Statistical Methods for Biotechnology Products－Part II: Biochip Diagnostic Products Evaluation of Performance of Microarray Products

(1)

110/07/17 Copyright by Jen-pei Liu, PhD, Pro

duct names only for illustration 1

Statistical Methods for

Biotechnology Products

Part II: Biochip Diagnostic Products

Evaluation of Performance of

Microarray Products

by

Professor, Jen-pei Liu, PhD

Division of Biometry, Department of Agronomy

National Taiwan University, and

Division of Biostatistics and Bioinformatics

National Health Research Institutes

(2)

Outline



Introduction



Performance Evaluation of Affymetrix Gene

Chip



Performance Evaluation of Scanner



Performance Evaluation of GeneChip

Fluidics Station 450



Examples of Association between Expressi

on Levels and Clinical Outcomes



Performance Evaluation of Different Platfor

(3)

(4)

Introduction



Special instrument requirements for Roc

he AmpliChip CYP450 Microarray:



Affymetrix GeneChip Microarray Instrument

ation System (Platform)



GeneChip Fluidics Station 450DX



GeneChip Scanner 3000DX with Autoloader



Data station for GeneChip Operating Software a

nd AmpliChip CYP450 Data Analysis Software



GeneChip Operating Software (GCOS) Version 1.

(5)

Types of Microarray Platforms



Characteristics of AmpliChip CYP450 Microarray



15,129 probes with ~10

7

of the specific oligonucle

otide probe each in 11 m square position for a ~

2cm glass



Length of probe sequence: 16-22 bases



A single Probe Set consists of 4 Probes (or Feature

s) which have a fixed target except for at the subs

titution position wherean A, C, G, and T are includ

ed to generate four unique probes: one Perfect M

atch (PM) and three Mismatch (MM)



High-density oligonucleotide microarray probre-tili

ng approach to genotying

(6)

(7)

Types of Microarray Platforms



Characteristics of AmpliChip CYP450 Microarray



A Probe Set Pair consists of a Wild-type Probe Set

(for Wild-type allele) and a Mutant Probe Set (for

a known polymorphism).



To distinguish 29 polymorphisms in CYP2D6

including gene duplications and gene deletion



To identify 27 distinct alleles including 7 CYP2D6

gene duplication genes

(8)

Types of Microarray Platforms



25mer oligonucleotide probe sets printe

d by photolithography



Single spotted 30mer oligonucleotide



Single spotted 60mer oligonucleotide sy

nthesized in situ



Covalent attachment of prefabricated oli

gonucleotide

(9)

Introduction



Statistical Evaluations for



Each Instrument



a Complete System



Different Platforms



Between Laboratories



Issues Concerning Statistical Methods for

(10)

Performance Evaluation of Aff

ymetrix GeneChip





Technology



Semiconductor fabrication techniques



Solid phase chemistry



Random access combinatory chemistry



Molecular biology

(11)

Performance Evaluation of Aff

ymetrix GeneChip





Quality Controls for Manufacturing



Probe Design and Selection



Synthesis Process



Random Access Combinatory Chemistry



Signal Intensity

(12)

Performance Evaluation of Aff

ymetrix GeneChip





Manufacturing (Fabrication)



Photolithographic process



Light-sensitive chemical compound to

prevent coupling



Lithographic masks



5”x5” wafer surface flooded with solution

of A,T,C or G



Nucleotides also coupled with

(13)

duct names only for illustrationhttp://www.affymetrix.com/technology/manufacturing/index.aff13

x

Photolithographic fabrications

Performance Evaluation of Aff

(14)

(15)

Performance Evaluation of Aff

ymetrix GeneChip





Combinations of probes are

simultaneously synthesized through

repeated cycles of random access

combinatorial chemistry



High-density arrays over 10

7 probes



Each wafer is diced into tens, hundreds or

thousands of individual arrays



Sampling of arrays from every wafer by

running control hybridization using

standardized control probes

(16)

Performance Evaluation of Aff

ymetrix GeneChip





Random Access Combinational Chemistry



The number of compounds of length N, composed

of Y different subunits, is equal to Y

N



One can synthesize each of Y

N

compounds in Y

times N steps



25mer probes of A,C,G,T



Synthesis of 4

15

probes in 4 times 25 or 100 steps



Can examine the entire 3.1 billion bases in the

human in 100 steps or less



Easy implementation of in-process control for

(17)

Performance Evaluation of Aff

ymetrix GeneChip





Probe Design and Selection



25mer probes (why?)



