I am working with Triticum urartu, a diploid ancestor of bread wheat. Haven't tried tophat, but bowtie gave a very poor alignment for the seq data against the reference.

Here's an article..might be you get answer of your queries.

http://www.nature.com/nmeth/journal/v8/n6/full/nmeth.1613.html

Not so sure: I think you can use trinity pipeline for denovo transcriptome reconstruction. I heard it works well with plant poly ploidies.

I am not really looking 'build' a transcriptome, but rather am trying to decipher gene expression information on the basis of mapped reads from RNA-seq data.

There are multiple solutions to your problem.

I assume you don't have well annotated reference genome , since you are using some contigs assembled etc . Here two things you can do.

1. Use tuxedo protocol and cuff links to assemble all the splice variants and get your results. But you don't have well annotated genome then you should not use reference guided (gtf) cuff links, but since you used contigs in the starts (while using top hat) which is also not well annotated, it's likely you won't get a good mapping on your annotation.

2. A simple solution to your problem i think is. Assemble your transcriptome using trinity then do differential gene expression using your assembled transcriptome. There is a nature protocol paper about trinity you can use that for this purpose. Triniy pipeline uses edgeR for differential expression.

Thanks

My apologies for not responding earlier.

Yes, my reference genome isn't well-annotated. I have not used contigs, but individual genes, and used BWA for alignment, which so far has given me the best mapping.

I will read up on Trinity, but I don't see how it would help, since I am working with a genome that is over 85% repetitive, and a polyploid.

I am sure you must have seen this paper, i think you might need to play with the parameters of some programs which deal with repetitive genomic elements etc:

http://www.nature.com/nrg/journal/v13/n1/pdf/nrg3117.pdf%3FWT.ec_id%3DNRG-201201

in case you don't have access i am also attaching .pdf

also see this:

http://www.amjbot.org/content/99/2/267.full

I haven't read any of these so my guess is these can direct you well for choosing best strategy for your data.

Remember these de novo assemblers can take hundreds of gb of ram.

Thanks

Here is second attachment.

Thank you, the papers look interesting. I will have a detailed read.

Hello,

I would suggest you to check the following in order

- What is the depth of your sequencing experiment. If the sequencing is not deep enough many lowly expressed tissue and developmental stage specific genes will be incompletely mapped.

- The reason is because of PCR amplification (in the current sequencing protocols), abundantly expressed genes can mask the the small fraction of RNA coming from less abundant and specific genes.

- Check what percentage of partially mapped genes are protein coding, simply by BLASTx search against a protein database. Non-coding sequences have a higher probability to be incompletely represented in a RNAseq experiment specially if they overlap a repetitive region.

- Try both de-novo and reference based assembly. Compare the gene models of both manually on a genome browser. It may give you a fair idea of how the assembly can be improved.

- For reference based assembly, Tophat used with the --prefilter-multihit option can reliably reduce error due to repeats. I would allow all reads mapped 10 or less times on the reference genome for the downstream assembly of tophat reads by cufflinks. This option of allowing reads with 10 or less mapping is a balance between specificity and sensitivity of the mapping.

Thank you Swaraj. These are wonderful suggestions.

I am very glad I decided to discuss this on research gate. The suggestions and feedback are all very useful.