How will trimming low-quality ends of Illumina reads affect TopHat and Cufflinks?

[ecabot]

Hello Bowtie/TopHat/Cufflinks developers and user community,


I want to use TopHat and CuffLinks with 75 bp long paired-end Illumina data (i.e. 2x75). With this particular data set many of the reads suffer from low quality base-calls on either end and I would like to trim the individual reads, rather than simply lop-off a set number of bases off the end of each.

I really don't know the implications of doing this with respect to TopHat and Cufflinks.

-Will TopHat have any inherent problem with non-uniform read lengths?

-Trimming will likely affect the distribution of inner paired-end distances (presumably making them larger). I already know the distribution from the non-trimmed data and I realize that with some initial mapping I can get more accurate representation of the distribution after trimming, but I wonder: How sensitive is TopHat to the settings of the -mate-inner-dist and -mate-std-dev parameters?

-What are the implications of trimming with respect to CuffLinks? In particular, is this likely to this introduce any severe bias in the calculation of FPKM?

Can anyone suggest any other considerations about running this suite of programs with trimmed data?

Regards,

Eric L. Cabot,
Genome Center of Wisconsin
University of Wisconsin, Madison

[dan]

I already know the distribution from the non-trimmed data and I realize that with some initial mapping I can get more accurate representation of the distribution after trimming

Can you expand on this point? i.e. what distribution do you see before and after trimming?

My (untested) feeling was that mapping should not be quality dependent, as either the read maps over a reasonable length or it doesn't. i.e. trimming should not be necessary for read mapping.