Parameters | Genetics | Multi-Marker Model Selection

Multi-Marker Model Selection
Check this box to do multi-marker model selection using an approach proposed by Rakitsch et al. (2013)1.
The first step in the selection uses SAS PROC MIXED to estimate the background genetic variance and the residual variance under the model with no marker effects.
The second step consists of selecting a multi-marker model based on rotated markers and phenotypic variables. PROC GLMSELECT is used to fit and select models with multiple markers in the second step. The following Model Selection Methods are available: Forward, Backward, Stepwise, LAR, LASSO, and ElasticNet.
A third step (not described in the reference paper) was also implemented. In this step, selected variables from step 2 are fit to the unrotated data for filtering out non-significant variables. This third step is important because variables selected on the rotated data might not be significant when fitted on the unrotated data set, especially if there are missing values in the unrotated data.
To Specify This Option:
*

1
Rakitsch, B., Lippert, C., Stegle, O., and Borgwardt, K. (2013). Lasso multi-marker mixed model for association mapping with population structure correction. Genetics 29(2): 206--214.