The
Results
tab generated by this process depends on the compression method specified.
The
Results
tab generated when
Optimized
compression is specified is shown below:
This plot shows the output of the
Optimized
K matrix
compression algorithm. The
mixed model
for trait
association
is fit without the
SNP
marker in the
model
at varying levels of
K
matrix compression (performed by
hierarchical clustering
of the
K
matrix between all samples). The resolution of the intervals of
K
matrix compression/clustering can be controlled by options on the process
dialog
. At each level, the model is evaluated for performance with metrics such as AIC, AICC, and BIC (smaller values indicate a better fitting model). The plot on the
Results
tab displays these model performance criteria on the
y
-axis versus the levels of compression of the
K
matrix on the
x
-axis. The optimized compression method (as described by Zhang
et al
. Nature Genetics. 2010) chooses the level of compression at the minimum value of the chosen metric for the
Criterion for Optimal Compression Level
parameter.
Text above the plot details the final dimension of the
K
matrix that was chosen as optimal based on the mixed model for association testing. In the example settings, the
K
matrix was found to be optimal for the association mixed model at 178, indicating that the 193 individuals in the study could be clustered into 178 groups to reduce size of the random genetic effect in the model. This corresponds well to the data as there were known to be a handful of related samples that had a close family structure.
The
Results
tab generated when
Interactive
compression is specified is shown below:
When run using the
Interactive
compression method, the
Results
tab displays the full
K
matrix of relatedness between samples with a heat map and
dendrogram
. Based on the structure (shown best by patterns in the colored heat map), you can choose to cluster the samples into groups using diamond widget tool with the dendrogram on the right. Once you have set the desired level of clustering, click
Create Compressed K Matrix
to compute the aggregated estimate of relatedness and produce the compressed matrix for the samples in the groups defined by the clustering algorithm.