22 probes for expression measurements



40 probes for genotype calls



Balance between sensitivity and specificity



Discrimination between signal and

(18)

(19)

Performance Evaluation of Aff

ymetrix GeneChip





Mismatch Probes



To detect and eliminate false or contaminat

ing fluirescence



Hybridization to nonspecific sequences as e

ffectively as perfect match



To serve internal control for perfect match



To quantify cross-hybridization

(20)

Performance Evaluation of Aff

ymetrix GeneChip





Design Verification



Design of photolithographic masks



Automatic software tests to verify the desig

n by array

in silico



Goal: the correct probe sequence are synth

esized in the correct (x,y) position in the ar

ray

(21)

Performance Evaluation of Aff

ymetrix GeneChip





Synthesis Verification



Manufacturing execution software system



Correct reagents



Correct mask



Each mask in a design set has a unique identification

number



Correct wafer



Correct timing

(22)

(23)

Performance Evaluation of Aff

ymetrix GeneChip





Combinatorial Chemistry Strategy



Goal: oligonucleotide probes on the array

are synthesized correctly



Control probes



Probe 1 is built in cycles 1,2,4, and 5



Probe 2 is built in cycles 3,5,7, and 8



Analyze probes 1 and 2 over all 8 cycles



No need to analyze all probes in an array

(24)

(25)

Performance Evaluation of Aff

ymetrix GeneChip





Signal Verification



Use of a second panel of control probes



Recombinase gene from P1 bacteriophage



cre



Genes in the biotin synthesis of E. coli

(26)

Performance Evaluation of Aff

ymetrix GeneChip





Analytical Sensitivity



Quantitative Testing: The change in

response of a measuring system or

instrument divided by the corresponding

change in the stimulus



Quality Testing: The test’s ability to obtain

positive results in concordance with

positive results obtained by the reference

method

(27)

Performance Evaluation of Aff

ymetrix GeneChip





Analytical Specificity



Quantitative Testing: the ability of an analytical m

ethod to determine only the component it purport

s to measure or the extent to which the assay res

ponds only to all subsets of a special analyte and

not to other substances present in the sample



Quality (Semi-quantitative, i.e., ordinal) Testing: T

he method’s ability to obtain the negative results i

n concordance with negative results by the refere

nce method

(28)

Performance Evaluation of Aff

ymetrix GeneChip





New Statistical Algorithms



Statistical significance for detection and

change calls



Confidence limits for log ratio values (fold

changes)



Fine tuning parameters to vary the

stringency of the analyses



Elimination of negative expression values

(29)

Performance Evaluation of Aff

ymetrix GeneChip





Single Array Analysis



Detection Algorithm



Detection p-value



Detection call



Signal Algorithm



Comparison Analysis



Change Algorithm



Robust Normalization



Change p-value



Change Call

(30)

Performance Evaluation of Aff

ymetrix GeneChip





Detection p-value against

user-definable cut-offs to determine the

Detection call



A Call indicates whether a transcript is

reliably detected (present) or not

detected (Absent)



A signal value is calculated which assigns

a relative measure of abundance to the

transcript

(31)

Performance Evaluation of Aff

ymetrix GeneChip





A transcript is represented as a probe set



A probe set is made up of probe pairs of Perfect Matc

h (PM) and Mismatch (MM)



Intensity of probe pair: key ingredients for expressio

n measurement



This measurement is calculated for each probe set in

the form of qualitative and quantitative values



Expression measurements of a baseline and experime

nt array can be compared to understand relative cha

nge in abundance of a transcrip

(32)

Performance Evaluation of Aff

ymetrix GeneChip





Detection p-value



Calculation of the Discriminant score for ea

ch probe pair



R = (PM-MM)/(PM+MM)



Test the Discriminant score against the use

(33)

(34)

(35)

Performance Evaluation of Aff

ymetrix GeneChip





PM intensity is 80



MM intensity increases from 10 to 100



Discriminant score decreases as MM intensity

increases



The ability to discriminate between PM and M

M decreases as MM increases



The dashline is the threshold  = 0.015



The one-sided Wilcoxon’s signed rank test to

(36)

Wilcoxon Signed-rank Test

Example:



A frozen foods manufacturer

A: own frozen product

B: competitors frozen product



Purchasing managers: 10-point Likert scale

(High score: better product)



