Multiple mappings and Cufflinks

[Rachelly]

Hi,

Does anyone knows if Cufflinks (and Cuffdiff) use reads that were mapped to more than one location?
(The additional locations appear in my sam file after XA:Z attribute, I used BWA aligner).

Thanks,
Rachelly.

[Rachelly]

From Lior Pachter:
Currently they are assumed to have an equal probability of originating
from each of the locations they mapped to. We have improved on this
approach recently and can now estimate different mapping probabilities
(in the same way we probabilistically map reads to transcripts), in a
way that is similar to the RESCUE procedure in ERANGE.
This is already implemented and will be released in a forthcoming
version of Cufflinks (1.0) imminently.

[DavidMatthewsBristol]

Hi,

I have been worrying about multihit reads as well. I have a published workflow on the Galaxy server (Bristol workflow to get....) that takes a sam file of reads mapped with tophat using the default settings (i.e. with up to 40 hits per read). The workflow then sorts out the unique reads that are mapped in a proper pair from the multi hit reads and keeps both sets. In that way you can use cufflinks to estimate gene expression of uniquely mapped reads and in a separate cufflinks run highlight genes where reads are mapping ambiguously. In this way you at least know which genes may have expression estimates that could be a bit "iffy". This is designed for tophat I'm afraid but you get the idea.
Hope this helps!
David

[frankyue50]

at least for the current version, I remember a read can be mapped to as many as 20 locations.

From Lior Pachter:
Currently they are assumed to have an equal probability of originating
from each of the locations they mapped to. We have improved on this
approach recently and can now estimate different mapping probabilities
(in the same way we probabilistically map reads to transcripts), in a
way that is similar to the RESCUE procedure in ERANGE.
This is already implemented and will be released in a forthcoming
version of Cufflinks (1.0) imminently.