Dear  Marcio

Good one! Are you sure the interim sequences are not non-coding?  Did you find any specific hits of sub-ranging from those larger chunks of transcript? 

I would frame the hits into two:

1.   Based on Poly As :  Which further can be used to read splice variant data, if any. 

2.  Similar sequences clustered or mapped in the same chromosome. They might again be just isoforms.  In such cases, use some of the transcript reconstruction methods.

The first one is ruled out in this case, as I see.  You would, however see such similar or predicted hits. 

Hope this helps

Prash  

Hi Marco - it is in fact very common for de novo assembled transcriptomes to contain several contigs per gene. If you are using Trinity assembler, there are two reasons. First,  as Prash indicated above, the assembler would keep alternative splice forms as different contigs. Such contigs would share considerable sequence identity but represent different assembly alternatives. Trinity keeps track of their relatedness by presenting them as the parts of the same "component".  Second, the assembler often fails to join parts of the same transcript into a single contig, which result in contigs representing different parts of the gene (looks like this is your case). Such contigs would not share identical sequence and would belong to different "components", but blast to the same gene. The measure of transcriptome fragmentation is called "contiguity", I have some not very optimized but working scripts with which you can compute it for yours: http://www.bio.utexas.edu/research/matz_lab/matzlab/Methods_files/transcriptomeContiguity-1.tgz . The reference of interest in Martin and Wang 2011: doi:10.1038/nrg3068 .

Now, for gene expression quantification, the first kind of multiplicity presents a  bigger problem because the read mapper will think that some reads do not map uniquely when they come from the shared-sequence parts of the same-component contigs. The second kind -  fragments of the same gene - are simply quantified independently, no problem for functional summary there. There are solutions to this, but please let me know if you want me to continue...

In agreement with what Mikhail said, there are a few ways to move forward to visualize and identify trends in your data. You can create tables by curating your transcripts into groups by common Entrez/Ensembl/Uniprot gene IDs and count total number of transcripts, number of differentially expressed transcripts, and number up- and down-regulated transcripts per gene. I have some scripts I wrote in R I can send you if you message me. You can also use heatmaps to see if there is specificity in the expression of your transcripts across samples (I like GENE-E but there are plenty of other options too).

Thank you all for the answers!

Mikhail, please, you can continue your explanation.

Lara, I will contact you by e-mail or message shortly to get these scripts.

Still regarding RNA-seq, in this same set of contigs I got some pairs of contigs that were annotated with same terms, e.g. contigs A and B were annotated as 2-alkenal reductase (nadp(+)-dependent), and they ARE significantly similar according to blastn (see attached file).

Is the alignment in the attached file an indicative of assembly error? It seems that these two contigs could be merged in only one larger contig. Can anyone try to explain to me what is going on here?

The assembler used was the CLC Genomics Workbench software. I don't know the parameters used because I just received the contigs for further analysis.

Best,

Marcio

Hi Marcio,

There are many tool for generating good denovo assemblies. We carryout the following: 1. Use high quality reads after stringent PRINSEQ v0.20.4 filtration.

2. Use trinity  for denovo assembly.

3. Use CD-HIT-EST clustering to reduce redundancy by similar seq cluttering with user define parameter for identity. default 90% identity parameter is used.

4. You will good transcripts, predict CDS and also you can use tool like Annocript0.2 for rest of your work. 

hope this info is useful.

regards

Surendra

Hi everybody,

I would like to thank you all for the suggestions.

Best,

Marcio

Hi Marcio - sorry, I got distracted a bit - so, if you are still looking for solutions to account for multiple contigs per gene: the pipeline here offers an easy way to derive counts from SAM files if you have a table defining which contigs belong to which gene:

https://sourceforge.net/projects/tag-based-rnaseq/

I don't think it is much of a problem if a gene is fragmented int non-overlapping parts. But it is a problem if you have contigs sharing extensive sequence stretches (potentially splice isoforms). In the readme file in the pipeline above there is a simple solution involving clustering such contigs using cd-hit and deriving a table of contigs to "genes" from the results.

Hope this is useful

Dear Mikhail,

Thank you for your useful help! And sorry for taking such a long time to thank you!

Cheers,

Marcio