Major Minor

We allow the major and minor to be determined from either the counts of nucleotides, based on genotype likelihoods, specified by the ancestral/reference or even force both major minor to specific bases Sites, which can be useful if you compare with HapMap data etc.

Brief Overview

../angsd0.567/angsd -doMajorMinor 
	-> angsd version: 0.567	 build(Dec  7 2013 17:25:57)
	-> Analysis helpbox/synopsis information:
-------------------
analysisMajorMinor.cpp:
	-doMajorMinor	0
	1: Infer major and minor from GL
	2: Infer major and minor from allele counts
	3: use major and minor from bim file (requires -sites afile.bim)
	4: Use reference allele as major (requires -ref)
	5: Use ancestral allele as major (requires -anc)

Details

From counts of data

-doMajorMinor 2

If you input sequencing data like the bam format you can choose to infer the major and minor allele by picking the two most frequently observed bases across individuals. This is the approach from here: citation. To use this appraoch choose

From genotype likelihood data

-doMajorMinor 1

From input for either sequencing data like bam files or from genotype likelihood data like glfv3 the major and minor allele can be inferred directly from likelihoods. We use a maximum likelihood approach to choose the major and minor alleles. Details of the method can be found in the theory section of this page and for citation use this publication Skotte2012.

Forcing Major/minor

You can force the major and minor according to your reference or ancestral states if you have defined those -ref/-anc. We first estimate the major/minor from the data using -doMajorMinor 1/-doMajorMinor 2, and swap these accordingly with the major we are trying to force. If that is not the case the site will be discarded from downstream analysis.

Theory

From Genotype Likelihoods

Assuming that the considered site is diallelic, we infer those two alleles using the genotype likelihoods. Let $\{M,m\}$ denote the two possible alleles at the diallelic site, then the maximum likelihood estimate of this pair is found using the likelihood function

$P(D|\{m,M\})=\prod _{i}P(D_{i}|\{m,M\})=\prod _{i}\sum _{A_{1},A_{2}\in \{m,M\}}P(D_{i}|G=A_{1}A_{2})p(G=A_{1}A_{2}|\{m,M\}),$

where $P(D_{i}|G=A_{1}A_{2})$ is the genotype likelihood. We then assume that the two alleles within an individual are independent and randomly drawn from the set $\{m,M\}$ with equal probability, ignoring the fact that the two alleles at a diallelic site are not observed equally frequent. This gives us $p(G=A_{1}A_{2}|\{m,M\})=1/4$ for all four possible combinations of $A_{1},A_{2}\in \{m,M\}$. Therefore we estimate the two possible alleles at the diallelic site by

$argmax_{\{m,M\}}\prod _{i}\sum _{A_{1},A_{2}\in \{m,M\}}P(D_{i}|G=A_{1}A_{2})p(G=A_{1}A_{2}|\{m,M\}).$

To infer which of these two alleles is the minor allele, we estimate the allele frequencies (only one iteration of the EM algorithm is needed if the starting point is frequencies of 0.5).

This is the approach described in Skotte2012.