Price, package design, variety, company pr

(37)

Wilcoxon Signed-rank Test

Likert Score

Manager

A

B

Difference(A-B)

1

6.5

4.0

2.5

2

4.0

8.0 -4.0

3

5.5

6.0 -1.0

4

5.5

10.0 -4.5

5

7.0

8.0 -1.0

6

4.0

9.0 -5.5

(38)

Wilcoxon Signed-rank Test

Likert Scale

Impression



The manager is asked to mark his/her respon

se on the scale



The mark defines an assumed measure of acc

eptability for the product in question

Vary poor Acceptable outstanding

(39)

Wilcoxon Signed-rank Test

Exact Methods for Small Sample(n ≤ 30)

1.

Calculate the difference( x

_A

- x

_B

)for each of the n pairs.

Differences equal to zero are eliminated, and the num

ber of pairs, n, is reduced accordingly.

2.

Rank the absolute values of the differences, assigning

1 to the smallest, 2 to the second smallest, and so on.

Tied observations are assigned the average of the ran

k that would have been assigned with no ties.

(40)

Wilcoxon Signed-rank Test

3.

Calculate the rank sum for the negative differences

and label this value T

-

. Similarly, calculate T

+

, the

rank sum for the positive differences.

4.

For a two-tailed test we use the smaller of these two

quantities, T, as test statistic to test the null

hypothesis that the two population relative

frequency histograms are identical.

(41)

Wilcoxon Signed-rank Test

Methods for Larger Samples (n>30)

1.

Null hypothesis: H

₀

: The population relative

frequency distributions for A and B are identical.

2.

Alternative hypothesis: H

_a

: The population relative

frequency distributions differ in location (a

two-tailed test). Or H

a

:The population relative frequency

distribution for A is shifted to the right (or left) of the

relative frequency distribution for population B (a

one-tailed test).

(42)

Wilcoxon Signed-rank Test

3.

Test statistic:

and T can be either T

+

or T

-



24 /

)]

1 n

2 )(

1 n

(

n

[

4 /

)

1 n

(

n

T

z







(43)

Wilcoxon Signed-rank Test

4.

Rejection region:

Reject H

0

if or for a two-tailed

test. For a one-tailed test, place all of

 in

one tail of the z distribution. To detect a shift

in the distribution of A observations to the

right of the distribution of B observations, let

T=T

+

and reject H

0

when . To detect a

shift in the opposite direction, let T=T

-

and

reject H

₀

if . Tabulated values of z

are given in Table 3 in the Appendix.

2 /

z



_

z





z

__/₂



 z

z





 z

z

(44)

Wilcoxon Signed-rank Test

Example(continued)

Likert score

Manager

A

B

Difference

(A-B)

Rank of

Abs. Diff

Signed

Rank(+)

Signed

Rank(-)

1

6.5

4.0

2.5

3

0

3

2

4.0

8.0 -4.0

4

0

3

5.5

6.0 -1.0

1.5

0

4

5.5

10.0 -4.5

5

0

5

7.0

8.0 -1.0

1.5

0

6

4.0

9.0 -5.5

6

0 Total

T

-

=18

T

+

=3

(45)

Wilcoxon Signed-rank Test

N=7<30  Exact Method

Two-tailed: T=min(T

-

, T

+

)=min(18, 3)=3

=0.05, T

0 =2  T=3 > T

0 =2

(46)

(47)

(48)

(49)

(50)

(51)

(52)

(53)

(54)

Performance Evaluation of Aff

ymetrix GeneChip





↑   the number of false present calls (↑sensitiv

ity) and   the number of true present calls ( sp

ecificity)



Before the detection call, the level of photomultipl

ier saturation for each probe is evaluated.



If all probed pairs in a probe set are saturated, the

probe set is immediately given a Present call.



