1. GeneChips and Microarray Expression Data
David Paoletti

2. The Problem Determine gene expression (activity)
What proteins are being produced by a group of cells?

3. The Assumption
The RNA present in the cell determines what proteins are being produced
Efficiency

4. The Why Understanding Toxicology Drug design Evaluation Specificity
Response

5. What is a GeneChip? 1.28 x 1.28 cm glass wafer  500,000 features
24 x 24 m probe site
25 mer oligo, complementary
PM: perfect match
MM: mismatch
 2.5 M copies
GeneChip

6. The Solution

7. The Gains
Speed
Possibility
Sensitivity
Reproducibility

9. The Process Poly-A RNA Biotin-labeled Antisense cRNA Cells cDNA IVT L
AAAA
Fragment (heat, Mg2+)
Labeled
fragments
Hybridize
Wash/stain
Scan L L L

10. Hybridization and Staining
GeneChip
Biotin
Labeled cRNA
Hybridized Array L L L L + L L L + L L L L
SAPE
Streptavidin-
phycoerythrin L

11. Specialized Equipment

13. How Features Are Chosen5’
Gene Sequence 3’
Multiple
oligo probes
25 mers
Perfect Match
Mismatch

14. Feature Values Remove outermost rows and columns83
112 96 32 47
382
165 87 55
246
140 93
104
552
187 65
Remove outermost rows and columns
Find 75th percentile of remaining values
This value is taken as representative of this feature

15. Background Noise Removal
The array is divided into 16 equal sectors
For each sector
Find the lowest 2% of the feature intensities
Average these
Subtract this average from the intensity value of all features in the sector

16. Noise Calculation

17. Average Difference Intensity
For a given gene
For each probe pair for the given gene
Calculate the difference PM-MM
Calculate ,  for this set
If abs( (PM – MM) - )  3, delete from set
Remaining set is pairs in avg

18. Positive & Negative Probe Pairs
PM-MM  SDT
PM/MM  SRT
MM-PM  SDT
MM/PM  SRT
If both true, mark as positive
If both true, mark as negative
SDT = Q · STDmult
By default, SRT = 1.5, STDmult = 2.0 (low density), 4.0 (high)

19. Voting Methods for Absolute Call
Positive/negative ratio
PNR = #pos / #neg
Positive fraction
PF = #pos / #used
Log average ratio

20. Decision Matrix Absent Marginal Present PNR 3.00 4.00 PF 0.33 0.43 LA
0.90
1.30

21. Average Difference and Absolute Call
Which of these do you base a decision on, for whether a gene is being expressed?
Use the absolute call for decision
Use average difference to compare those which are present

22. Conclusions
Incredible amalgam of biological and computational processes
Allows analyses that would not be performed otherwise
Already of proven worth

23. References
Moore, S K; Making chips to probe genes, IEEE Spectrum, March 2001,
GeneChip Gene Expression Algorithm Training, Part I: Absolute Analysis; Affymetrix.
Berberich, S, and McGorry, M; GeneChip protocols; Wright State University.