DAM: Discriminant Analysis Module

Initialization Dialog

Parameter Information


General DAM Terminology

The Primary function of the DAM is to serve as a method for multi-class classification. It incorporates the dimensional reduction method Multivariate Partial Least Squares (MPLS) and two classification analysis methods Polychotomous Discrimination (PDA) and Quadratic Discriminant Analysis (QDA). Either PDA or QDA are performed after the starting data has been reduced by MPLS.

Classification Selection

The classification selection option indicates whether to classify genes or experiments.

Parameters

Number Of Classes   The number of classes.

 

Number Of Gene Components   The number of gene components for Dimension Reduction.

 

Alpha   The alpha value chosen for the t-distribution in preliminary gene screening.

Skip Gene Screening step (ANOVA)   Skips the Gene Screening / Selection step when executing the Initial Classification and Assessment algorithms(A0, A1 or A2)

 
Assessment Algorithm Selection

The Initial Classification Algorithm and assessment algorithms A0, A1 & A2 essentially contain the same fundamental stages (i.e. Preliminary Gene Selection, Dimension Reduction(MPLS) & Classification/Prediction based on Leave One Out Cross Validation (LOOCV)) but they are ordered in a different sequence.

Initial Classification algorithm

  1. Preliminary Gene Selection: Use ANOVA to select genes from given gene expression matrix.
  2. Dimension Reduction: Use MPLS algorithm to obtain gene components matrix from gene expression matrix
  3. Classification:.
    Use the fitted classifier from training to predict all the test samples in the gene components matrix obtained from step 2.  

A0 algorithm

  1. Preliminary Gene Selection: Use ANOVA to select genes from given gene expression matrix.
  2. Dimension Reduction: Use MPLS algorithm to obtain gene components matrix from gene expression matrix
  3. Classification/Prediction: Classification is based on LOOCV.
    For each sample in the gene components matrix obtained from step 2, leave out this sample, and fit classifier to the remaining samples. Use the fitted classifier to predict left out sample.
     
    NOTE: For a given expression matrix, steps 1 (Preliminary Gene Selection) & 2 (Dimension Reduction) are fixed with respect to LOOCV. Thus, the effect of gene selection & dimension reduction on the classification cannot be assessed.

A1 algorithm

1.      Preliminary Gene Selection: Use ANOVA to select genes from given gene expression matrix.

 

For each sample in the gene expression matrix, leave out this sample to obtain a sub expression matrix.

2.      Dimension Reduction: Use MPLS algorithm to obtain gene components matrix from the sub gene expression matrix.

3.      Classification/Prediction: Fit classifier to the remaining samples in gene components matrix. Use the fitted classifier to predict left out sample.

 

NOTE: The first modification to A0 given as A1 assesses the affects of dimension reduction. The dimension reduction as well as the classifier is refitted N times, one for each sample left out.

A2 algorithm

For each sample in the gene expression matrix, leave out this sample to obtain a sub expression matrix.

  1. Preliminary Gene Selection: Use ANOVA to select genes from the sub expression matrix.
  2. Dimension Reduction: Use MPLS algorithm to obtain gene components matrix from the sub expression matrix.
  3. Classification/Prediction: Fit classifier to the remaining samples in gene components matrix. Use the fitted classifier to predict left out sample.

 

NOTE: The second modification to A0 given as A2 assesses the affects of gene selection. The gene set is reselected for classification each time a sample is left out.

Classification Method Selection

The Classification methods are performed after the data has reduced by dimension reduction (MPLS)

PDA

Polychotomous Discrimination

QDA

Quadratic Discriminant Analysis