1. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 A NOVEL APPROACH TO IMPROVE THE NOISE IN DETECTING COPY NUMBER VARIATIONS USING OLIGONUCLEOTIDE MICROARRAYS 12 November 2008 Noushin Farnoud, Marco Marra, Jan Friedman, Stephane Flibotte, Allen Delaney Canada’s Michael Smith Genome Sciences Centre

2. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Outline What are Copy Number Variations (CNVs)? Why is it important to study copy number variations? How can we study CNVs? What are the issues associated with studying CNVs? How can we deal with them?

3. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 What is Copy Number Variation (CNV)? The DNA copy number of a region of a genome is the number of copies of genomic DNA. In humans the normal copy number is two for majority of autosomes. However, recent discoveries have revealed that many segments of DNA, ranging in size from kilobases to megabases, can vary in copy- number. These DNA copy number variations (CNVs) are a result of genomic events causing discrete gains and losses in contiguous segments of the genome.

4. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Why is it important to study CNVs? CNVs are common in cancer and other diseases. For example, a review paper by Charles Lee have listed 17 conditions of the nervous system alone – including Parkinson’s Disease and Alzheimer’s Disease – that can result from copy number variation (Neuron Oct 06) CNVs are also common in normal individual and contribute to our uniqueness. These changes can also influence the susceptibility to disease. Since CNVs often encompass genes, they can have important roles both in characterizing human disease and discovering drug response targets. Understanding the mechanisms of CNV formation may also help us better understand human genome evolution.

5. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 How can we detect CNVs? Two-color arrays One-color arrays Patient Reference

6. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Main issue of oligonucleotide microarrays Log2 Ratio of Intensity 22 4 -4 -2 0 Position (Mb) 0 50 100 150 200250 * http://dsgweb.wustl.edu/qunyuan Although high density microarrays provide genome wide data on copy number, they are often associated with substantial amount of noise that could affect the performance of the analyses.

7. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 How can we improve this noise? Can we improve the oligonucleotide microarray noise by analyzing individual oligonucleotide probes? Hypothesis Each SNP probe set has : # oligonucleotide probes (10K array): 647,080 oligos # oligonucleotide probes (100K array) : 4,648,160 oligos # oligonucleotide probes (500K array): 12,013,632 oligos

8. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008

9. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Therefore… We can conclude that a major source of the noise is the different behavior of the individual oligonucleotide probes in the SNP probe-set. This points out to the fact that averaging all PM oligos is not a proper approximation of information content of a SNP.

10. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Novel Algorithm: Oligonucleotide Probe-level Analysis of Signal intensities (OPAS) Clusters the individual oligos in each SNP probe-set Apply Null-hypothesis testing : estimates the likelihood (p-value) that each cluster of oligos have log-ratio- intensity =0; >0 or <0 Based on these p-values and ML classification algorithms; identify the “most significant cluster of oligos”. The other cluster(s) of oligos is noise; exclude them from analysis.

11. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Example of Improving the SNP Noise by OPAS Before After

12. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 How does OPAS Affect CN analysis?

13. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 What's next? The next-generation of DNA microarray-based technologies will allow equal detection of large and small CNVs. Also on the horizon are new DNA sequencing technologies enabling rapid (and ultimately inexpensive) 'personalized' genome sequencing projects. Coupled together, these technologies will capture almost all the variation in a genome.

14. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Acknowledgments Funding Contact: nfarnoud@bcgsc.ca GSC Marco Marra Stephane Flibotte Allen Delaney Irene Li Hong Qian Robert Holt Sussana Chan BC Children’s & Women’s Hospital Jan Friedman Patrice Eydoux

15. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008

16. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Advantages of Array CGH

17. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Log2 Ratio Classification (each SNP is classified to be deleted, normal or amplified, based on comparing the P’s of its consisting clusters of PM oligos Likelihood Estimation Apply a series of Null-Hypothesis Tests, to determine the likelihood : P Hs (cluster = 0) P Hs (cluster< -0.5) P Hs (cluster> +0.6 Clustering PM oligos (using Fuzzy Clustering approach) Post Processing the Results Test Array (Normalized Log2 Raw- Intensity) Ref Set (Pool of Normal Parents)

18. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 What is Copy Number Introduction - What is a SNP? - What is a SNP array? Array Design + Target Preparation Applications of SNP arrays (other than genotyping) - Copy number analysis Genotyping using SNP arrays - Generations of methodologies - Properties of SNP arrays

19. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Schematic Representation of DNA Copy Number Change Normal cell deletion amplification CN=0 CN=1 CN=3 CN=4 CN=2

20. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 T T T T T T A A A A A C C CG G G G A T T T T T T A A A A A C C CG G G G A CG GCTA Single Nucleotide- Polymorphism (SNP) Background (1) : What are SNPs? Definition: SNPs are variations in single base pairs that are randomly dispersed throughout the genome

21. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Major conclusions so far* … There is a considerable variation among the numbers and types of candidate CNVs detected by different analysis approaches. Multiple programs are needed to find all real aberrations in a test set. The frequency of false positive deletions is substantial, but can be greatly reduced by using the SNP genotype information to confirm loss of heterozygosity. * Friedman et. al, AJHG 2006 Baross et. al, BMC Bioinformatics, 2007 Delaney et. al, in progress

22. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Profile of SNP probe sets Deleted SNPs SNPs in ‘Normal’ Region

23. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 3 Generation of Affymetrix SNP arrays 10K100K500K Number of SNPs 11,555116,204500,568 Number of Oligonucleotide Probes 647,080 (14 quartets / SNP) 4,648,160 (10 quartets / SNP) 12,013,632 (6 quartets / SNP) Number of Arrays 1 Xba I 2 Xba I + Hind III 2 Nsp I + Sty I Number of SNPs per Array -- ~58,960 : Xba ~57,244 : Hind ~262,000 : Nsp ~238,000 : Sty Median inter-marker distance (kb) 1058.52.5 Mean inter-marker distance (kb) 21023.65.8 Average heterozygosity 0.370.3 % genome within 10kb of a SNP --40%85%

24. ____ __ __ _______Birol et al :: AGBT :: 7 February 2008 Background : Structure of Affy SNP array Each SNP probe set has : 57 oligonucleotide probes (10K array): 647,080 oligos 40 oligonucleotide probes (100K array) : 4,648,160 oligos 20 oligonucleotide probes (500K array): 12,013,632 oligos