Calculating RPKMs from Tophat results : Your advice requested

[apratap]

I am trying to calculate RPKM on the tophat data but have come across this issue that I believe could skew my results.

My #input reads to tophat are ~49 million. The number of reads reported by tophat to be mapped are ~55 million. I assume I am getting more reads mapped than the total input due to the "--max-multihits 15" option I had set.

Now for RPKM calculation I am not sure what number should I use for total mapped reads.

1. Total reads mapped by Tophat including multireads
2. Total uniquely mapped reads

If I go with #2 then I think I should also remove all multi reads when I am doing the counting for reads mapping to my genes.

Thanks!
-Abhi

[carmeyeii]

Apratap,

I also have a similar question.

When TopHat maps a read to more than one place in the genome, thus producing 2+ alignments for read X, at the end of the mapping process we will have more alignments than reads originally input.

When a program such as Cufflinks does the normalization for a gene/transcript by its length and million mapped reads - which value is it actually using: the total million UNIQUE mapped reads, or the total alignments TopHat found?

My guess is it takes the former, but I cannot find a logical explanation in my head to support this gut feeling....

Any thoughts?

Carmen