This folders contains a script (sparseREML_v0.6_toshare.R) ro run linear mixed model analyses with sparse genomic relationship matrices and an example script (Example.R) simulating data for siblings pairs and using the sparseREML() function to re-estimate parameters.

REQUIREMENTS

This code runs under any OS that can run the software package R. The main function depends on the R package Matrix, which handles sparse matrices. The simulation function also requires the R package MASS to simulate multivariate normal distribution.

RUNNING THE EXAMPLE

To run the example you can simply "source" the file as

R_prompt> source("Example.R")

The code should run in less than 30 seconds and print the results of the analysis. The simulation data consist of N=100,000 sibling pairs with their simulated relatedness (object G: summary(G)) ranging between 0.34 and 0.68 and their simulated phenotypes. Phenotypes are simulated to have an expected mean of 0 and an expected variance of 1. The heritability of the trait is hsq=0.6 and the variance explained by shared environmental effects is csq=0.1.

The result of analysis are in the R object model_ACE

R_prompt> print(model_ACE$h2_g)

The estimated heritability is 0.676 (s.e. 0.064) and the estimated shared environmental effect is 0.064 (0.032).

HOW TO RUN THIS FUNCTION ON YOUR DATA

The main function is sparseREML(). It requires 4 main parameters

• prefixGRM: this is the prefix of a sparse GRM generated by the software tool GCTA
• phenoFile: this is the path to phenotype file in PLINK format (i.e., columns are Family ID, Ind. ID, Phenotype)
• covarFile: this is the path to covariate file in PLINK format. If no covariate then set to NULL
• model: this takes values in AE (additive model), ACE (additive model with shared environmental effect, AME (additive model accounting for assortative mating) and AAE (additive and non-additive model).