de-novo differential transcription/splicing question

[karl_s]

Hi folks,
I am relatively new to this kind of analysis. I am interested in discovering novel splice variants and differentially transcribed genes. I have several pairs of inputs (pre and post relapse CLL for 3 individuals). I am using STAR to map paired reads to hg19 without a GTF file. I also am using cufflinks/cuffcompare/cuffdiff to look for differential events. Is there an easy way to remove known transcript variants from the output of cuffdiff so as to focus on the novel stuff? Or, should I be using STAR with an annotated set of indexes and using a GTF file in the search? For example, to create the index for STAR:

STAR --runMode genomeGenerate --sjdbFileChrStartEnd hg19_intron_loci.txt --sjdbOverhang 75 --genomeDir ../blabla.fa

then to run it on a pair of reads:

STAR --genomeDir /data/Genomes/UCSC/hg19/STAR_ANNOTATED --sjdbGTFfile /data/Genomes/UCSC/hg19/knownGene_standardchromonly.gtf --readFilesIn sample1_1.clipped.fastq sample1_2.clipped.fastq --runThreadN 32

Or is there a better way? BTW I can already do the general differential expression analysis using a mapping to the transcriptome (i.e using BWA-eXpress and edgeR, or inverse beta binomial for the statistical test).

I'm also using diffsplice on the star output, and that is kind of interesting. I annotate the regions with annovar.

cheers,
karl_s

[karl_s]

OK, i think i figured out part of this. It seems that cuffcompare is used with the -r flag to compare to a reference, and the output has some flags to indicate how the transcripts are related to a known set of transcripts. Providing annotation in the mapping step maybe just helps the mappers accuracy.

Anyhow comments on this sort of analysis from more experienced folks would be welcome. Generally, the quesiton I am trying to answer is, given two biological conditions, am I seeing anything new.