TPM vs RPKM

Benh · 09-03-2014, 08:46 AM

Hello SeqAns community,

This may be a naive question, but after reading the paper from Wagner, 2012 and watched the video from Lior, I still don't see the benefit of switching from R/FPKM to TPM.

From my understanding:
- TPM and RPKM are both relative abundance units
- Both are acceptable units when comparing genes from the same sample, but being relative quantity, it is not appropriate to compare TPM/RPKM values from different samples directly. Some (further) normalization is needed.
- In Wagner's paper it was mentioned that avg TPM is consistent (like relative molar concentration) while avg RPKM is not. What is the significant of this?

Thank you!

crazyhottommy · 09-03-2014, 09:08 AM

see here Paper: RPKM measure is inconsistent among samples http://www.ncbi.nlm.nih.gov/pubmed/?term=22872506
and here http://blog.nextgenetics.net/?e=51

Brian Bushnell · 09-03-2014, 09:17 AM

From glancing at the paper, it looks like the difference is that FPKM normalizes by relative mass (or number of nucleotides) while TPM normalizes by relative molarity (or number of transcripts). I don't really see why one is inherently better than the other, but it would make a difference in some cases.

Suppose you have an organism with 3 genes, A, B, and C, such that A is your gene of interest; B is some random short gene and C is some random long gene. You have 2 samples with exactly 1 molar A (in other words, no expression change). Sample 1 has 1 molar B and 0 molar C. Sample 2 has 0 molar B and 1 molar C.

Because C is longer than B, each transcript will generate more fragments. So sample 2 will have a lower FPKM of A than sample 1, even though the molarity of A did not change. Therefore FPKM is not an ideal measure of relative molarity.

However, from a cell's perspective, it doesn't really allocate a certain molarity of different transcripts; rather, it has a finite amount of nucleotides from which it can construct RNA. Furthermore, it has a finite amount of amino acids from which it can construct proteins. So I suspect that transcripts of different genes don't really compete with each other on a relative molarity scale, but on a relative mass scale - the amount of protein generated from some gene should depend on the fraction of nucleotides representing that gene, not the relative molarity. And therefore, to measure effective expression, FPKM should be a better metric.

Thus, in summary, it seems to me that the premise of the paper is basically wrong.

Benh · 09-03-2014, 09:42 AM

Quote:

Originally Posted by crazyhottommy

see here Paper: RPKM measure is inconsistent among samples http://www.ncbi.nlm.nih.gov/pubmed/?term=22872506
and here http://blog.nextgenetics.net/?e=51

Thank you for your response Tommy.

Indeed I have read both articles (actually it's the same paper that I posted), but it seems to me that there is no clear benefit why TPM is better... Simply because the avg_TPM is consistent with avg_rmc?

On a practical level, RPKM, just like TPM, works for comparing genes expressed within the same sample. Meanwhile, neither units are directly suitable for direct comparison (for differential expression anyways)... so nothing's changed here except for the choice of tool .... Please correct me if I'm wrong.

Wonder if anyone noticed anything different with their calculations (e.g. fold change..etc) when different units are used.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
The easiest/fastest way to get from BAM to TPM or RPKM	feralBiologist	Bioinformatics	22	11-01-2016 06:50 AM
Cutoffs for TPM values	mht	Bioinformatics	1	03-16-2015 09:46 AM
A question about TPM normalization	NikTuzov	Bioinformatics	2	08-18-2014 10:58 AM
RNASeq TPM vs. FPKM - What's the point if differential expression tools use counts?	ndovu9	Bioinformatics	1	11-25-2013 10:37 AM
rpkm	Rbc	Bioinformatics	0	01-24-2013 11:38 AM

09-03-2014, 08:46 AM	#1
Benh Member Location: Pacific Join Date: Aug 2012 Posts: 11	TPM vs RPKM Hello SeqAns community, This may be a naive question, but after reading the paper from Wagner, 2012 and watched the video from Lior, I still don't see the benefit of switching from R/FPKM to TPM. From my understanding: - TPM and RPKM are both relative abundance units - Both are acceptable units when comparing genes from the same sample, but being relative quantity, it is not appropriate to compare TPM/RPKM values from different samples directly. Some (further) normalization is needed. - In Wagner's paper it was mentioned that avg TPM is consistent (like relative molar concentration) while avg RPKM is not. What is the significant of this? Thank you!

09-03-2014, 09:08 AM	#2
crazyhottommy Senior Member Location: Gainesville Join Date: Apr 2012 Posts: 140	see here Paper: RPKM measure is inconsistent among samples http://www.ncbi.nlm.nih.gov/pubmed/?term=22872506 and here http://blog.nextgenetics.net/?e=51

09-03-2014, 09:17 AM	#3
Brian Bushnell Super Moderator Location: Walnut Creek, CA Join Date: Jan 2014 Posts: 2,706	From glancing at the paper, it looks like the difference is that FPKM normalizes by relative mass (or number of nucleotides) while TPM normalizes by relative molarity (or number of transcripts). I don't really see why one is inherently better than the other, but it would make a difference in some cases. Suppose you have an organism with 3 genes, A, B, and C, such that A is your gene of interest; B is some random short gene and C is some random long gene. You have 2 samples with exactly 1 molar A (in other words, no expression change). Sample 1 has 1 molar B and 0 molar C. Sample 2 has 0 molar B and 1 molar C. Because C is longer than B, each transcript will generate more fragments. So sample 2 will have a lower FPKM of A than sample 1, even though the molarity of A did not change. Therefore FPKM is not an ideal measure of relative molarity. However, from a cell's perspective, it doesn't really allocate a certain molarity of different transcripts; rather, it has a finite amount of nucleotides from which it can construct RNA. Furthermore, it has a finite amount of amino acids from which it can construct proteins. So I suspect that transcripts of different genes don't really compete with each other on a relative molarity scale, but on a relative mass scale - the amount of protein generated from some gene should depend on the fraction of nucleotides representing that gene, not the relative molarity. And therefore, to measure effective expression, FPKM should be a better metric. Thus, in summary, it seems to me that the premise of the paper is basically wrong.