cdna Microarray Analysis

Transcription

1 cdna Microarray Analysis with BioConductor packages Nolwenn Le Meur Copyright 2007

2 Outline Data acquisition Pre-processing Quality assessment Pre-processing background correction normalization summarization

3 Two-color Microarray Probe (gene reporter) Oligonucleotides or cdna clones Scan Target Test Reference Excitation Laser 2 Laser 1 RT-PCR Label with Fluor dyes PCR amplification and purification Cy5 Cy3 Emission Spotting Hybridize target to microarray Computer analysis (adapted from Duggan et al., Nat. Gen., 1999)

4 Image Analysis 2. Segmentation 3. Quantification Intensity CH2(635) 1. Location CH1(532) Intensity Raw data 60000

5 Terminology Target: DNA hybridized to the array, mobile substrate. Probe: DNA spotted on the array (spot). print-tip-group : collection of spots printed using the same print-tip (or pin), aka. grid. G, Gb: Cy3 signal and background intensities R, Rb: Cy5 signal and background intensities

6 BioC Task View: 129 packages

7 Import data... using limma package limmausersguide() library( limma ) targets <- readtargets( Targets.txt ) RG <- read.maimages(targets$filename, source= genepix )

8 ...into RGList > names(rg) R source G Rb printer Gb targets > RG$R[1,] s1 s2 s3 s4 [1, ] genes

9 Quality Assessment vs Control Quality Assessment: computation and interpretation of metrics that are intended to measure quality Quality Control: possible subsequent action, removing bad array or re-doing part of the experiment

10 Quality Assessment For each array: Diagnostics plots of spot statistics e.g. R and G log-intensities, M, A, spot area. Boxplots; 2D spatial images; Scatter-plots, e.g. MA-plots; Density plots. Stratify plots according to layout parameters, e.g. print-tip-group, plate. summary statistics

11 Quality Filtering Intensity (635) Intensity (635) Intensity (532) Intensity (532) Intensity (532) Background Foreground Intensity (635) Intensity (635) Intensity (532) 6000

12 Spatial Effects Image Plots R Rb R-Rb color scale by rank another array: print-tip color scale ~ log(g) color scale ~ rank(g)

13 Spatial Effects 1 pin 1 block

14 Spotting Pin Quality Decline after delivery of 5x105 spots after delivery of 3x105 spots H. Sueltmann DKFZ/MGA

15 MA plot M = log2(r) - log2(g) A = 1/2(log2(R) + log2(g))

16 Print-tip Effects ECDF plot (a42-u07639vene.txt) by spotting pin ^ F F(q) 0.6 1:1 1:2 1:3 1:4 2:1 2:2 2:3 2:4 3:1 3:2 3:3 3:4 4:1 4:2 4:3 4: q (log-ratio) -0.4 log(fg.green/fg.red)

17 PCR Plates - Boxplots

18 Bioconductor QA packages Several packages arraymagic arrayquality arrayqualitymetrics pdf or html

19 Diagnostic plot with arrayquality

20 Data Exploration with limma (Limma user Guide)

21 Quality Assessment: Summary For each array: Diagnostics plots Stratify BioC packages: arrayquality arraymagic

22 Outline Data acquisition Pre-processing Quality assessment Pre-processing background correction normalization summarization

23 Sources of Variation RNA extraction Systematic similar effect on many reverse transcription measurements labeling efficiencies corrections can be estimated from data Scanner settings PCR DNA concentration Printing or pin cross-hybridization Calibration Stochastic too random to be explicitely accounted for noise Error Model

24 Variance-Bias trade off <- Precision Variance -> <- Bias Accuray ->

25 Background Correction none subtraction, movingmin Minimun,edwards, normexp, More details limma >?backgroundcorrect

26 Background Correction none substraction normexp

27 Why Normalize? Theory Cy5 vs Cy3 Theory Reality Cy5 vs Cy3 Raw Data - Cy5 vs Cy Cy5 Cy5 Raw Cy Cy3 Cy5 Cy Cy Cy3 Raw

28 Normalization Identify and remove the effects of systematic variation Normalization is closely related to quality assessment. In a ideal experiment, no normalization would be necessary, as the technical variations would have been avoided. Normalization is needed to ensure that differences in intensities are indeed due to differential expression, and not some printing, hybridization, or scanning artifact. Normalization is necessary before any analysis which involves within or between slide comparisons of intensities, e.g., clustering, testing.

29 Data Transformation measured intensity = offset + Yik = Bik + gain true abundance ik Sk Example: log transformation Intensity measurements adapt a distribution that is closer to the normal distribution Muliplicative noise becomes additive noise: variance more independent of intensity

30 Normalization methods median loess 2D loess print-tip loess variance stabilisation Two-channel Separate-channel Smyth, G. K., and Speed, T. P. (2003). In: METHODS: Selecting Candidate Genes from DNA Array Screens: Application to Neuroscience

31 Two channel normalization Location: centers log-ratios around zero using A and spatial dependent bias

32 Two channel normalization Location: centers log-ratios around zero using A and spatial dependent bias Scale: adjust for different in scale between multiple arrays median centered median centered & MAD scaled Scaling

33 One channel normalization As technology improves the spot-to-spot varation is reduced Development of normalization techniques that work on the absolute intensities Ex: quantile normalization (limma) variance stabilization (vsn)

34 Quantile Normalization Before After Bolstand et al.(2003)

35 Variance Statibilizing Transformation log arsinh intensity Meaningful around 0 Original intensities may be negatives (Huber et al. 2004)

36 Variance stabilization (vsn) linear log arsinh

37 Variance stabilization (vsn) xi log xj log-ratio 'glog' (generalized log-ratio) log xi + xi2 + ci2 xj + xj2 + cj2 - interpretation as "fold change" + interpretation even in cases where genes are off in some conditions (negative values) + visualization + can use standard statistical methods (hypothesis testing, ANOVA, clustering, classification ) without the worries about low-level variability that are often warranted on the log-scale

38 Preprocessing : Summary For each array: Background correction or not Normalization: bias-variance trade-off Diagnostic plots BioC pacakges: marray limma

39