Biostatistics Grand Rounds March 13, 2001

1
Biostatistics Grand RoundsMarch 13, 2001

• Discussion of Richard Simons talk
• Using DNA Microarrays to Improve Cancer
  Diagnostic Classification

2
Outline

• How is classification different?
• Is classification and/or clustering always
  appropriate?
• Whats the contribution of Statistics?
• Are current analyses tools always appropriate?

3
Classification

• http//linus.nci.nih.gov/brb/TechReport.htm
• NN classifiers and DLDA had the lowest error
  rates, whereas FLDA had the highest
• Dudoit, Fridlyand, and Speed (2001) JASA 97 77-87

4
(No Transcript)
5
Clustering/Classification always appropriate?

• Common problem
• Find differentially expressed genes in two
  populations
• A good design is important.

6
Dye effect?
7
Contribution of Statistics

• Not as simple as looking for red and green
  spots
• How big is big?
• Critical exploration of Data
• Garbage in, garbage out

8
Saturation problem

• Probes reaching maximum in experiment 1
• Scanner Chip PM MM Value
• 2 NV21a 354 25 46140
• 2 NV21b 564 57 46144
• 2 V21a 1004 83 46141
• 2 V21b 665 51 46139
• 1 NV100a 1917 328 46154
• 1 NV100b 1265 168 46160
• 1 V100a 3399 1085 46155
• 1 V100b 2267 446 46149

9
Scanner Effect
10
Remove Scanner Effect
11
The red-green ratios can be spatially biased
Top 2.5of ratios red, bottom 2.5 of ratios green
12
Another example

13
Looking for large FC
14
Looking for large FC