A probe pair is rejected for further analysis when

(55)

Performance Evaluation of Aff

ymetrix GeneChip





Signal Algorithm



Signal is a quantitative metric calculated for each

probe set which represented the relative level of e

xpression of a transcript



Weighted mean by One-step Tukey’s Biweight Esti

mate



The vote: an estimate of the real signal due to hy

bridization of the target



The MM intensity is used to estimate stray signal



The real signal is estimated by taking the log of th

e PM intensity after subtracting the stray signal es

timate

(56)

Performance Evaluation of Aff

ymetrix GeneChip





Signal Algorithm



The probe pair signal close to the median

for a probe set is weighted more strongly



The quantitative metric Signal is the mean

of the weighted intensity values for a probe

set



MM > PM because of cross hybridization

and no additional information



Use of an imputed value = Change

(57)

Performance Evaluation of Aff

ymetrix GeneChip



(1) If MM < PM, MM is informative and is an

estimate of stray signal

(2) If MM is generally informative for a probe

set and only a few noninformative MM, use

adjusted MM for the noninformative MM

(3) If MM is general noninformative for a probe

set, this probe set is called Absent by the

(58)

(59)

Performance Evaluation of Aff

ymetrix GeneChip





Comparison Analysis (Experiment vs.

Baseline)



Change Algorithm



Robust Normalization



Change p-value



Change Call

(60)

Performance Evaluation of Aff

ymetrix GeneChip





Change Algorithm



Paired comparison: each probe set on the

experimental array is compared to its

counterpart on the baseline array



Change p-value: increase, decrease, or no

change in gene expression (quantitative)



Change calls: Increase, Marginal Increase,

No change, Decrease, or Marginal

Decrease

(61)

(62)

Performance Evaluation of Aff

ymetrix GeneChip





Normalization and scaling is applied to the

data from a selected user-defined group of

probe sets or to all probe sets



Normalization: intensities of the probe sets on

the experimental array are normalized to those

on the baseline array



Scaling, intensities of probe sets from both

experimental and baseline arrays are scaled to

a user-defined target intensity



Global scaling is recommended for comparing

(63)

Performance Evaluation of Aff

ymetrix GeneChip





Additional normalization is performed

by adjusting the normalization factor up

and down using a user-modified

parameter called perturbation



The range of perturbation is from 1 to

1.4 and the default value is 1.49



The higher value of perturbation, the

(64)

(65)

(66)

Performance Evaluation of Aff

ymetrix GeneChip





Calculation of Change p-values



Wilcoxon signed rank test



PM – MM of experimental array



PM – background of baseline array



Two-tailed p-value



P-value near 0  increase



P-value near 1  decrease

(67)

Performance Evaluation of Aff

ymetrix GeneChip





Change Call



The change p-value is categorized by t

wo cutoff values 1 and 2



1 and 2 are derived from iL and iH, i

=1,2



Default for 1: 1 = 1L = 1H = 0.0025

(68)

Performance Evaluation of Aff

ymetrix GeneChip





The final  cutoffs for a given probe set are

calculated using a linear interpolation

between the L and the H limits, based on the

probe set’s Signal position over the entire

array signal change



This fine tuning  cutoffs are used if Change

call accuracy is not optimal at one or both

extremes of Signal range of the array

(69)

(70)

(71)

(72)

(73)

(74)

Performance Evaluation of Aff

ymetrix GeneChip





The Signal Log Ratio estimates



Magnitude



Direction



Cancel out differences due to different

probe binding coefficients

(75)

Performance Evaluation of Aff

ymetrix GeneChip





A mean of the log ratios of probe pair in

tensities across the two array by a

one-step Tukey’s Biweight method



A 95% confidence interval is also provid

(76)

The biweight location estimate is defined as:

where

and

C=6(using 6 means that residuals up to approximately 4 are

included).

(77)

Performance Evaluation of Aff

ymetrix GeneChip





Comparison between MAS 5.0 and MAS 4.0



Selection of the components for the new algo

rithm



A “training set”



Each transcript group was spiked into a labele

d mixture of RNA from a tissue source



A 14x14 Latin square to monitor the detectabi

lity of transcripts over a range of concentratio

n

(78)

(79)

(80)

(81)

(82)

(83)

(84)

(85)

(86)

(87)

(88)

Performance Evaluation of Aff

ymetrix GeneChip





Validation of GeneChip



Human Genom

e U133 set



Key feature: The number of probe pair per

sequence reduces from 16 to 11



To increase sequence content per array



Validation as compared to HG-U95



Sensitivity



Concordance



Control genes

(89)

Performance Evaluation of Aff

ymetrix GeneChip





Sensitivity



A set of 50 human control clones as control

set



14x14 Latin square design for validation

experiment with 14 spiked concentrations

described previously



Detection Calls

(90)

