1 Estimating chromosomal copy number InCoB2007, Hong Kong, 30 August, 2007.

1

Estimating chromosomal copy number

InCoB2007, Hong Kong, 30 August, 2007

2

Copy number variation (CNV)What is it?

• A form of human genetic variation: instead of 2 copies of each region of each chromosome (diploid), some people have amplifications or losses (> 1kb) in different regions– this doesn’t include translocations or inversions

• We all have such regions – the publicly available genome NA15510 has between 5 & 240 by various estimates

– they are only rarely harmful (but rare things do happen)

3

Copy number variationPopulation genomics

The genomes of two humans differ more in a structural sense than at the nucleotide level; a recent paper estimates that on average two of us differ by

~ 4 - 24 Mb of genetic due to Copy Number Variation

~ 2.5 Mb due to Single Nucleotide Polymorphisms

4

Copy number variationAs it relates to human disease

Is responsible for a number of rare genetic conditions. For example, Down syndrome ( trisomy 21), Cri du chat syndrome (a partial deletion of 5p).

Is implicated in complex diseases. For example: CCL3L1 CN HIV/AIDS susceptibility; also, some sporadic (non-inherited) CN variants are strongly associated with autism, while

Tumors typically have a lot of chromosomal abnormalities, including recurrent CN changes.

5

Trisomy 21

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

6

Partial deletion of chr 5p

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

7

Large amplifications/losses can be seen by eye; smaller ones

are hard to see

8

A cytogeneticist’s story

“The story is about diagnosis of a 3 month old baby with macrocephaly and some heart problems. The doctors questioned a couple of syndromes which we tested for and found negative. Rather than continue this ‘shot in the dark’ approach, we put the case on an array and found a 2Mb deletion which notably deletes the gene NSD1 on chr 5, mutations in which are known to be cause Sotos syndrome. This is an overgrowth syndrome and fits with the macrocephaly.

The bottom line is that we are able to diagnose quicker by this approach and delineate exactly the underlying genetic change.”

9

2Mb deletion

Chromosome 5

10

NSD1

11A lung cancer cell line vs matched normal lymphoblast,from Nannya et al Cancer Res 2005;65:6071-6079

Many tumors have gross CN changes

12

Research into gonad dysfunction: Human sex reversal

• 20% of 46,XY females have mutations in SRY

• 80% of 46,XY females unexplained!

• 90% of 46,XX males due to translocation SRY

• 10% of 46,XX males unexplained!

Suggests loss of function and gain of function mutations in other genes may cause sex reversal. We’re looking at shared deletions.

13

Plan

To introduce the Single Nucleotide Polymorphism (SNP) arrays, the probes, and the associated assays. Then I’ll discuss the first bioinformatic aspect of Copy Number (CN) analysis, what I call low-level analyses, then show one way of assessing the outcome.

For simplicity I concentrate on Affymetrix arrays, called GeneChips, though similar considerations apply in whole or in part to some other array technologies, including Illumina.

14

Genomic DNA

ATCGGTAGCCATTCATGAGTTACTAPerfect Match probe for Allele A

ATCGGTAGCCATCCATGAGTTACTAPerfect Match probe for Allele B

A SNP

GTAGCCATCGGTA GTACTCAATGAT

Affymetrix SNP chip terminology

Genotyping: answering the question about the two copies of the chromosome on which the SNP is located:

Is a sample AA (AA) , AB (AG) or BB (GG) at this SNP?

Affymetrix GeneChip

1.28cm 6.4 million features/ chip

1.28cm

5 µ5 µ

5 µ

> 1 million identical 25 bp probes / feature

* **

***

16

250 ng Genomic DNA

RE Digestion

Adaptor Ligation

GeneChip® Mapping Assay Overview

Xba XbaXba

Fragmentationand Labeling

PCR: One Primer Amplification

Complexity Reduction

AA BB AB

Hyb & Wash

17

Principal low-level analysis steps

• Background adjustment and normalization at probe level These steps are to remove lab/operator/reagent effects

• Combining probe level summaries to probe set level summary: best done robustly, on many chips at once

This is to remove probe affinity effects and discordant observations (gross errors/non-responding probes, etc)

• Possibly further rounds of normalization (probe set level)

as lab/cohort/batch/other effects are frequently still visible

• Derive the relevant copy-number quantities Finally, quality assessment is an important low-level task.

