Wet lab experiment has been done by different group of people. Say, I assume its normalised (I shall ask them for the same though :-( ), then what next?

Thanks..

[Jeremy]

What platform was used? And was it paired-end sequencing?

You should be able to mine the information that you want out of the data that is available. A good way to gauge how well you have covered the transcriptome is by the number of singletons you obtain.

[ritzriya]

What platform was used? And was it paired-end sequencing?

You should be able to mine the information that you want out of the data that is available. A good way to gauge how well you have covered the transcriptome is by the number of singletons you obtain.

It was Illumina GAIIx and yes it is paired-end sequencing.

That was indeed a useful information! Thanks.. but for genome we can calculate, it cannot be calculated for that particular tissue/celltype related RNA-seq reads for an organism??

[seidel]

Wouldn't the way to calculate average coverage for a transcriptome be to just take a description of all the known exons (e.g. from UCSC or Ensembl), and calculate the average depth across the features? Sure there may be novel exons, but as an estimate, average coverage across known features would give an average depth. No?

[Jeremy]

Wouldn't the way to calculate average coverage for a transcriptome be to just take a description of all the known exons (e.g. from UCSC or Ensembl), and calculate the average depth across the features? Sure there may be novel exons, but as an estimate, average coverage across known features would give an average depth. No?

If you have normalised RNA maybe, and even then there is still a fairly large copy number difference between highly expressed and lowly expressed genes. Don't forget though that a different set of genes is expressed in each tissue type, so that method would give zero for many genes simply because they naturally are not present.
If its not normalised then you are taking the average across genes that have expression differences of several fold, which isn't a very good indicator.

[dariober]

Hello,
I agree in that looking for a 'transcriptome coverage' is not sensible since in contrast to the genome, the transcriptome varies over time and tissue.

Maybe a more meaningful statistics to assess the depth of a transcriptome sequencing is in terms of 'transcript detection threshold', i.e. What is the minimal expression level that my sequenced library can detect? So, if in a typical (!?) human cell you have approximately 300000 mRNA molecules (see http://bionumbers.hms.harvard.edu/bi...r=3&hlid=43015) than with 3 million reads you are able to assign ~10 reads to a transcript expressed at a level of 1 molecule/cell. (...I'm aware there is a lot of hand waving here).

Does it make sense, at least in principle?

My 2p
Dario

[ritzriya]

If you have normalised RNA maybe, and even then there is still a fairly large copy number difference between highly expressed and lowly expressed genes. Don't forget though that a different set of genes is expressed in each tissue type, so that method would give zero for many genes simply because they naturally are not present.
If its not normalised then you are taking the average across genes that have expression differences of several fold, which isn't a very good indicator.

I agree with Jeremy. But I know what normalization is - though I am unclear what normalization would mean in terms of RNA-Seq reads. Is it filtering of low quality reads or plainly removing redundant reads from the data? Apologies for the silly question

[Jeremy]

By normalised RNA, I mean the use of a normalised RNA library. Not sure if this is the most appropriate reference, but it gives the idea and a starting point - Construction and characterization of a normalized cDNA library. I think I remember reading about someone who has used that approach (but probably not the exact same method from that reference) in here somewhere, but not sure.

This review RNA-Seq: a revolutionary tool for transcriptomics cites some work where they investigate coverage. But again its not something that can be meaningfully applied to a single sequence run.

[Thorondor]

the coverage wont be homogeneously in your transcriptome.

you could cluster your reads which will give you an idea about the coverage.

but how do your clienst know that he sends you a 20x cov RNA-seq, seems to be more like a wild guess.

[ritzriya]

Thanks Jeremy, I have read the articles and I have a better idea of normalization of a library now.

@thorondor - If I am not wrong, a sequencer can be set to produce the reads with particular coverage required once the sample preparation has been done keeping in mind that approximately 20x coverage needs to be obtained.
P.S.Do correct me .. I am more into Bioinformatics and have lesser knowledge in wetlab..

And on what basis shall I cluster my reads?

[Thorondor]

that might be so, but then you imply that the sample is perfectly normalized, which is not the case. ;-) I am also more into bioinformatics. ;-)

Still it is recommended to trim your reads and i doubt that after trimming your coverage is still around 20x. Or is it not your job to assemble them?

only clustering the reads is not the best option, that's my new conclusion after some more thinking.
better first trim, then assemble, then map back the reads to your assembled contigs. Or assemble with velvet, in the contigs id there will be the coverage of the contig. e.g. >NODE_xxxx_length_xxxx.xxxx_cov_xxxxxx.xxxxx

[colindaven]

If I were you I wouldn't worry so much about these coverage claims. As others have suggested just try looking at how many reads were aligned to each gene / exon/ transcript and look at some summary boxplots.

I have found the bioconductor package edgeR and typical R boxplots to be excellent for this. There's an excellent guide for edgeR on this site too.