inconsistent results of rsem quantification

Question

[gnilihzeux]

I find rsem-calculate-expression put out two different results after my twice running with default parameters.
I use the bowtie2 aligner and foud parameter -k will search for one or more alignments and RSEM has a default set --bowtie2-k 200.
So, I guess that rsem will assign multiple alignment reads randomly.
Anyone suffers this ?

[Kristoffer Vitting-Seerup]

How different are the results? Could you do a log2 exp vs log2 exp of the two runs?

[Rob]

The -k 200 parameter does _not_ mean that "rsem will assign multiple alignment reads randomly." The -k 200 parameter means that Bowtie2 will report up to 200 distinct alignments for a read, which RSEM will then determine how to probabilistically allocate in accordance with its model. When there are > 200 distinct "valid" alignments (alignments with a sufficiently hight score to be found and reported by Bowtie2), then, for such reads, Bowtie2 will report only up to 200 alignments. However, it is worth noting that this parameter for use with RSEM was certainly chosen with some care, and very few reads are likely to produce > 200 distinct valid alignments.

[gnilihzeux]

Thanks for your reply. And sorry for my delay. After a long time, I've had a deeper understanding to the alignment parameters of bowtie2.