Shift caBIG

							     DWD for Micro-Array
Meta-Analysis (Data Combining)

     J. S. Marron & C. M. Perou

     Lineberger Cancer Center
     University of North Carolina

(& Department of Statistics and O. R.)
                   Goal
• Stat’ically Adjust for Systematic Biases
  – Removing Lab/Analyst Effects
  – Adjusting Cross-Platform Differences


• For merging micro-array data-bases
         Statistical Challenge
•   High Dimension – Low Sample Size data
•   Classical statistical methods fail
•   Pressing need for new methodologies
•   Wide open research area
            Conceptualization
•   Cases are long vectors
•   Gene expression levels are components
•   Genes are dimensions
•   Vectors are points in high dim’al space
•   Data set is “point cloud”
•   Bias Adjustment is “Match up point clouds”
Illustration of Conceptualization
Effective Bias Adjustment
 The Past of Bias Adjustment
• Principal Component Analysis
    i.e. Singular Value Decomposition
• Can be useful (e.g. above toy data)
• But fails too often
      (since finds dir’n of max’al var.)
• Key Hope: same as bias dir’n
• Weakness: dir’n ignores class info
 The Past of Bias Adjustment
• Principal Component Analysis (poor result)
• Only feels “variation in data”, not class labels
  The Present of Bias Adjustment
• Distance Weighted Discrimination (much better)
• HDLSS improvement of Support Vector Machines
  Two Improvements of DWD


1. DWD finds better dir’s for adjustment



2. Shift subpopulations, not collapse
DWD Adjustment for Batch Bias
DWD Combo of cDNA and Affymetrix Micro-Array Data Sets
    DWD, a look under the hood
•   An improved Support Vector Machine
•   Aimed at HDLSS situations
•   Avoids “data piling at margin”
•   Can lead to spurious direction
•   Non-shiftable populations
          Projection Plots
• Project data on adjustment direction

• Useful diagnostic

• Shows DWD finds better direction
  Projections more Gaussian (shiftable)
SVM Projection Plot
DWD Projection Plot
PCA 2-d Projections, Pre-Adj.
PCA 2-d Projections, Post-Adj.
   Future Research Problem
• Handling “non-center” differences
• Proposed Continued work, funding requested
           Future Directions
•   Adjusting for Variation about Center
•   Diagnostics (Develop – Deploy)
•   Better Tuning
•   Validation & Comparison
             References
• Benito, M., Parker, J., Du, Q., Wu, J.,
  Xiang, D., Perou, C. M., and Marron, J.
  S. (2004). Adjustment of systematic
  microarray data biases. Bioinformatics.
  Jan 1;20(1):105-14.

• Matlab software available:
  https://genome.unc.edu/pubsup/dwd/
        Requested Support
• DBA - Programming
      Data base work, and DWD integration


• Graduate student support
        Refinements & Next Generation