Delete irrelevant variables
This function was designed to remove less relevant variables with regard to a target, or dependent, trait or variable. It involves the following steps:
- Under Multivariate, click Find Associated Testers.
- An input box will appear, asking for the probability level for a trait to be considered as 'associated' with the target trait. Put a number between 0 and 0.5, and click OK.
- A panel will appear above the biplot.
- Click the drop-down arrow by the textbox, select the target trait, and then click Find.
- An input box will appear, asking you if you want to use any traits/markers as covariates.
- Click No. (If you click Yes, then you are asking for a Covariate Analysis)
A new biplot will appear that has the following changes:
- traits/markers that were chosen as covariate and traits/markers that do not meet the association criteria are deleted from the biplot;
- the target trait is indicated by a larger font and is framed;
- only trait/marker labels are shown to focus on the relationships among the traits/markers (the entries can be called back by clicking Full Names or Show Both under View);
- the vectors of the trait/marker (lines from biplot origin to the trait labels) are drawn to facilitate visualization of the relationships among them.
Interpretations when the target variable is a trait and the other variables are genetic markers:
- Markers that have longer vectors and smaller angles (close to 0° or 180°) with the target trait are more closely associated with the target trait. An acute angle means positive correlation and an obtuse angle means negative correlation.
- Markers that have short vectors or that have angles close to 90° with the target trait are more loosely associated with the target trait. However, isolated traits or markers may have stronger associations with the target trait than they the angles suggested.
- Clustered markers may be closely linked and therefore suggest a single QTL.
- An isolated marker may suggest a QTL.
Here is a real example of QTL identification based on a GGE biplot.