18

AA

TTAT

Our preprocessing for total CN using SNP probe pairs (250K

chip)

Modification by H Bengtsson of a method due to A Wirapati developed some years ago for microsatellite genotyping; similar to the approach used by Illumina.

19

Background adjustment and normalization

Outcome similar to that achieved by quantile normalization

20

Low-level analysis problems haven’t been solved once and

for all; why?

• The feature size keeps and so the # features/chip keeps;

• Fewer and fewer features are used for a given measurement, allowing more measurements to be made using a single chip

These considerations all place more and more demands on the low-level analysis: to maintain the quality of existing measurements, and to obtain good new ones.

21

SNP probe tiling strategy

TAGCCATCGGTA N

SNP 0 position

A / G

GTACTCAATGAT*

ATCGGTAGCCAT T

ATCGGTAGCCAT CATCGGTAGCCAT G

ATCGGTAGCCAT ACATGAGTTACTACATGAGTTACTA

CATGAGTTACTA CATGAGTTACTA

PMMM

PMMM

AA

B B

0 Allele0 Allele

0 Allele0 Allele

Central probe quartet

22

SNP probe tiling strategy, 2

TAGCCATCGGTA N

SNP

+4 PositionA / G

GTA C TCAATGATCAGCT*

GTAGCCAT T

GTAGCCAT CGTAGCCAT C

GTAGCCAT TCAT G AGTTACTAGTCGCAT C AGTTACTAGTCG

CAT G AGTTACTAGTCGCAT C AGTTACTAGTCG

PMMM

PMMM

AA

B B

+4 Allele+4 Allele

+4 Allele+4 Allele

+4 offset probe quartet

SNP probe tiling strategy, 3

1 2 3 4 5 6 7

PMA PMA PMA PMA PMA PMA PMA

MMA MMA MMA MMA MMA MMA MMA

PMB PMB PMB PMB PMB PMB PMB

MMB MMB MMB MMB MMB MMB MMB

Central quartet

Offset quartets Offset quartets

This was repeated on the opposite strand giving 56 probes for the 10K chip.

The 100K chip had 40 chosen from offsets and strands by performance.

The 5.0 chip had 8 well chosen probes/SNP; no MMs.

The current 6.0 chip has just 6: 3 replicates of a PMA and 3 of a PMB. Also,

there are a large # of unreplicated non-polymorphic probes for CN inference.

24

What comes next?

• Using SNP chips to identify change in total copy number (i.e. CN ≠ 2)

• Outline a new method (CRMA)

• Evaluate and compare it with other methods

• Make some closing remarks on further issues

25

Copy-number estimation using Robust Multichip Analysis (CRMA)

CRMA

Preprocessing(probe signals)

allelic crosstalk (or

quantile)

Total CN PM=PMA+PMB

Summarization (SNP signals )

log-additivePM only

Post-processing fragment-length(GC-content)

Raw total CNs R = Reference

Mij = log2(ij/Rj) chip

i, probe j

A few details are passed over. Ask me later if you care about them.

26

CRMA, 1

CRMA


allelic crosstalk (or

quantile)

Total CN PM=PMA+PMB


log-additivePM only

Postprocessing fragment-length(GC-content)

Raw total CNs Mij = log2(ij/Rj)

Already briefly described.

27

CRMA, 2

CRMA


allelic crosstalk (quantile)

Total CN PM=PMA+PMB

Summarization (SNPsignals )

log-additivePM only

Postprocessing

fragment-length(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

That’s it!

28

CRMA, 3

CRMA



Total CNs PM=PMA+PMB


log-additivePM only

Postprocessing

fragment-length(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

log2(PMijk) = log2ij + log2jk + ijk

Fit using rlm

29

CRMA, 4a

CRMA



Total CN PM=PMA+PMB


log-additivePM-only

Postprocessing

fragment-length

(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

100K

Longer fragments get lesswell amplified by PCR and so give weaker SNP signals

30

CRMA, 4b

CRMA



Total CN PM=PMA+PMB


log-additivePM-only

Postprocessing

fragment-length

(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

500K


31

CRMA, 4c

CRMA



Total CN PM=PMA+PMB


log-additivePM-only

Postprocessing

fragment-length

(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

500K


32

CRMA, 5

CRMA



Total CN PM=PMA+PMB


log-additivePM-only

Postprocessing

fragment-length

(GC-content)

Raw total CNs Mij = log2(ij/Rj)

