I'm using the tophat/cufflinks/cuffcompare tools to deal with the RNAseq to assemble the transcriptome. When I use cuffcompare to compare Cufflinks transcript gtf file with the annotation gtf file, the output tmap lost some transcript id in my Cufflinks transcript gtf file. As the Cufflinks' manual saying, "there is one row per Cufflinks transcript". However, the cufflinks transcript has 44499 unique id and the tmap has 44246 lines. Most of the lost id has 0 fpkm, but there are also 5 lost id whose rpkm is not 0. I don't know why the cuffcompare filter some Cufflinks transcript.