1. HWW Gene Expression Experiments: How? Why? What’s the problem?

2. High Throughput Experiments Bioinformatics Functional Genomics

3. DNA Hybridization The principle: have two denatured DNA strands bond together, then check double strand amount (florescent dye, radioactive label) “Traditional”: Southern/Northern/Western Blot The great advance: micro array DNA chips – automation, material eng., computer aided (including algorithmic solutions)

4. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, History cDNA microarrays have evolved from Southern blots, with clone libraries gridded out on nylon membrane filters being an important and still widely used intermediate. Things took off with the introduction of non-porous solid supports, such as glass - these permitted miniaturization - and fluorescence based detection. Currently, about 20,000 cDNAs can be spotted onto a microscope slide. The other, Affymetrix technology can produce arrays of 100,000 oligonucleotides on a silicon chip.

5. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, THE PROCESS Building the Chip: MASSIVE PCR PCR PURIFICATION and PREPARATION PREPARING SLIDESPRINTING Preparing RNA: CELL CULTURE AND HARVEST RNA ISOLATION cDNA PRODUCTION Hybing the Chip: POST PROCESSING ARRAY HYBRIDIZATION PROBE LABELING DATA ANALYSIS

6. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, MASSIVE PCR PCR PURIFICATION and PREPARATION PREPARING SLIDES PRINTING Building the Chip: Full yeast genome = 6,500 reactions IPA precipitation +EtOH washes + 384-well format The arrayer: high precision spotting device capable of printing 10,000 products in 14 hrs, with a plate change every 25 mins Polylysine coating for adhering PCR products to glass slides POST PROCESSING Chemically converting the positive polylysine surface to prevent non- specific hybridization

7. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Preparing RNA: CELL CULTURE AND HARVEST RNA ISOLATION cDNA PRODUCTION Designing experiments to profile conditions/perturbations/ mutations and carefully controlled growth conditions RNA yield and purity are determined by system. PolyA isolation is preferable but total RNA is useable. Two RNA samples are hybridized/chip. Single strand synthesis or amplification of RNA can be performed. cDNA production includes incorporation of Aminoallyl-dUTP.

8. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Hybing the Chip: ARRAY HYBRIDIZATION PROBE LABELING DATA ANALYSIS Cy3 and Cy5 RNA samples are simultaneously hybridized to chip. Hybs are performed for 5-12 hours and then chips are washed. Two RNA samples are labelled with Cy3 or Cy5 monofunctional dyes via a chemical coupling to AA-dUTP. Samples are purified using a PCR cleanup kit. Ratio measurements are determined via quantification of 532 nm and 635 nm emission values. Data are uploaded to the appropriate database where statistical and other analyses can then be performed.

9. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Printing Microarrays Print Head Plate Handling XYZ positioning –Repeatability & Accuracy –Resolution Environmental Control –Humidity –Dust Instrument Control Sample Tracking Software

10. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Ngai Lab arrayer, UC Berkeley

11. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Microarray Gridder

12. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Printing Approaches Non - Contact Piezoelectric dispenser Syringe-solenoid ink-jet dispenser Contact (using rigid pin tools, similar to filter array) Tweezer Split pin Micro spotting pin

13. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Micro Spotting pin

14. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research,

15. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Practical Problems —Surface chemistry: uneven surface may lead to high background. —Dipping the pin into large volume -> pre-printing to drain off excess sample. —Spot variation can be due to mechanical difference between pins. Pins could be clogged during the printing process. —Spot size and density depends on surface and solution properties. —Pins need good washing between samples to prevent sample carryover.

16. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Post Processing Arrays Protocol for Post Processing Microarrays Hydration/Heat Fixing 1. Pick out about 20-30 slides to be processed. 2. Determine the correct orientation of slide, and if necessary, etch label on lower left corner of array side 3. On back of slide, etch two lines above and below center of array to designate array area after processing 4. Pour 100 ml 1X SSC into hydration tray and warm on slide warmer at medium setting 5. Set slide array side down and observe spots until proper hydration is achieved. 6. Upon reaching proper hydration, immediately snap dry slide 7. Place slides in rack.

17. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Practical Problems 1 Comet Tails Likely caused by insufficiently rapid immersion of the slides in the succinic anhydride blocking solution.

18. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Practical Problems 2

19. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Practical Problems 3 High Background 2 likely causes: –Insufficient blocking. –Precipitation of the labeled probe. Weak Signals

20. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Practical Problems 4 Spot overlap: Likely cause: too much rehydration during post - processing.

21. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Practical Problems 5 Dust

22. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Steps in Images Processing 1. Addressing: locate centers 2. Segmentation: classification of pixels either as signal or background. using seeded region growing). 3. Information extraction: for each spot of the array, calculates signal intensity pairs, background and quality measures.

23. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Steps in Image Processing Spot Intensities –mean (pixel intensities). –median (pixel intensities). –Pixel variation ( IQR of log (pixel intensities ). Background values –Local –Morphological opening –Constant (global) –None Quality Information Signal Background 3. Information Extraction

24. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Addressing This is the process of assigning coordinates to each of the spots. Automating this part of the procedure permits high throughput analysis. 4 by 4 grids 19 by 21 spots per grid

25. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Addressing Registration

26. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Problems in automatic addressing Misregistration of the red and green channels Rotation of the array in the image Skew in the array Rotation

27. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Segmentation methods Fixed circles Adaptive Circle Adaptive Shape –Edge detection. –Seeded Region Growing. (R. Adams and L. Bishof (1994) :Regions grow outwards from the seed points preferentially according to the difference between a pixel’s value and the running mean of values in an adjoining region. Histogram Methods –Adaptive threshold.

28. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Examples of algorithms and software implementation

29. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Limitation of fixed circle method SRGFixed Circle

30. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Limitation of circular segmentation —Small spot —Not circular Results from SRG

31. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Information Extraction —Spot Intensities —mean (pixel intensities). —median (pixel intensities). —Background values —Local —Morphological opening —Constant (global) —None —Quality Information Take the average

32. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, Walter and Eliza Hall Institute of Medical Research Local Backgrounds

33. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Summary of analysis possibilities Determine genes which are differentially expressed (this task can take many forms depending on replication, etc) Connect differentially expressed genes to sequence databases and perhaps carry out further analyses, e.g. searching for common upstream motifs Overlay differentially expressed genes on pathway diagrams Relate expression levels to other information on cells, e.g. known tumour types Define subclasses (clusters) in sets of samples (e.g. tumours) Identify temporal or spatial trends in gene expression Seek roles for genes on the basis of patterns of co-expression ……..much more Many challenges: transcriptional regulation involves redundancy, feedback, amplification,.. non-linearity

34. Department of Statistics, University of California, Berkeley, and Division of Genetics and Bioinformatics, The Walter and Eliza Hall Institute of Medical Research, Biological Question Sample preparation Microarray Life Cycle Data Analysis & Modeling Microarray Reaction Microarray Detection Taken from Schena & Davis

35. Oligonucleotide Arrays

36. Schadt et al., Journal of Cellular Biochemistry, 2000

37. Oligonucleotide Arrays Tech. ~20 probes per “gene”, 25bases each* Probe size: 24x24 micron (contain ~10 6 copies of the probe) Probe is either a Perfect Match (PP) or a Miss Match (MM) MM: –usually at the center of the probe –Aim: to give estimate on the random hybrd.

38. Motivation Data is noisy, missing values. Each array is scanned separately, in different settings → To extract biological meaningful results we need: 1.Good expression estimations 2.Scale/Normalize across arrays

39. What we need Image segmentation Background/Gradient correction Artifact detection Allow array to array comparison (scale/normalize) Assess gene presence (quantitative “Measure”) Find differentially expressed genes

40. Why isn’t “Normalization” Easy? No ability to read mRNA level directly Various noise factors → hard to model exactly. Variable biological settings, experiment dependent. Need to differentiate between changes caused by biological signal from noise artifacts.

41. Variability Sources 1.Real Biology – 1.Biological noise 2.Biological Signal 2.Sample preparation related 3.Technical dependent

42. dChip MBEI Based on several papers by Li & Wong (PNAS, 2001 vol 98 no.1 and others) Implemented on their freely available dChip software Model based: The estimation is based on a model of how the probe intensity values respond to changes of the expression levels of the gene

43. dChip Model i is the array index j is the probe index is the baseline response of the probe due to non specific hybridization is the additional rate of increase of the PM response is the rate of increase of the MM response

44. dChip “Reduced” Model Basic idea: Least square parameter estimation, iteratively fitting and

45. dChip “Reduced” Model For one array, assume that the set has been learned from a large number of arrays, and therefore known and fixed Given this set, the linear least square estimate for theta is An approx. Std. can be computed for this estimator:

46. dChip “Reduced” Model Similarly, we regard the set as known, and compute std. for each phi We use these estimated Std. to find outlier and exclude them from the computation:

47. Dchip – Array outliers detection

48. Dchip – Probe outliers detection

49. Normalization/Scaling We saw how to get MBEI from dchip, i.e measure “quantitation “ We still need to scale the different arrays: –Arrays usually differ in overall image brightness (differ in time, place, exper. Cond….) This is usually done PRIOR to the “measure quantitation” manipulations (as dChip’s MBEI we just described).

