How to identify transcripts that are full-length?

[ccard28]

Hello Everyone,
I am working with bovine RNA-Seq samples in which I have a genome to work with. Currently I am just using the simple pipeline of tophat and cufflinks to determine what transcripts are present in my samples. A key aspect to our research is determining what transcripts are possibly present in our sample full-length(completely in tact) because we are interested in what transcripts could be translated into proteins downstream.

Can you use the coverage value from cufflinks to determine this? Is there a tool that is out there to determine full-length transcripts from the seq data? Currently we have done some manual checks through the UCSC browser to determine if all exons and 3'/5' UTRs are covered by our tophat accepted hits but this is tedious and impossible to do for thousands of transcripts in a reasonable amount of time manually. Any help would be much appreciated.

Thank You,
Chris

[mbblack]

I'm a little confused about what you are trying to do? RNA-Seq is by definition randomly fragmented short reads. You have not sequenced anything even close to a full length transcript. Trying to infer which ones were originally present as full length is going to be next to impossible with RNA-Seq data, at least with any confidence of being even close to correct.

There are some methods for generating data to do exactly what you want - one I read recently was RNA-PET (methods in mol. biol. 2012. vol 809:535-562). But it seems to me you really need very different data to identify whole length transcripts.

I do not know how good cufflinks EM algorithm is at inferring what you wish, but it is an inherently very difficult and problematic thing to even try. I don't pretend to understand the math behind it all, but you should read

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1475746/

and then

http://arxiv.org/abs/1104.3889

[ccard28]

I realize that you can not infer that transcripts are full-length because they are in fact generated off of small reads. However, we have a cell population that includes a lot of degenerated transcripts so many of our transcripts if you look at the mappings are not even fully covered by fragments generated from our sequencing run. We need to separate out the transcripts that get 100% coverage from base 1-X and then we take leftover sample and do PCRs to check if these transcripts are indeed full-length by using primers that cover the entire transcript as well as by doing RACE RT/PCRs. I know the sequencing data itself can not infer full-length transcripts but due to our high degree of degraded transcripts where entire ends of many of our transcripts are not covered by our sequencing(ex. no sequencing hitting the last couple of exons) I am looking for a way to filter out only those transcripts that get 100% base coverage from our sequencing and then we can do further testing via various PCR methods to determine if the transcript is actually fully intact.

So I guess my question is better phrased is there a way to pull out transcripts that get 100% base coverage as opposed to those that are "full-length". I hope this makes more sense.

Thank You,
Chris

[Jeremy]

so many of our transcripts if you look at the mappings are not even fully covered by fragments generated from our sequencing run.

This sounds like you do not have the required sequence depth. Which sequencing technology did you use and how many reads per sample do you have?

The fastest way to exclude genes that did not get fully covered would be to apply a min read count threshold.

[ccard28]

We have 37,000,000 x 2 paired-end Illumina reads that are 100bp in length.

We are working with a unique cell population that has been shown to have degraded/truncated transcripts. Many of our transcripts do appear to have full coverage but there are also a lot that do not. I believe that many of our transcripts not being fully covered by reads is indicative of our unique cell population and not a lack of sequence depth(though I am a novice when it comes to RNA-Seq so I could be wrong here). Due to the nature of our cell population that is why I want to filter out reads that have all bases covered from the dataset so that we can further verify/explore these transcripts that could be full-length.

How would you go about applying a min read count threshold to filter these out? Wouldn't this be difficult as reads could all be on one end of a transcript and the fact that transcript lengths are variable?

[Jeremy]

Right, the sequencing depth is good. I see what you are trying to do now, I don't know of anything that can do that for you, you may need to script something yourself. I haven't done or read of anything like this but I have done de-novo RNA assembly and differential expression analysis (disclaimer: so what follows are just ideas)

If you expect there to be a mixture of partial/degenerate transcripts and normal transcripts then it gets tricky. If you have control samples that should be normal then you could look at differential expression on a per exon basis. Use HTSeq to get the read counts per exon then something like DESeq2. If you don't have controls, read depth variance per transcript would be helpful but you run the risk of false positive with sequence that just amplifies poorly on Illumina.

Mapping to a transcriptome might be better than mapping to the genome for this, but then you have the problem of splice variants.