Case study: multi-trait mapping of 41 traits in yeast
The yeast dataset used to show the applicability for multi-trait mapping of a large number of traits was first published by Bloom and colleagues in 2013 Bloom et al, (2013) "Finding the sources of missing heritability in a yeast cross" Nature.
-
Phenotypes For analysis with LiMMBo, samples must be fully phenotyped. The missing phenotypes in this yeast dataset were imputed when feasible. Before imputation, missingness mechanism of data was examined and tested for missingness completely at random (MCAR) and missingness at random (MAR). The best predictor variables for each phenotype were chosen based on simulated missingness on the subset of fully phenotyped samples. Samples with high missingness rates and phenotypes that could not reliably be imputed were excluded from the study. All analysis were done in phenotypes_yeast.R, creating Figure S8-S10 (publication) and Figure 5.1-5.4 (thesis)
-
Genotypes Genotypes were filtered for samples that passed the phenotype pre-processing as described above and formated for plink (.ped/.map) via genotypes_yeast.R.
-
Relationship Genetic relationship between the samples in the test cohort was estimated in relationship_yeast.sh via plink's grm option. Different pruning set-ups were tested.
Genotype and phenotype data was formated for the linear mixed model software limix via format_data4limix_yeast.sh.
The variance decomposition of into genetic and noise variance components Cg and Cn was done via runLiMMBo, the association mapping in uni-variate or multi-variate mode providing Cg and Cn via gwas.py. Parameter set-ups for both function calls can be found in GWAS_yeast.sh.
The results of the association mapping are analysed and summarised in plots via GWAS_analysis.R, creating Figure 5 (publication) and Figure 5.5, 5.6 and B2 (thesis).