Can Differential Isoform expression analysis can be performed using DESeq2 package
1
0
Entering edit mode
priyanka.m • 0
@03ddb485
Last seen 1 day ago
India

Hello,

I am want to perform differential isoform expression (DIE) analysis for RNAseq data from human. Can I use DESeq2 for this by inputting the transcript level abundance and getting differential expression of transcripts then performing some downstream manual analysis to find which of the differential transcripts belong to same gene. Or if there is any other R package that can perform DIE analysis by taking in salmon quant files directly.

I know DESeq2 is used for gene-level analysis, I am not sure if it can be extended to transcript level and if there are any underlying parameters I should take into consideration before performing such analysis.

Any advice will be highly appreciated.

Thank you.

RNASeqData isoform DESeq2 • 296 views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 7 hours ago
United States

In the Swish paper we show that, while overall the sensitivity and precision are roughly on target for many gene-level tools, for certain classes of transcripts with high inferential uncertainty, we found the Swish method had better control of FDR (1 - precision). See for example Fig 3 where Swish controls error well for the transcripts with highest uncertainty of assignment. To run Swish, you would just need to have run Salmon with --numBootstraps or --numGibbsSamples of 20 or 30. Swish has a DE vignette similar to DESeq2:

https://mikelove.github.io/fishpond/articles/swish.html

ADD COMMENT
0
Entering edit mode

Thank you so much Michael.

Both the paper and links are very useful.

ADD REPLY
0
Entering edit mode

Hi Michael,

I performed Differential isoform expression analysis using Swish and have got some isoforms. I had previously performed the gene level analysis using DESeq2 package. Since the two methods are quite different can I compare the two results? I also performed gene level analysis using swish which also gives different result than DESeq2. I just need some advice on how I can still retain my DESeq2 results along with swish since we have already interpreted the DESeq2 result.

Sorry to continue it in this post. If required I can delete it and repost.

ADD REPLY
0
Entering edit mode

hi,

Good question. So you can aggregate your swish isoform results to gene level (for which genes are there any significant isoforms) and then plot gene significance vs isoform level significance (e.g. -log p or LFC for the top isoform per gene). I would then examine these genes and isoforms by eye with plotCounts (DESeq2) or plotInfReps (swish).

I have likewise compared DESeq2 and swish results, to show how they qualitatively differ. Take a look here:

https://github.com/mikelove/swish-demos/blob/master/differential-transcript-airway.knit.md#comparison-with-deseq2-at-transcript-level

While this is a txp-level comparison, the points here are still valid for gene level. You can see what type of features will show up as significant for one or the other method. As significance is not deterministic, but also a random variable, and a complex function of the input data (including aspects of the entire count table), it's hard to pin down exact numbers e.g. how many features that DESeq2 finds significant will swish find significant, but we can make some qualitative comparisons.

ADD REPLY
0
Entering edit mode

Thank you for the suggestions. I will try to compare my results and accordingly interpret the two.

ADD REPLY

Login before adding your answer.

Traffic: 231 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6