Are all existing softwares on RNA-Seq analysis still useful for Single-Cell RNA-Seq?

wilson90 · 09-17-2013, 07:04 PM

Dear all,

I am currently pursuing my interest in Single-Cell transcriptomics, but I worry that the statistical model optimized for rna-seq might not be applicable to Single-cell or Single-nuclei rna-seq transcriptomics analysis anymore.

Is there any software/packages that are designed for single-cell transcriptomics?

What do you think?

Best Regards,
Wilson

EduEyras · 09-17-2013, 09:14 PM

That's a very interesting question. Do you have some data already?

Eduardo

dpryan · 09-17-2013, 09:55 PM

They're still useful, they're just not directly geared toward that case. Have a look at SAMstrt, which is a tweak to SAMseq, for one example. I suspect that the biggest change is how normalization needs to be done (with spike-ins, though DESeq/edgeR/etc. can handle that, it's just not detailed). The SAMstrt authors mention some additional differences, though lacking data of this type I can't really judge that.

wilson90 · 09-18-2013, 01:05 AM

I have played with some data, and generated some data. Theoretically, it should be a "pristine" dataset.

I wonder if cuffdiff2 is as robust when it comes to single-cell rna-seq data?

wilson90 · 09-18-2013, 01:06 AM

I have played with some data, and generated some data. Theoretically, it should be a "pristine" dataset.

I wonder if cuffdiff2 is as robust when it comes to single-cell rna-seq data?

kopi-o · 09-18-2013, 08:55 AM

I would not count on cuffdiff2. Why not try SAMstrt? The first author of it did extensive tests of different software packages for single-cell data with many replicates and found SAMSeq performed best for his data, whereupon he tweaked it a little bit to improve some things.

dpryan · 09-18-2013, 09:08 AM

I'm with kopi-o in suspecting that cuffdiff2 won't work for you, at least without modification. An absolute prerequisite for any package for this sort of data is to either be able to directly perform the size normalization on the spike-ins and then ignore them, or take user inputable size factors. I don't recall cuffdiff2 being able to do that. I would really recommend that you first give SAMstrt or something similar a try.

Simon Anders · 09-19-2013, 08:00 AM

You should at least give some information on your experimental design to get some useful advice. How many cells are we talking about? How many cell types? And what is the analysis goal? (The majority of single-cell RNA-Seq studies I am aware of are not about differential expression calling.)

dpryan · 10-22-2013, 02:15 AM

Quote:

Originally Posted by Melisa Stewart

All one have to do is search for the best and optimized alternative for themselves and accentuate the liability of transforming the related business into the profitable investment.

Is someone putting posts through google translate multiple times? Alternatively, if you're playing "buzz-word bingo", you need to throw "synergize" in there.

jparsons · 10-22-2013, 09:35 AM

Quote:

Originally Posted by dpryan

I'm with kopi-o in suspecting that cuffdiff2 won't work for you, at least without modification. An absolute prerequisite for any package for this sort of data is to either be able to directly perform the size normalization on the spike-ins and then ignore them, or take user inputable size factors. I don't recall cuffdiff2 being able to do that. I would really recommend that you first give SAMstrt or something similar a try.

Is 'giving SAMstrt or something similar a try' possible yet? As far as i can tell, the github is still just an unorganized pile of R scripts with next to 0 documentation.

dpryan · 10-22-2013, 12:11 PM

Quote:

Originally Posted by jparsons

Is 'giving SAMstrt or something similar a try' possible yet? As far as i can tell, the github is still just an unorganized pile of R scripts with next to 0 documentation.

It's possible, though it's really not as easy as it should be. You can download a zip-file from github (I haven't checked if the .tar.gz file linked from the paper is the most recent) , unzip, and install that (note, it requires samr and impute from Bioconductor). There's a vignette, though the vignette command doesn't actually seem to bring it up. The spike-ins that it's looking for needs to have names starting with RNA_SPIKE_. There's a function SAMstrt.normalization() that seems to actually normalize according to these. For the most part, the library just modifies SAMseq() behind the scenes, so the normal workflow there should work. Allegedly, that is. I don't have any data like this so I can't really proof things. If you're interested in the package, drop an email to Shintaro Katayama and mention that maybe the package should get finalized and uploaded to Bioconductor.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
RNA-Seq: A simple and novel method for RNA-seq library preparation of single cell cDN	Newsbot!	Literature Watch	0	08-23-2012 03:00 AM

09-17-2013, 07:04 PM	#1
wilson90 Member Location: Singapore Join Date: May 2012 Posts: 48	Are all existing softwares on RNA-Seq analysis still useful for Single-Cell RNA-Seq? Dear all, I am currently pursuing my interest in Single-Cell transcriptomics, but I worry that the statistical model optimized for rna-seq might not be applicable to Single-cell or Single-nuclei rna-seq transcriptomics analysis anymore. Is there any software/packages that are designed for single-cell transcriptomics? What do you think? Best Regards, Wilson

09-18-2013, 01:05 AM	#4
wilson90 Member Location: Singapore Join Date: May 2012 Posts: 48	Interesting Indeed I have played with some data, and generated some data. Theoretically, it should be a "pristine" dataset. I wonder if cuffdiff2 is as robust when it comes to single-cell rna-seq data?

09-18-2013, 01:06 AM	#5
wilson90 Member Location: Singapore Join Date: May 2012 Posts: 48	Interesting Indeed I have played with some data, and generated some data. Theoretically, it should be a "pristine" dataset. I wonder if cuffdiff2 is as robust when it comes to single-cell rna-seq data?

09-17-2013, 09:14 PM	#2
EduEyras Member Location: Barcelona, Spain Join Date: Dec 2012 Posts: 17	That's a very interesting question. Do you have some data already? Eduardo

09-17-2013, 09:55 PM	#3
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	They're still useful, they're just not directly geared toward that case. Have a look at SAMstrt, which is a tweak to SAMseq, for one example. I suspect that the biggest change is how normalization needs to be done (with spike-ins, though DESeq/edgeR/etc. can handle that, it's just not detailed). The SAMstrt authors mention some additional differences, though lacking data of this type I can't really judge that.

09-18-2013, 08:55 AM	#6
kopi-o Senior Member Location: Stockholm, Sweden Join Date: Feb 2008 Posts: 319	I would not count on cuffdiff2. Why not try SAMstrt? The first author of it did extensive tests of different software packages for single-cell data with many replicates and found SAMSeq performed best for his data, whereupon he tweaked it a little bit to improve some things.

09-18-2013, 09:08 AM	#7
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	I'm with kopi-o in suspecting that cuffdiff2 won't work for you, at least without modification. An absolute prerequisite for any package for this sort of data is to either be able to directly perform the size normalization on the spike-ins and then ignore them, or take user inputable size factors. I don't recall cuffdiff2 being able to do that. I would really recommend that you first give SAMstrt or something similar a try.

09-19-2013, 08:00 AM	#8
Simon Anders Senior Member Location: Heidelberg, Germany Join Date: Feb 2010 Posts: 994	You should at least give some information on your experimental design to get some useful advice. How many cells are we talking about? How many cell types? And what is the analysis goal? (The majority of single-cell RNA-Seq studies I am aware of are not about differential expression calling.)