Performance Evaluation of Aff

ymetrix GeneChip





Detection Call



Over 80% of the spikes are called present at a co

ncentration of 1.5 pM which is corresponding to a

pproximately one transcript in 100,000 or 3.5 copi

es per call



False positive rate is less than 10%



Comparison Call



2-fold change detected in 80% of spikes between

1.5 pM and 3.0 pM



4-fold change detected in 80% of spikes between

(91)

(92)

(93)

Performance Evaluation of Aff

ymetrix GeneChip





Concordance Call with HG-95



The percent of concordance calls (percent

agreement) is 80% using MAS 5.0



The percent of concordance calls (percent

agreement) dropping to 78% using MAS

5.0 for HG-133 and MAS 4.0 for HG-95



R

2 of signal log ratios between hG-95 and

HG-133 is only 0.54 for all calls and

(94)

(95)

McNemar 配對樣品檢定法

適合度檢定，獨立性檢定與同質性檢

定中，每一個觀測值均來自不同的個體

，觀測值是互相獨立。

選民在看電視辯論兩次投票行為結果

皆來自同一位選民，故此觀測值是具有

例：立委候選人電視辯論對選民投

票之影響

辯論後

辯論前

欲投甲

欲投乙

和

欲投甲

₄₉₁

₉

₅₀₀

欲投乙

₁

₄₉₉

₅₀₀

和

₄₉₂

₅₀₈

₁₀₀₀

(97)







電視辯論有影響

電視辯論無影響

:

0 a

H

0

1. .1

:

a

H P

P

H P

P







_



0

12

21

12

21 :

:

a

H P

P

H P

P







_



1.

11

12 .1

11

21 P

P









(98)

辯論後

辯論前

甲

乙

和

甲

_n

₁₁

_(p

₁₁

₎

_n

₁₂

_(p

₁₂

₎

_n

_1.

_(p

_1.

₎

乙

_n

₂₁

_(p

₂₁

₎

_n

₂₂

_(p

₂₂

₎

_n

_2.

_(p

_2.

₎

和

_n

_.1

_(p

_.1

₎

_n

_.2

_(p

_.2

₎

_{n (p)}

n

P

n

P

n

P

_ij



_ij

,

_i

_.



_i

_.

,

_.

_j



_.

_j

(99)



僅需考慮 n

₁₂

及 n

₂₁



令Ｓ＝ n

₁₂

+n

₂₁



若 H

₀

為真 n

₁₂

為二項 Bin(n

₁

,1/

2)



其期望數為：





(n

₁₂

+ n

₂₁

)/2

























0 2 1 , 2 21 12 2 21 11 21 12 2 21 12 21 21 12 2 21 12 11 2 2

,

2

2 H

n

E

O

i i i

拒絕

決策分法

































(100)

電視辯論









84 .

3

4 .

6

10

8

1

9

1

9

2

1 ,

05 .

0

2

21

12

2

21

12

2 





















n

拒絕 H

o

，電視辯論對選民投票行為有影響

(101)

Call Concordance Data









2

12

21

2

12

21

2 ₂

2 0.05,1

3616 2678

₉₃₈

3616 2678

6294

139.79

3.84 n

n





 















Reject Ho ， the proportion of present call by

HG-U133 is different from that of HG-U96

(102)

Continuity Correction





2

12

21

2

12

21

1 c

n



 



(103)

(104)

(105)

Performance Evaluation of Aff

ymetrix GeneChip





Tissue-Specific Expression Studies



Fifteen gene exhibiting tissue-specific expression in

one of four tissues – heart, fetal, pancreas and

placenta – were analyzed and confirming using

real-time RT-PCR



Change threshold values were plotted against log

signal data



Specific genes produce a high log signal value with

(106)

(107)

Performance Evaluation of Aff

ymetrix GeneChip





Normalization Control Probe Sets



a set of 100 normalization control genes



common characteristic of being called Present



over a wide range of expression levels



relatively low variability



Identify the corresponding probes via BLAST



100 normalization probe sets ID number 200000

(108)

Performance Evaluation of Aff

ymetrix GeneChip





Bacterial Control Probe Sets



Indicators of array quality, proper

hybridization and staining



18 11-probe pair bacterial probe sets vs.

(109)

(110)

(111)

Performance Evaluation of Aff

ymetrix GeneChip





Validation of Human Genome U133 Plus

2.0 and Human Genome U133A2.0



U133 Plus 2.0: 54,000 probe sets for

47,000 transcripts



U133A2.0: 22,000 probe sets for 18,400

transcripts

(112)

(113)

Performance Evaluation of Aff

ymetrix GeneChip





Spikes are detected at a concentration

of 0.75 pM, approximately one transcrip

t per 200,000.



