Sailfish expression value

[Pedamallu Chandra Sekhar]

Hi,

I am new to Sailfish, could you please help me to find which one of the measures are good one out of TPM/ RPKM/ KPKM / Estimated NumKmers?

Also i saw a paper in https://genomebiology.biomedcentral.com/articles/10.1186/s13059-015-0734-x talking about bias-corrected measure is not comparable one.

Can we ignore that and use the quant.sf?

Also, can i compare the TPM in one group to another group of samples directly?

Best

[Rob]

Hi,

On Thursday, March 31, 2016 at 3:38:40 PM UTC-4, Pedamallu Chandra Sekhar wrote:

Hi,

I am new to Sailfish, could you please help me to find which one of the measures are good one out of TPM/ RPKM/ KPKM / Estimated NumKmers?

 

  The new versions of Sailfish (which I highly recommend you use) actually output only two real measures for abundance estimation; TPM and Estimated Number of Reads.

Which one you should use depends on your analysis.  For example, tools for differential expression almost universally require count data (limma voom being the only exception 

of which I know), and therefore you should use the estimated number of reads for this purpose.  For other exploratory analysis (e.g. looking at correlations, doing PCA, etc.) TPM

is the most natural choice.

 

Also i saw a paper in https://genomebiology.biomedcentral.com/articles/10.1186/s13059-015-0734-x talking about bias-corrected measure is not comparable one.

Can we ignore that and use the quant.sf?

  Again, this issue with bias correction has been addressed in the newer versions of Sailfish.  It was most likely a result of the fact that the simulated data in that paper followed a very different distribution than that for which that bias correction

methodology was designed.  Regardless, the new bias correction methodology is much more "conservative" in that regard.  That being said, unless you have a reason to believe that there

is a strong sequence-specific bias in your data, you should be fine running without bias correction and just taking the results in quant.sf.

  

Also, can i compare the TPM in one group to another group of samples directly?

  This is actually a rater deep question.  I would say that for exploratory analysis, this is probably fine.  However, if you're looking for e.g. statistically significant differential expression, you should feed the 

estimated number of reads to a tool that is capable of dealing with that type of data (e.g. DESeq2 or EdgeR work well with Sailfish).

Best,

Rob

 

Best