Thanks for the question, and to Alex and Jeremy for providing some answers. The good news is that VarScan doesn't make Mendelian or diploid assumptions, so you can set any --min-var-freq threshold you like in searching for RNA editing. The bad news is that RNA editing is tough to investigate with next-gen sequencing even with a good caller. See this article, brought to my attention on a different thread by Sean Davis:

http://nar.oxfordjournals.org/content/early/2013/01/08/nar.gks1443

You might try the BEDOPS suite conversion script vcf2bed to convert your VCF file to BED format (our script tries to preserve as much information as possible — see the comments in the script to see how the various columns are mapped), along with the Noble lab's gtf2bed script to convert GTF to BED.

Finally, use the BEDOPS application bedmap to map the genes in the converted GTF file to the sites in the converted VCF file. For instance:

$ vcf2bed.py < sites.vcf | sort-bed - > sites.bed
$ gtf2bed genes.gtf | sort-bed - > genes.bed
$ bedmap --echo --echo-map-id --delim '\t' sites.bed genes.bed > annotatedSites.bed

The file annotatedSites.bed will be tab-delimited and contain each VCF site datum in BED format, with one additional column containing the GTF-sourced gene name (or names) that overlap with that VCF site by one or more bases.

(If there are multiple gene names associated with the site, these are placed into one column and delimited with semi-colons. Custom overlap criteria can be specified, as well.)

not to evade your annotation question, but i don't VarScan would be a good tool for RDDs if it assumes mendelian alleles in individuals. The problem will be with sites that are edited, say 10%-30% of the time. Maybe the tumor module would call those.