BSNP - Bayesian Genotype Caller w/ No Reference Bias

[bgulko]

We just released the Bayesian BSNP genotype caller we used to compare whole genome human sequences generated from differing technologies (454, Illumina, SOLiD, Sanger and Complete Genomics), and at differing mean depths. It requires a samtools pileup as input, but seems to compare favorably to maq and produces full joint distributions P(Geno & Data) so you can use both the posterior and the amount of data that supports that posterior, in downstream analysis. BSNP also offers full control over the genotype priors, and some sequencing technology specific error compensation (Illumina, Sanger, 454 and SOLiD). Finally, genotype calls are not directly dependent on the reference sequence, beyond any biases already embodied in the alignment.

The C++ source is available (no external dependencies) at http://compgen.bscb.cornell.edu/GPhoCS/BSNP/

Any feedback is welcome!
-- Brad

[aaronh]

Looks interesting! I checked out your website and documentation. However, without some support of insertions/deletions I don't see myself using it. Any chance this will be developed? Also, in the supplement you mention that you compared your snp caller to MAQ/pileup. Have you compared it to the updated mpileup?

[bgulko]

Thank you for your interest!

BSNP relies on pileup's representation for indels, so it accommodates their presence but doesn't call the genotypes for inserted regions. If I get a requests or two I might do so, but its a fair amount of coding. I haven't yet compared it with mpileup.

While mpileup looks like it can produce a genotype likelihood distribution via the AF1 tag, (can be combined with priors to generate a posterior), I'm not familiar with model it uses to generate the likelihoods, so its hard to comment. Unfortunately, I understand that the option to report alignment qualities for each base at each position has been removed from mpileup, so it would be hard to update BSNP to use this new format (BSNP uses this information). The simple consensus -c option has also been removed, but that would be easy to code around.