[arthur.yxt]

well i don't know. In fact I find it kinda strange that the flag=19 alignment holds mismatch=3, but flag=83 alignment holds mismatch=1. Can a sequence map to one location but with different mismatches?

[pengchy]

It is indeed strange. But there are 258*3 lines like this of the total 76308 lines in accepted_hits.sam file.

[sdriscoll]

so your files show that it is for sure mapping the same read to multiple locations. if that's the case then what i have to do is scan my sam files and create a count myself filtering out duplicates. then i can use that count to calculate my percentages.

[Cole Trapnell]

Hey Cole - would it be possible in an upcoming release of Cufflinks for you to write the gene expression files with all genes from the GTF file and not only those that have non-zero expressions? In order for one to compare one set to another it would be helpful to have consistent row alignments between those sets of gene expression files. I've written a script to do this already but I feel like it would be a great help for people using this program who don't code.

Just to clarify my question at the top of the thread - and I'm not sure you addressed it - is there a way to determine the percentage of source reads that were successfully aligned? Or does the accepted_hits.sam file contain only unique read alignments (one alignment max per read)? For example when my lab got it's first set of sequencing done they had the alignment run with ELAND which produced a table that showed us this percentage. If you could explain a way I could figure that out from the output files it would be a big help.

Thanks!

A quick question about Cufflinks. I've got output from several data files in the same experiments where we are comparing wild type mice to mutant mice and while comparing the gene expressions I see a lot of the time the expression values for genes that should be similar between two samples jump to other isoforms of the gene. For example I've seen a value of 50 on a gene at a certain isoform and then in the next lane that value of 50 is in a different isoform. What we are wondering is if that is something we should ignore and just interpret the expression of that gene to be the same...but if that is the case then when we are comparing the mutant to the wild type, and we don't have that control, how can we trust that the differences in expression are actual differences or just these isoform jumping events?

To answer the original question: the accepted_hits.sam file may contain more than alignment for each read, as long as they are "equally good" according to TopHat. The specific rules for determining goodness of an alignment will be added to the manual when I get to time to do it. Briefly, TopHat prefers to map mates within a gene's length of each other, contiguously over spliced, and with as few mismatches as possible. These rules aren't perfect, but they work well for me. So to get the actual percentage of source reads that map, you'll need to awk out the unique list of read ids from the SAM file (adding back the /1 and /2 as necessary), and then diff that against the ids from your FASTQ file.

To answer your other questions:

1) About the GTF files: I don't understand - isn't cuffcompare already matching these guys up for you?

2) About the differential expression vs. differential splicing. This kinds of analysis comes with a somewhat complicated statistics problem that we are still addressing. I hope to have a tool (and associated documentation explaining our thinking and approach) added to the Cufflinks distribution in a few weeks that handles this. Keep an eye on the site. I can't say too much more about this right now.

[sdriscoll]

Thanks Cole. Great work, by the way. These tools are moving things forward around here in a big way.