annotation of local sequence "uniqueness"

[gruberjd]

Hi all-- I am attempting to filter out false "SNPs" in my bowtie assemblies (yeast genomic DNA). One major predictor of SNP falseness is being located in a regions that shows sequence similarity with somewhere else in the genome (e.g., paralogs, telomeres).

I've attempted to limit the read mismapping that underlies these SNPs by tweaking the bowtie parameters, but I still get quite a few. So an alternative would be to ask, for each region containing a putative SNP, what is the local uniqueness?

Does anyone know of a previously generated track that catalogs this property? Or better still, a tool that would perform a sliding window analysis (probably by parsing BLAST or BLAT) to record it for any given sequence?

Thanks!

[twaddlac]

Wouldn't you be able to infer true positives based on the depth and nucleotide ratios at that position? '

Since you have the SAM/BAM output you could try varscan (http://varscan.sourceforge.net/) or the samtools pipeline (http://samtools.sourceforge.net/mpileup.shtml). You can filter these SNPs with these tools and they both provide nucleotide resolution for each putative SNP. More so the samtools pipeline, but I prefer varscan sine it is a bit cleaner.

[gruberjd]

Hi Twaddlac--

Thanks for the reply. I don't think depth can get me all the way. Sometimes the depth is lower than "regular" sequence, but not always. It can also look normal or be higher (which would make sense if reads from two locations in the genome are mapping to the same place). I must tolerate the possibility of real heterozygotes so allele frequency can only do so much, also.

I am already using SAMtools, but I'll check out varscan, Thanks!

[gaffa]

The term "mappability" has emerged to refer to this - using this as a search term is likely to yield results. There are a few tools that compute mappability measures, for example GEM: http://sourceforge.net/apps/mediawik...ility_man_page.

[gruberjd]

thanks gaffa, the right search term is exactly what I needed and I didn't know 'mappability'... something definitely wasn't right about "uniqueness". Thanks!