...
Let's revisit that pipeline diagram here.
Paths through the Tuxedo workflow
...
For each separate dataset representing a specific replicate and condition, cufflinks assembles a map of genomic areas enriched in aligned reads. Cuffmerge then takes the set of individual assemblies and merges them into a consensus assembly for all the provided datasets. The consensus may include known annotations if you have provided those to the program.
Step 4: What does cuffcompare do? (Optional)
Cuffcompare allows you to compare your assembled transcripts to existing annotation.
Step 5: What does cuffdiff do?
Next, cuffdiff uses the consensus splice variant annotations (and/or the known splice variants) to quantify expression levels of genes and isoforms, using FPKM (fragments per kilobase per million reads) metrics.
Let's look at the commands to perform these steps and how the output files look...