I want to compare gene deferentially expressed in two conditions (1 and 2). I don't have a GTF reference annotation. 

In my first attempt, I mapped my Illumina reads with TopHat. Then I run Cufflink, and Cuffmerge on my Cufflink outputs for the conditions 1 and 2. At the end, I run Cuffdiff using the Cuffmerge output (1 + 2) and the TopHat output for the two samples of interest. I got genes and transcripts differentially expressed, but I was expecting a quite different output (based on the literature).

Do you guys think I did anything wrong? 

Also, I am not sure the question makes sense, but is the Cuffmerge step essential for the comparison of my two samples? Or could I use my Cufflink outputs for samples 1 and 2 for running Cuffdiff?

Regards,