[galaxy-user] Mapping paired end reads with Tophat

Hi all, After mapping, I used IGV to have a look at the mapping. There are a lot of mapped reads without pair reads. Should I keep these reads? or Is this a problem for cufflinks analysis? What I tried is: 1. BAM to SAM 2. Filter SAM: set Read mapped in a proper pair to Yes. The result is that only 1/5 reads were left. Can anybody tell me if this operation is proper?? How do you normally optimize the mapping rerults from Tophat? Which considerations should I take into account? Looking forward to your reply. Jiwen