alternative splicing in RNA-Seq

[reut]

I am analyzing RNA-Seq data, and need to focus on detecting alternative splicing events.
I need to know which alternative splicing events occurred in the data and how do they differ between samples (increased/decreased inclusion level).

I have started using Cufflinks tools, and I know the splicing.diff file should provide me with some information, but it's on the transcript level and not exon/alt event level.

Does anyone have any recommendations on how to approach this?

Thanks,
Reut

[fongchun]

MISO might be able to help you as it can look at the individual exon level:

http://www.nature.com/nmeth/journal/v7/n12/abs/nmeth.1528.html

I, myself, haven't tried it yet. But the paper claims it can. So that might be one option. A simpler approach might be to just use the simple splicing index metric.

Hope that helps,

Fong

[hong_sunwoo]

I am trying MISO and I think that the tool is very useful for detecting difference in alternative splicing.
It uses pre-annotated AS data sets with multiple categories such as exon skip and alternative last exon. It means you can not discover 'novel' alternative splicing event which is not included in the pre-annotated AS data.
However, for me, it supplies lots of information about AS changes in my samples.

[reut]

Thank you for your replies,
hong_sunwoo, can you please show me/send me an example of it's output?

[hong_sunwoo]

Thank you for your replies,
hong_sunwoo, can you please show me/send me an example of it's output?

Because the data will be used for publication, I can not share you the whole data generated from MISO.

The below is one line of the running result file.

event_name sample1_posterior_mean sample1_ci_low sample1_ci_high sample2_posterior_mean sample2_ci_low sample2_ci_high diff bayes_factor isoforms sample1_counts sample1_assigned_counts sample2_counts sample2_assigned_counts
blank 0.47 0.24 0.76 0.51 0.18 0.75 -0.04 0.75 blank (1,1):63 0:45,1:18 (1,1):36 0:15,1:21

You may check the document for MISO for getting the meaning of this output.

http://genes.mit.edu/burgelab/miso/docs/

[malachig]

Tools for detecting alternative splicing events

In addition to MISO and Cufflinks, ALEXA-seq is another tool that focuses on analysis at the feature level (exon, exon-exon junction, intron, etc.). The manuscript describes the approach. A project level summary of the output can be found here for a series of RNA-seq libraries corresponding to normal breast tissues FAC sorted into various subtypes. Here is a specific example of a gene reported by ALEXA-seq as alternatively expressed: CA12. Another simpler example from a comparison of 5-FU sensitive and resistant cells: UMPS.

[xguo]

Does Alex-Seq allow statistical comparison of two groups of samples for alternative splicing, especially paired analysis for multiple normal-tumor paired samples? As far as I know, Cuffdiff doesn't have the capability to do paired comparison right now.

thanks

[malachig]

Yes, you can define arbitrary numbers of pairwise and groupwise comparisons in any combination.

e.g.

Tumor #1 vs. Normal #1 (pairwise)
All Tumors vs. All Normals (groupwise)
Myc Amplified Tumors vs. Non-Amplified (groupwise)

[xguo]

thanks, malachig,

I'm trying to use Alexa-Seq for the splicing analysis. The comprehensive comprison of different features is very useful for the identification of splicing events. I understand exonJunction and exonBoundary annotation files with different length should be downloaded for sequence libraries with different read length. I'm wondering if I can use Alexa-Seq to analyze sequence data with different read length together.

The second question is about the pre-processing of raw reads. I have fastq file for each library. Can I omit the pre-processing and start the pipeline with the mapping of fastq files to repeats, exons, introns, etc?

thanks
Xiang

[shobbir]

i got back my first dataset from an 75bp paired end mRNAseq run. we have run a cuffdiff, but our bioinformatician, bless his cotton socks, is a complete novice, and I know next to nothing about cuffdiff. i have some simple questions:

we are interested in differences in global splicing. we are comparing normal vs patient samples and the affected patient has a splice site mutation in the causative gene. however although there is a very significant difference for the causative gene in the isoform_exp.diff table (we know that in the patient exon 6 is skipped and in the isoform_exp.diff table we observe that the isoform which includes exon 6 is massively downregulated in the patient data), the gene does not show up in the spicing.diff table. but it should do since we know that the patients have this splice site mutation. why is this?

is the isoform_exp.diff table itself useful for information regarding differences in alternative splicing? or should i only rely on the splicing.diff table?

thanks so much

[Simon Anders]

You could also try out tool, DEXSeq. See our paper (doi:10.1101/gr.133744.111) for a discussion of its advantages compared to the other tools mentioned in this tread.

[shobbir]

looks interesting, thanks. will definitely give it a go. we are currently running MISO - what do you think of this one?

[areyes]

MISO does not take into account biological variation. The disadvantages of this are explained in the paper that Simon mentioned.

[shobbir]

Ok thanks guys, we have run the DEXseq and the results look more promising. I have a question regarding statistics:

We ran 2 experiments, the first was mum vs affected female (4 replicates each) and the second was scrambled siRNA vs knockdown (4 replicates each) - the affected female has a mutation in the same gene we knocked down. In the mum vs affected female the DEXseq detected 2951 genes with differential exon usage with FDR 0.1. However, in the scrambled vs knockdown it only detected 81 genes with FDR 0.1 (39 of these were also present in the mum vs affected female list).

I was wondering could we be less stringent for our cut-off for the knockdown data - when we filter using p-value<0.01 for example we get 1057 genes with 360 of them also found to be present in the patient data.

I know next to nothing about stats, but are p-values completely worthless in this type of analysis? Are they ok to use when we have four replicates each for each condition. Or do p-values not take into account the consistency/variation of counts across replicates? Is this something that only adjusted p-values do?