Using RPKM in ribosome profiling

[Rogue1]

I am interested in using ribosome profiling to compare the translation efficiencies (TE) of different transcripts in a single condition (as opposed to changes in TE of each transcript between conditions). TE is calculated by comparing the levels of ribosome-protected fragments (RPFs) in one library to the transcript levels in a separate library, so TE = RPKM RPFs/ RPKM mRNA. However, the compositions of the two libraries are very different. For instance, RPF libraries tend to have much higher levels of rRNA contamination than mRNA libraries. I suspect that the two samples will have very different read lengths, transcript levels and feature length distributions.

I'm very new to this field, but reading posts in this forum has made me wary of comparing RPKM values between samples. Is RPKM the best way to normalize these different libraries? Will it introduce different biases in each library, or will this mostly be a scaling issue? Is there a more appropriate between sample normalization method for ribosome profiling?

Thanks for any ideas/advice.

[qqseq]

I think this issue is quite complicated. Use of RPKM for ribosome profiling is by no means the most obvious or appropriate method. The reason why RPKM made a lot more sense for RNA-Seq was due to the nature of random fragmentation that should lead to proportionally higher number of fragments from longer transcripts compared to shorter ones. However, this assumption is not necessarily true for ribosome profiling. I think it becomes especially problematic for samples that have relatively low translational activity. Hence, in conditions where this is the case RPKM may not be suitable for example as shown in http://genome.cshlp.org/content/earl...2.115.abstract.

In terms of calculating TE, I would strongly suggest you take a look at http://www.ncbi.nlm.nih.gov/pubmed/24048356 and http://www.ncbi.nlm.nih.gov/pubmed/21115840
The simple ratio is a biased estimator and several newer studies have been addressing these problems as described in the above studies or using linear modeling type approaches

[Rogue1]

These are great! Thanks so much for replying. They address a lot of problems beyond my initial question.