In your question, you mentioned that you performed BLAST analysis.

Was it BLAST for annotation ? something like BLASTx against protein sequence database in order to annotate your assembled contigs ?

If yes, one can not conclude that contigs (or transcripts) sharing the same annotation come from the same gene. Depending on the tissues that you studied, you can find transcripts coming from large gene families. For example, by studying a growing tissue, one can expect to find transcripts of genes involved in cell cycle such as cyclins which is a large multi-gene family.

What I mean is that finding a large number of transcripts sharing the same functional annotation is not necessarily surprising. And contrasted expression patterns of genes belonging to multigene family is also frequent.

You also mentioned that the contigs hitting the same gene did not necessarily overlapping each other. This can be due to the fact that these transcripts came from different genes sharing a similar functional annotation (i.e genes belonging to the same family).

Because RNA-Seq is a high throughput technology, I think that looking at all the "problematic" sequences to draw a conclusion is not a good idea.

Before RPKM computation, what you can try in order to optimize gene expression patterns is to remove the reads exhibiting multihits (i.e that mapped on different contigs with the same alignment score) because these reads can not be unambiguously assigned to a unique contig.

I don't know which software you used to do the transcriptome assembly, but I guess that it is usual to assemble many more transcripts than there are genes. Maybe you can optimize a bit the parameters of the transcriptome assembler to reduce the number of assembled contigs. But I think that without a reference genome to guide the assembly, it will be difficult to discriminate splice variant, members of multigene families and bioinformatic bias.

Maybe you can manually curate your assembly by aligning the contigs that "contain" the same gene in order to distinguish splice variants but again, because RNA-Seq is a high throughput technology, I think that it can be an endless and counterproductive task.

Hello,

As you mentioned, genes located on multiple contigs after assemby can be due to duplicated genes, splicing variant or bioinformatic bias.

The question is : Is it a biological fact or a bioinformatic bias ? I think that it is not easy to answer if you don't have a reference genome for the mapping step.

For me the best thing to do to avoid mis-interpretation is to consider each contig as a unique locus.

Combining the RPKM values sounds risky.

Hello Summer,

This might be helpful: http://dwheelerau.com/2013/04/15/how-to-use-deseq-to-analyse-rnaseq-data/

http://compbio.mit.edu/cummeRbund/manual_2_0.html

best wishes

Shab

Hi David,

Thank you for answering my question! That make sense. But I saw one Histone methlyase gene have more than 70 transcripts in the transcriptome. The expression level of these transcripts vary a lot between tissue A and tissue B. Do you have any good way to visualize and describe this? I can easily find other examples like this. Do you usually go back and look at all the sequences and then draw a conclusion? Thank you again!

In your question, you mentioned that you performed BLAST analysis.

Was it BLAST for annotation ? something like BLASTx against protein sequence database in order to annotate your assembled contigs ?

If yes, one can not conclude that contigs (or transcripts) sharing the same annotation come from the same gene. Depending on the tissues that you studied, you can find transcripts coming from large gene families. For example, by studying a growing tissue, one can expect to find transcripts of genes involved in cell cycle such as cyclins which is a large multi-gene family.

What I mean is that finding a large number of transcripts sharing the same functional annotation is not necessarily surprising. And contrasted expression patterns of genes belonging to multigene family is also frequent.

You also mentioned that the contigs hitting the same gene did not necessarily overlapping each other. This can be due to the fact that these transcripts came from different genes sharing a similar functional annotation (i.e genes belonging to the same family).

Because RNA-Seq is a high throughput technology, I think that looking at all the "problematic" sequences to draw a conclusion is not a good idea.

Before RPKM computation, what you can try in order to optimize gene expression patterns is to remove the reads exhibiting multihits (i.e that mapped on different contigs with the same alignment score) because these reads can not be unambiguously assigned to a unique contig.

I don't know which software you used to do the transcriptome assembly, but I guess that it is usual to assemble many more transcripts than there are genes. Maybe you can optimize a bit the parameters of the transcriptome assembler to reduce the number of assembled contigs. But I think that without a reference genome to guide the assembly, it will be difficult to discriminate splice variant, members of multigene families and bioinformatic bias.

Maybe you can manually curate your assembly by aligning the contigs that "contain" the same gene in order to distinguish splice variants but again, because RNA-Seq is a high throughput technology, I think that it can be an endless and counterproductive task.

Thank you for all of your answers and discussion! I had deeper understanding about the analysis!