I have a number of exome (and some wgs) sequences of tumors with no matched blood sequence data. In order to call somatic mutations, the approach that I'm taking is to compare the tumor sequence to the human reference genome, filter out variants that are known SNP sites, and assume the rest are somatic mutations. Not perfect, but reasonable.
The problem I'm having is finding a control .bam file to use with mutation callers like MuTect, SomaticSnpier, etc. Basically, I need a .bam file corresponding to the reference human genome (mostly assembly hg19 for the tumor data) to compare to the tumor .bams, but I don't know how to go about creating one from the reference fasta in the absence of read coordinates. Is there a straightforward way to get an input bam file that uses the consensus sequence of hg19?
Wouldn't a more sensible approach be to simply use GATK/samtools/freeBayes/etc. to call variants and then just filter the resulting VCF file with dbSNP and 1000 genomes variants?
Actually, I agree, that sounds like a better idea.
I don't think that I can do that with most of the variant callers that I like to use, they all require a reference .bam file.
All of the callers you like to use are designed around having matched control samples. You don't have that and using a consensus BAM file is not going to help you appreciably. Use the right tool for the job, don't try to shoe-horn all problems into the same solution pipeline.
Devon has the right answer. Those callers are not appropriate for your data. They make specific assumptions related to having a matched normal, and the results will be at best, disappointing, and at worst, flat out wrong if you try to use them.