27
edits
Changes
From Genome Analysis Wiki
Jump to navigationJump to search→Model for Variant Calling
== Model for Variant Calling ==
== Model for Variant Calling ==
glfSingle uses a likelihood-based model for variant calling. Starting from genotype likelihoods ''Pr(reads| genotype)'' per genomic position, computed from appropriate tools (eg. Samtools BAQ), the likelihoods combine with an individual-based prior ''p(genotype)'' to generate posterior probabilities ''Pr(genotype| reads)''.
Ingredients that go into prior:
*All sites have an equal probability of showing polymorphism:
**P(non-reference base) = 0.001
*When a site shows polymorphism, it is usually heterozygous:
**P(non-reference heterozygote) = 0.01 * 2/3
**P(non-reference homozygote) = 0.01 * 1/3
*Mutation model: Transitions (C <-> T or A <-> G) accounts for most variants, while transversions account for minority of variants
**transition has 2/3 probability
**each transversion has 1/6 probability
*'''New implementation''': Alternative mutation model with uniform (uninformative) prior for transition to transversion ratio
**updated by Yancy Lo, 9/24/2012
**each mutation has a 1/3 probability
**add --uniformTsTv in command line to enable this alternative mutation model
**download glfSingle with this new implementation here: [[File:Generic-glfSingle-2013-09-25.tar.gz]]
== Download ==
== Download ==
