RNA-Seq simulation

[Zimbobo]

Hello,

does anyone know of any software that produces simulated RNA-Seq data. I am interested in questions like how many reads are needed for a good assembly with velvet for example, what read errors produce which problems in the assembly. Thanks in advance for any pointers.

[krobison]

You are probably better off downloading an RNA-Seq dataset from the Short Read Archive -- this is more likely to represent the real biases you will find in RNA-Seq data (such as strand, 3' vs 5', etc)

[brentp]

you could check out FluxSimulator.

[krobison]

The FluxSimulator looks interesting, but someone needs to fix the pages -- there is a consistent typo which makes it impossible to read them aloud in polite company!

[vinay052003]

These are no doubt good suggestions. One simple way would be to take the mRNA sequences from the public database and chop them up randomly (in-silico). Repeat this process couple of times untill you don't get the desired coverage.

[lexa]

you can use dwgsim from the dnaa package. you can give read number, read length (even for paired-end data) and a reference.

maq contains also a simulation tool which seems similar to dwgsim.

[micha]

The FluxSimulator looks interesting, but someone needs to fix the pages -- there is a consistent typo which makes it impossible to read them aloud in polite company!

Keith,
someone fixed the page, I think you can read it aloud now the timestamp of the former html file read 00:58, was probably not the best moment of that day. Thanks for bringing this typo to attention!

[seqmagician]

Could any one please point to me the link to FluxSimulator paper? I do find out in their web pages. Thanks.

[lexa]

as far as I know, there is no paper about FluxSimulator.

[catbus]

There's also "RNA Seq Simulator," which is part of USeq---note that this requires REAL RNA-seq data as an input, and then it simulates various types of factors that cause differential gene expression.

http://useq.sourceforge.net/cmdLnMen...NASeqSimulator

[greggrant]

Originally Posted by Zimbobo

Hello,

does anyone know of any software that produces simulated RNA-Seq data. I am interested in questions like how many reads are needed for a good assembly with velvet for example, what read errors produce which problems in the assembly. Thanks in advance for any pointers.

Please try our simulator BEERS:

http://cbil.upenn.edu/BEERS/

[jingjinghao]

Originally Posted by greggrant

Please try our simulator BEERS:

http://cbil.upenn.edu/BEERS/

Hi,sir, I have tried BEERS. Thank you for your good simulator. It did me a big favor.
Some questions:
(1)BEERS generates reads like: genes from combined gene models(RefSeq, AceView...)-->transcripts-->add polymorphisms-->reads-->add sequence error and position bias. Is it right?
(2)How does BEERS decide which and how many gene and transcript to be "expressed"?
(3)Reads are generated from transcripts according to which distribution?

Thank you very much!

[Jegar]

There seems to be a broad range of RNA-Seq simulators. Has anyone done a comparison, or know of a paper that examines them empirically? I'd just be interested to know how their different features compare. The list I have compiled is this so far (apologies for doubles), some of these may be only for DNA-Seq simulation.

http://www.biomedcentral.com/1471-2164/13/74 - GemSim

http://bioinformatics.oxfordjournals...rmatics.btr708 - ART

https://github.com/jstjohn/SimSeq - SimSeq

https://popmodels.cancercontrol.canc...lux-simulator/ -FLUX Simulator

https://github.com/lh3/wgsim - wgsim in SAMtools

http://omictools.com/simulators2/ - Massive range of DNA-seq simulators

http://useq.sourceforge.net/cmdLnMen...NASeqSimulator RNAseq simulator

http://cbil.upenn.edu/BEERS/ BEERS


Importantly, is it better to use one of these simulators than to just download something from a Short Read archive? In which contexts?

[dpryan]

I imagine such a paper would be difficult to get published (I can already see the reviewer comments of "not novel" and "too trivial", even though such a paper would end up being useful for the community).

Regarding when you might use a simulator vs. an actual experiment, the only benefit to a simulator is that you can know exactly where the reads should align and where their mismatches are. If you need to test the accuracy of an aligner, then that's something you need. Similarly, if you want to test methods for calling SNPs or finding RNA editing sites, then you need a dataset with known changes. Of course the error profiles of the resulting reads are never perfect, so you end up needing to use a real dataset too, just to compare raw alignment/call rates (you obviously can't know accuracy from that).

[Jegar]

Thanks for your reply and helpful clarification.

It sounds like a comparison of simulators might be good for a blog post - shame that a helpful piece of work like that wouldn't get published (I have to agree it would not be considered novel, despite there being no existing published comparison paper).

I am exploring error signatures produced through biological processes in the sequencing workflow, and am attempting to reproduce the workflow in silica. From what I gather, Flux Simulator might have some love for me but if not I'll get Python to do the heavy lifting.

[dpryan]

If you do happen to perform such a comparison, please do post a link to it here and/or on biostars, since I expect many people would find it interesting.

[mbblack]

Originally Posted by Zimbobo

Hello,

does anyone know of any software that produces simulated RNA-Seq data. I am interested in questions like how many reads are needed for a good assembly with velvet for example, what read errors produce which problems in the assembly. Thanks in advance for any pointers.

Why not download a real data set from NCBI and then randomly sample from that to derive pseudo-data sets of varying read depths? That way you would have an realistic baseline to compare to.

[dpryan]

Originally Posted by mbblack

Why not download a real data set from NCBI and then randomly sample from that to derive pseudo-data sets of varying read depths? That way you would have an realistic baseline to compare to.

Note that the person you're replying to posted that ~4 years ago...