Problem: I have low-coverage sequencing data from humans in vcf format. For each sample, I have PL scores / preliminary genotype calls from the software ATLAS (it's a bit like GATK for those who don't know). Because the samples are low coverage (mean 1x, with lots of missing data), I am looking to enhance the genotype calls by using a reference panel and genetic map, before go on to phase the samples. Briefly, the reference panel and genetic map can be used as a kind of prior to inform one about genotypes when there is little data present.

What I've done so far: So far I have been using Beagle4 with a command:

beagle4 gl=lowCoverageSamples.vcf ref=highCoverageReference.vcf map=geneticMap out=lowCoverageSamples.newCalls.vcf

However, Beagle4 is extremely slow, very memory inefficient and quite old. There have been many advances in phasing methods, which are now much more accurate, faster and memory efficient than the beagle4 model. However, I haven't come across any similar improved methods for the genotype re-estimation step.

Does anyone knew of any newer pieces of software which do such a thing, which may more faster and more accurate than Beagle4. It needs to have 2 things in particular:

1) Accept data input in PL score format, as to allow the appropriate representation of uncertainty in low coverage samples.

2) Produce a genotype call / genotype probability for each SNP in each individual as output.


1 Answer 1


In case anyone was interested, a new piece of software for rapidly imputing and phasing low coverage sequence data from genotype likelihoods has just been released. It is called GLIMSPE and is exactly what I was after.

It looks to be much faster (1200x speedup) compared to Beagle 4.0, and is more accurate at imputing low frequency variants. It is also more amenable to use with larger reference panels (such as the Haplotype Reference Consortium panel).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.