some questions about running tophat & cufflinks

[songyj]

After reading the manual, I still have some questions about running tophat & cufflinks, as following:

Question 1:
I am a bit confused with the options "anchor-length " and "segmet-length" when running tophat..

In my opinion, the default setting "--segment-length" 25 means a segment read no less than 25, and "--min-anchor-length " 8 means a cut of 8, which is smaller than 25, so what is the exact length of the final read? Is it 25 or 8? What is the specificity of using "--min-anchor-length "?

To address this question, I propose the following process to test a read length of 75, using the tophat mapped genome,
First, tophat will map initial reads to genome, the initial reads length is 75
Second, tophat will split all IUM reads into smaller segments to map again, now the length is at least 25
Final, still unmapped reads will map junctions in the tophat database of possible splice junctions, a mapped read longer than the "anchor length" will confirm a junction, am I right?

Question 2:
If the annotation file is download from public database such as ensembel, does the annotation file make a difference in the output result when running tophat? In other words, is there a difference running tophat with or without the annotation file?

Question 3:
In file “tophat/logs/prep_reads” it reads “6975 out of 28036024 reads have been filtered out”. What is the reason to filter the reads? Is it because the read’s quality is too low or the read can’t mapped genome?

Question 4:
*.diff files were obtained when cuffdiff is finished, if I set the minimum number of FPKM values, like 0.1, can I also keep the number of false positives/negative low after in *.diff files?

Thank you very much.
Best regards,
song