de novo transcriptome differential expression problem

[slavailn]

Hi guys,
The goal of the experiment is to assemble the transcriptome of non-model specie and find differentially expressed transcripts under two treatments.
I've assembled transcripts using reads combined from all libraries with Oases using different k-mer length; selected "best" assembly based on blatsx results and summary contig statistics. Next, reads from each treatment were mapped to assembled transcripts with Bowtie.

Now, the problem: How do I go from sam output to raw reads per transcript/contig in order to feed them into DEseq? Htseq-count script doesn't seem to be an option since it requires annotation.

Any help is greatly appreciated.
Cheers!
Slava

P.S. I'm a biologist, so I get stuck with problems like that.

[dariober]

Hi,
If you have transcripts/contigs it should be quite easy to produce a GTF file, i.e. the annotation file for htseq-count. Can you post more details about your input/output data files? In particular the format of the "best" assemblies.

Dario

[slavailn]

Hi Dario,
the format for input data was simply a fasta file, header of the sequence contains name, length, confidence level and a loci number; the output is a SAM file generated by -S option of bowtie.

Thanks for your reply!
Slava.

[wendy_11]

differential gene exp workflow

Hi slavin,
I am planning to do the same kind of experiment you did. I am having hard time to find a correct workflow to apply to our non model specie. As I can see, you used first Bowtie, and then DEseq.
Why did you choose these programs?
Any suggestions?
Thanks in advance

Hi guys,
The goal of the experiment is to assemble the transcriptome of non-model specie and find differentially expressed transcripts under two treatments.
I've assembled transcripts using reads combined from all libraries with Oases using different k-mer length; selected "best" assembly based on blatsx results and summary contig statistics. Next, reads from each treatment were mapped to assembled transcripts with Bowtie.

Now, the problem: How do I go from sam output to raw reads per transcript/contig in order to feed them into DEseq? Htseq-count script doesn't seem to be an option since it requires annotation.

Any help is greatly appreciated.
Cheers!
Slava

P.S. I'm a biologist, so I get stuck with problems like that.

[slavailn]

Hi Wendy,

Actually I ended up using RSEM (http://www.biostat.wisc.edu/~cdewey/software.html) by Colin Dewey to obtain expression values for genes and isoforms and than compared those between samples. Since I didn't have any biological replicates I simply ended up comparing log2 ratios between normalized expression values output by RSEM. RSEM uses bowtie as its internal aligner. I frequently use DESeq to compare libraries where biological replicates are available, other similar alternatives would be edgeR and baySeq, all are implemented as bioconductor packages, fairly straightforward to use and can account for overdispersion of the data caused by biological variation. All of them require raw read counts as an input, not normalized values.
I routinely use Bowtie for alignments due to its speed, flexibility and well written documentation, although there are quite a few popular aligners like BWA, BFAST, MAQ, SOAPalign, Novoalign etc.

Some help for de-novo transcriptomics can be found here http://seqanswers.com/wiki/How-to/de...cript_assembly.

Hope this helps!
Slava.

[wendy_11]

Thanks Slava,
You helped me.
As I have understood, you can use Bowtie also with non model species, then you proceed with Deseq?
Thanks
Wendy

[slavailn]

Hi Wendy,
Nothing prevents using Bowtie with a non-model specie, the problem may arise from the reads mapping to multiple transcripts that originate from the same locus and you will get a lot those with de novo transcript assembly. If you choose to stick only with uniquely mapping reads, you will be throwing out a lot of real mappings, If you decide to keep reads mapping to multiple locations you will be overestimating expression of a number transcript variants (real or misassemled). I used DESeq only with well annotated species, and I was counting only uniquely mapping reads within known gene boundaries.
For de-novo assembly I used RSEM which is supposed to be able to assign reads to transcript variants, although I'm not sure how well it really works. So far I had only one project with non-model specie it I found it quite hard to work with.

Cheers!