USC: Uncorrelated Shrunken Centroids
Initialization Dialog
Enter all Class Labels
Enter the labels for each of the classes that are in your training set. If there are more than 2 classes,
click the Up Arrow of the '# of Classes' spinner. Be advised that you should be entering each and every
unique class label here. You will be asked later to assign these labels to your known training hybs.
Analysis Options
The USC algorithm works performing through two main processes - training and classification.
One can elect to perform training (with the built in option of classification afterwords) or classification only.
The 'Training & Classify' option results in specific 'Delta' and 'Rho' values along with 'Training Ratios'
which can stored as a file and used for classification at a later time.
The 'Classification Only' option takes a file input of ratios, delta and rho generated from training and
results in a classification of the elements.
Advanced Parameters
# Folds refers to the # of times one would like 'Random Permutations' to occur during cross validation.
For example, if there are 10 training hybs, and # Folds is set to 5, 2 hybs will be removed and used as
the 'test subset' during each Fold. After 5 folds, all 10 hybs will have been removed and used in the
'test subset' once.
#CV Runs is how many times one would like to run cross validation.
# Bins is the # of Delta values that one would like to try during cross validation. For example, if 'Max Delta'
is 20 and '# Bins' is 50, the Delta values would range from 0 - 20 by increments of 0.4 (20 / 50 = 0.4).
Delta = { 0, 0.4, 0.8, 1.2... 20 }.
Max Delta is the maximum Delta (Shrinkage) value to use during cross validation.
Corr Low and Corr High are the minimum and maximum values one would like to try as a Correlation Threshold value.
When the correlation between 2 genes is greater than this value, 1 of the genes will
be removed from analysis.
Corr Step is the increment by which one would like to step between 'Corr Low' and 'Corr High'. By default,
cross validation will occur using the set Correlation Threshold = { 0.5, 0.6, 0.7, 0.8, 0.9, 1.0 }.