50. Global – Normalization/Scaling Suppose we have two arrays X,Y with values x 1 …x M and y 1.. y M “Global” normalization (MAS 5): find the constant “a” such that Which means: When we have multiple arrays then we choose Y to be the avg. of all arrays or compute a such that sum_i (x_i) = constant Better way: a(x) i.e adopt the fit parameter as a function of expression level ( as by dChip)

51. dChip – Normalization/Scaling Big question: Which gene to use for this scaling?? There are various ways to choose the set: –“House keeping” genes (Affy. chips) –Spiked controls added in various stages of the experiment, in a range of concentrations –Both of the above are very good in theory but (still) not in practice (esp. in Affy chips) –The result: several approaches suggested on how to use the set of genes tested in the experiments We’ll review dChip’s solution: The “Invariant set”

52. dChip “Invariant Set” Main idea: 1.Initialize: set of probes P = all probes 2.Order the probes in both arrays by their expression values 3.Give each probe in each array an index according to it’s relative expression order 4.Find a set of probes P’ who’s relative order is similar in both arrays 5.Set P = P’ and iterate from stage (2) until convergence 6.Use the resulting P to compute a piecewise linear running median line as the normalization curve

55. Normalization Tools – Current State Commonly Used: –RMA by Speed Lab –dChip by Li & Wong –GeneChip = MAS5 (Affy. built in tool) “The Future”: –New Chip design (both Affy. And cDNA) with better probes, better built in controls etc. –New algorithms – facilitating probes GC content (gcRMA), location etc. –New MAS tool (this year ?) is also supposed to incorporate RMA,dChip etc.

56. How to Measure Performance? 1.Theoretical Validation – use some theoretical assumptions and evaluate Statistical characteristics of the method at hand. 2.Experimental Validation – 1.Use public data sets to measure different aspects of performance 2.Evaluate relevant characteristics on your data set. Design your data set accordingly (if possible)

57. A Benchmark for Affy. Expression Measures* Main Idea: Define a “universal” test set & test statistics Based on 3 publicly available spike in data sets Tests for: –Variability across replicate arrays –Response of GE measures to change in abundance of RNA –Sensitivity of fold change measures to amount of actual RNA sample –Accuracy of fold change as a measure of relative expression –Usefulness of raw fold change score to detect differential expressed genes * Cope et al. Bioinformatics, 03 (Speed’s Lab)

58. MA Plot M 1 = X 1 – X 2 A = (X 1 + X 2 )/ 2 Where X i is the log 2 of expression measure

59. Variance across replicates plot Test Statistics: 1. Median std. 2. Avg. R 2 (squared corr. coef.) between two replicates

60. Observed Expression vs. Nominal Expression Plots Test Statistics: Fit a linear curve and compute 1. linear fit slope (should be 1) 2. R 2 to the linear fit

61. ROC Curves One of the chief uses of GE arrays is to identify differentially expressed genes ROC ( Receiver Operator Characteristic): A graphical representation of both Sens. and Spec. as a function of threshold value X axis: TPR (Sens.) Y axis: FPR (1-Spec.) In this case: Use fold change as the score, knowing which probes are spiked or not..

62. FC ROC Plots Here actual TP, FP numbers are used for the axes Test Statistic: AUC (area under the graph)

63. FC ROC Plots Same as before, but only for FC = 2 cases (harder)

64. The Benchmark – Bottom Line 15 parameters used to test performace 3 “synthetic” spike in data sets Automatic submission and evaluation tool + comparative results at: www.biostat.jhsph.edu www.biostat.jhsph.edu

65. Other Tests Evaluate separately normalization and expression measures techniques ( as by Huffman et al., Genome Biology, Vol. 3, 2002) How do we evaluate performance on our own, very specific, data??? ( hint: see next class..)