Genetic analysis in structured populations used mixed linear models where the variance matrix of the error term is a linear combination of an identity matrix and a positive definite matrix.
The linear model is of the familiar form: 饾懄 = 饾憢 尾 + 系.
- 饾懄: phenotype
- 饾憢: covariates
- 尾: fixed effects
- 系: error term
Further, V(系) = 蟿虏饾惥+ 蟽虏饾惣, where 蟿虏 is the genetic variance, 蟽虏 is the environmental variance, 饾惥 is the kinship matrix, and 饾惣 is the identity matrix.
The key idea in speeding up computations here is that by rotating the phenotypes by the eigenvectors of 饾惥 we can transform estimation to a weighted least squares problem.
This code is under development.
Guide to the directories:
src
: Julia source codedata
: Example data for development and testingtest
: Code for testingdocs
: Notes on comparisons with other implementations