At 0.75 pM, two-fold increases in conce

ntration were routinely detected



A labeling reagent based on biotinylated

(114)

(115)

(116)

(117)

(118)

(119)

(120)

(121)

(122)

(123)

(124)

(125)

(126)

Performance Evaluation of

Scanner



DNA Microarray Scanner



Independent of other system components



Nominally identical oligo microarrays



19,777 features of 60mer oligos



100 probes with 10 replicated



The same targets



All three microarrays were scanned 8 times

(127)

(128)

(129)

(130)

(131)

(132)

(133)

(134)

Performance Evaluation of

Scanner



Affymetrix GeneChip



Scanner 3000



Beta test plan: external testing at 6 customer sites

(academia, biotech, biopharam)



Alpha test plan: 144 internal site scans ( 2 probe arr

ay designs x 3 lots of each probe array design x 3 re

plicates for each lot x 4 spike targets x 2 scans



Evaluation Metrics



Present calls, false change between replicates, correlation

coefficients

(135)

(136)

(137)

(138)

(139)

(140)

(141)

(142)

(143)

Performance Evaluation of

Autoloader



Autoloader for Affymetrix GeneChip



Scanner

3000



Experiment 1: Impact of storage at 15

o

C on array

performance with the array’s performance



Experiment 2 and 3: Autoloader vs. Manual Scanni

ng



Performance metrics: Unscaled signal intensity, pr

esent call percentage, detection call accuracy, at a

concentration of 1.5 pM, false change rate, detecti

on call concordance

(144)

(145)

(146)

(147)

(148)

(149)

(150)

(151)

(152)

(153)

(154)

(155)

(156)

Performance Evaluation of

Fluidics Station 450



After loading the GeneChip arrays and al

l tubes, the FS 450 run unattended until

completion



Beta test plan



Pair of one FS 450 and one FS 400



Use of HG U133A from a single lot



53 spiked transcripts at 0, 0.75, 1.50, and 3

(157)

(158)

(159)

(160)

(161)

(162)

Limitations of Validation



Not independent evaluations by third party



Use of their own favorable



Designs



Target samples



With their own products



Performance metrics



Statistical methods



Correlation is for association not for agreement

(163)

Performance Evaluation



Standardization



Protocol



Design



Target samples



Reference samples



Performance criteria



Statistical methods

(164)

Michiels, Koscielny, and Hill

(2005, Lance

t)



To identify a subset of genes most

differentially expressed in patients with

different outcomes (a molecular signature) in

a training set



Estimate the proportion of misspecification in

an independent validation set of patients



To suggest a strategy by multiple random sets

to investigate the stability of molecular

signature and the proportion of

(165)

(166)

(167)

(168)

(169)

Two-gene expression ratio predicts

clinical outcome (Ma, et al, 2004)



Tamoxifen for breast cancer



A competitive inhibitor of estrogen binding to estrogen recep

tor (ER)



Reduction of 40%-50% in annual risk of recurrence



5.6% improvement in 10-year survival



ER and progesterone receptor (PR, an indicator of a function

al ER pathway) currently the best predicator of tamoxifen re

sponse



25% of ER+/PR+, 66% of ER+/PR-, and 55% of ER-/PR- fai

l to respond



To predict tamoxifen treatment outcome in early-stage brea

(170)

(171)

(172)

(173)

(174)

(175)

(176)

(177)

(178)

(179)

(180)

(181)

Evaluation of Commercial

Platforms (Tan et al, 2003)



Platforms



Affymetrix (U95Av2, GeneChips, 25mer olig

o probe sets)



Agilent (Human 1, cDNA probes)



Amersham (Codelink UniSet Human I Bioar

rays, 30mer Oligo probes)



A group of 2009 common genes present on

(182)

(183)

(184)

(185)

(186)

(187)

(188)

(189)

(190)

(191)

(192)

(193)

(194)

(195)

(196)

Standardization between Laboratories and

across Platforms

(MTRC, Nature methods, 2005)



Different platforms



Different protocols



Different computational and statistical

tools



Reproducibility



Standard array (intensity, ratio)



Resident arrays

(197)

(198)

(199)

(200)