Differentially expressed exons?

[jiwu2573]

I know Cufflinks and DEGseq can do DE gene analysis. Consider the resolution of mRNA sequence, it will be even more interesting to look at the DE at the exon level rather than the gene level. Does anyone know any available software or adapted method for identification of DE exons for mRNA sequencing?

[Xi Wang]

Certainly, DEGseq can do this job if you give the gene annotation per exon instead of gene. You need write some simple script to split genes into exons, but remember to follow the refflat format.

[jiwu2573]

Certainly, DEGseq can do this job if you give the gene annotation per exon instead of gene. You need write some simple script to split genes into exons, but remember to follow the refflat format.

Can you explain it in more details about making the exon annotation?

All those format conversions are a nightmare for a biology person without too much background in computer science like me. But if you can give more detailed insturctions, I can get help from IT person and get it done.

P.S. Affymetrix is providing exon microarrays now, and I think more and more people will be interested in exon analysis intead of gene. A function of exon expression analysis will certainly be an advantage of DEGseq, which will attract more users as well.

[Xi Wang]

Ok, let's start from the exon regions you already have, for example,

chr1  100    200 -
chr1  1000  1200 -
chr1  2000  2300 -

then you write into the format below:

Exon_1       Exon_1       chr1    -       100         200         100        200        1      100,      200,
Exon_2       Exon_2       chr1    -       1000        1200        1000       1200       1      1000,     1200,
Exon_3       Exon_3       chr1    -       2000        2300        2000       2300       1      2000,     2300,

A script will do this automatically.

If you don't have the data for every exon, you need extract the information from gene annotation, and be careful of the redundant exon regions.

Thanks for you suggestion. We may add this function of analyzing differentially expressed exons in our package in next update. That will be more direct to get what you want.

[Thomas Doktor]

An easier way would be to pull down the exons from UCSC in bed format and then use BEDTools' intersectBed utility to count the number of reads mapping to the individual exons. You can then do count statistics or calculate RPKM values for individual exons. If you save your alignment in SAM/BAM format BEDTools can convert your alignments to bed format, otherwise you will have to write your own script or use another utility to convert your aligned reads to bed format.

I think that you might find it difficult to identify differentially spliced exons this way, because you would find so many differentially expressed exons between samples that reflect differential expression rather than differential splicing. To really identify differentially expressed isoforms I think you need a more dedicated algorithm to compare the individual exon levels within a gene to other exons within the same gene. The problem here is that different exons are sequenced with variable efficiency and also mapped with different efficiency (some might contain repeat regions and). These problems are also present when comparing expression between two different genes, but I would think the problem increases as the length of the element decreases because it becomes more and more vulnerable to sequencing and mapping artefacts. Comparing expression of the same gene between to samples is another matter as they are subjected to the same biases, but of course the dynamic range of the measurement may decrease.

Btw, DEGseq seems like a really nice and easy to use package, my concerns regarding its use in this way is merely of a general nature and not specific to DEGseq.

[steven]

I think that you might find it difficult to identify differentially spliced exons this way, because you would find so many differentially expressed exons between samples that reflect differential expression rather than differential splicing. To really identify differentially expressed isoforms I think you need a more dedicated algorithm to compare the individual exon levels within a gene to other exons within the same gene. The problem here is that different exons are sequenced with variable efficiency and also mapped with different efficiency (some might contain repeat regions and). These problems are also present when comparing expression between two different genes, but I would think the problem increases as the length of the element decreases because it becomes more and more vulnerable to sequencing and mapping artefacts. Comparing expression of the same gene between to samples is another matter as they are subjected to the same biases, but of course the dynamic range of the measurement may decrease.

I totally agree.

[jiwu2573]

If I am particularly interested in a few genes, any easy way to compare those exon expressions within these genes?

[Xi Wang]

If I am particularly interested in a few genes, any easy way to compare those exon expressions within these genes?

If you have the well-defined exons for these genes, you can simply use some tools, such as BEDtools, to get the number of reads falling in these exonic regions. Than, statistic methods can be applied to identify the differentially expressed exons. Or some existing tools, such as DEGseq, can help you with the statistical inference.