Care required with the number and nature of Reference samples used

33

Summary comparison of 4 methods

CRMA dChip(Li & Wong

2001)

CNAG*(Nannya et al 2005)

CNAT v4(Affymetrix

2006)

Preprocessing

(probe signals)


quantile scale quantile

Total CN PM=PMA+PMB PM=PMA+PMB

MM=MMA+MMB

PM=PMA+PMB “log-additive”PM-only

Summarization (SNP

signals )

Log additive

PM only

Multiplicative

PM-MM

=A+B

Post-processing

fragment-length

(GC-content)

fragment-length

(GC-content)

fragment-length

(GC-content)

Raw total CNs

Mij = log2(ij/Rj)

Mij = log2(ij/Rj)

Mij = log2(ij/Rj)

Mij = log2(ij/Rj)

34

Evaluation: how well can we differentiate

between one and two copies?HapMap:Mapping 250K Nsp data30 males and 29 females (no children; one bad data set)

Chromosome X is known: Males (CN=1) & females (CN=2)5,608 SNPs

Classification rule:Mij < threshold CNij =1, otherwise CNij =2.

Number of calls: 595,608 = 330,872

35

Calling samples for SNP_A-1920774

# males: 30# females: 29

Call rule:If Mi < threshold, a male

Calling a male male:#True positives: 30

Calling a female male:#False positives : 5

TP rate: 30/30 = 100%FP rate: 5/29 = 17%M = log2(/R)

36

Receiver Operator Characteristic (ROC)

increasingthreshold

FP rate

TP

rate

37

Single-SNP comparison: a random SNP

TP rate

FP rate

38

A non-differentiating SNP

39

Distribution (density) of TP rates when

controlling for FP rate (5,608 SNPs)

TP rate(correctly calling males male)

FP rate: 1.0% (incorrectly calling females male)

CNAT: 10% SNPs poor

density

40

CRMA & dChip perform better for an average SNP (common

threshold)Number of calls:595,608 = 330,872

zoom in

41

Average across R SNPsnon-overlapping windows

threshold

A false-positive(or real?!?)

42

Better detection rate when averaging

(with risk of missing short regions)

R=1(no av)

R=2

R=3

R=4

43

CRMA does a bit better than dChip

CRMA

dChipControl for FP rate: 1.0%

CRMA: R=1 69.6%R=2 96.0%R=3 98.7%R=4 99.8%…

44

Comparing methods by “resolution”

CRMAdChip

CNAG*

CNAT

All @ FP rate 1%

45

Several further bioinformatic issues

Estimating copy number: needs calibration data

Segmentation (of chromosomes into constant copy number regions): an HMM-like algorithm

Analysing family CN data: a different HMMIncorporating non-polymorphic probes: independent HMM observations to be weighted and combined

Dealing with mixed normal-abnormal samplesUtilizing poor quality DNA samplesEstimating allele-specific copy number ……and more

46

Some results using trios

Data: one of seven trios, 250K, results from Jeremy Silver

47

Conclusions/comments

• Using chromosome X permits us to:–test how well a method detects deletions–compare methods–get a sense of resolution

• We plan to do further tests with known CN changes to see how well this generalizes

• We are working on some of the issues other mentioned

There is room for contributions from you!

48

Available in aroma.affymetrix ("google it")

“Infinite” number of arrays: 1-1,000s

Requirements: 1-2GB RAMArrays: SNP, exon,

expression, (tiling).Dynamic HTML reportsImport/export to existing

methodsOpen source: RCross platform: Windows,

Linux, Mac

49

Acknowledgements

Henrik BengtssonHenrik Bengtsson, , UC BerkeleyUC Berkeley

Andrew Sinclair & Howard Slater,Andrew Sinclair & Howard Slater, MCRIMCRI

Nusrat RabbeeNusrat Rabbee, , GenentechGenentech

Simon Cawley, Francois Collin & Srinka Simon Cawley, Francois Collin & Srinka Ghosh,Ghosh, AffymetrixAffymetrix

Rafael Irizarry & Benilton CarvalhoRafael Irizarry & Benilton Carvalho, , Johns Johns HopkinsHopkins

Nancy ZhangNancy Zhang, , StanfordStanford

Jeremy SilverJeremy Silver, WEHI, WEHI

50

Thank you!

1 Estimating chromosomal copy number InCoB2007, Hong Kong, 30 August, 2007.

Documents