why the TPM value is not same?

Question

[star]

I would like to do normalizing on my data using TPM methods like what explained https://www.rna-seqblog.com/rpkm-fpkm-and-tpm-clearly-explained/

TPM is very similar to RPKM and FPKM. The only difference is the order of operations. Here’s how you calculate TPM:

1. Divide the read counts by the length of each gene in kilobases. This gives you reads per kilobase (RPK).
2. Count up all the RPK values in a sample and divide this number by 1,000,000. This is your “per million” scaling factor.
3. Divide the RPK values by the “per million” scaling factor. This gives you TPM.

I used the below codes but I do not know why the output is not correct?

CODE:

RPK<- data.matrix(Data [-1] / Data$Length.Kbp)
TPM <- t(t(RPK)*1e6 / colSums(RPK))

Data:

                Length.Kbp    FB_1    FB_2    FB_3
1:15040-15500         0.46       0       4       0
1:108570-109500       0.93       1       5       0
1:248240-249110       0.87       2       1       1

RPK:

                                 FB_1           FB_2    FB_3
1:15040-15500                       0       8.695652       0
1:108570-109500              1.075269       5.376344       0
1:248240-249110              2.298851       1.149425       1.149425

TPM:

                                  FB_1             FB_2    FB_3
1:15040-15500                        0       2577162.0       0
1:108570-109500                70641.81       353209.1       0
1:248240-249110              2000000.00       1000000.0      1000000.0

while for the first row (related value to FB_2) should be like :

8.695652 * 1000000 / 15.221422 =571277.2

[Ram]

Did you try storing colSums2(RPK) in a vector and verifying a few values in it to ensure you're dividing by the right value? There is something odd about the third row - it seems to be exactly 1e6 x original_counts.

Also, your datasets don't conform to the code. If RPK <- data.matrix(Data / Data$Length.Kbp) is exactly what was run, then RPK would also have a column titled Length.Kbp with all values = 1. Did you remove that column?

[star]

Thanks for your reply! Yes, I have removed it and Edited the cod now.

In my cod I just used transpose :

TPM <- t(t(RPK)*1e6 / colSums(RPK))

and it looks work. but I don`t know what exactly happens after two times transposing?

[Ram]

Are you sure you should be using colSums and not rowSums? You're dividing transposed-RPK by per-sample RPK sums, not per-region RPK sums. Try using rowSums instead.

[star]

I want to divide RPK per-sample RPK based on the below explanation:

1) Divide the read counts by the length of each gene in kilobases. This gives you reads per kilobase (RPK).

2) Count up all the RPK values in a sample and divide this number by 1,000,000. This is your “per million” scaling factor.

3) Divide the RPK values by the “per million” scaling factor. This gives you TPM.

[Ram]

Please read those three statements and interpret them to get to the denominator you need to use. I can help you with specific questions, but I will not read English and translate it to reproducible code for you - you should be able to do that on your own.

[Ram]

I have edited your post and updated the TPM object with the formula above. Going forward, please give us the exact code you use - it is impossible to help you when you withhold critical information.