I am doing reference-guided transcript assembly and quantitifaction using Tophat and Cufflinks. Currently, I am running Cufflinks on each sample individually and then merging the assemblies with the cuffmerge tool. Would I get a better assembly by combining all the BAM files from all my samples into one and running Cufflinks on that? Or should it theoretically produce an identical assembly?
(Obviously the second strategy will yield nonsense FPKM values, but I can always requantify against each BAM file individually, so that's not a problem.)
Even if the samples are from the same biological group ?
Unless it is the same sample, you might be better off to run independent tophat analysis to find all the transcripts separately. Then there is always the option of combining all those to form the inclusive 'transcriptome'