limmaGUIlimmaGUIA Point-and-Click Interface for cDNA A Point-and-Click Interface for cDNA

Microarray AnalysisMicroarray AnalysisJames Wettenhall and Gordon Smyth

Division of Genetics and Bioinformatics

Walter and Eliza Hall Institute of Medical Research

limma, limmaGUI and affylmGUIlimma, limmaGUI and affylmGUI

• limma : linear models for microarrays• by Gordon Smyth• Also contains many useful functions

specifically for cDNA microarrays

• limmaGUI : A Graphical User Interface for cDNA analysis with limma.

• affylmGUI : A Graphical User Interface for Affymetrix analysis with limma.

R, G, M and AR, G, M and A

• Rf = Red Foreground Intensity

• Rb = Red Background Intensity

• R = Rf - Rb

• Gf = Green Foreground Intensity

• Gb = Green Background Intensity

• G = Gf - Gb

R G Plot for ApoAI Slide 1R G Plot for ApoAI Slide 1

loglog22(R) log(R) log22(G) Plot for ApoAI Slide 1(G) Plot for ApoAI Slide 1

M and AM and A

• Log Ratio :

M (“Minus”) = log2(R/G)

= log2R – log2G

• Average Log Intensity :

A (“Add”) = log2(RG)1/2

= (1/2)(log2R + log2G)

M A Plot for ApoAI Slide 1M A Plot for ApoAI Slide 1

Normalized M A Plot for ApoAI Slide 1Normalized M A Plot for ApoAI Slide 1

M and A Have Nicer DistributionsM and A Have Nicer Distributions

Linear Models in MicroarraysLinear Models in Microarrays

Suppose for one gene, we have:

R G

Array 1 4 (KO) 32 (WT)

Array 2 15 (WT) 2 (KO)

M1 = log2(R1/G1) = log2(4/32) = -3

M2 = log2(R2/G2) = log2(15/2) = 2.9

Linear Models in MicroarraysLinear Models in Microarrays

• This linear model has one parameter, MKO-WT to be estimated for each gene.

• This parameter was estimated using a simple (weighted) average.

• A factor of (-1) was used for the dye-swap.

Linear Models in MicroarraysLinear Models in Microarrays

What about confidence statistics?

• As M1 is close to -M2 , we are confident in our estimate for MKO-WT so we expect:

• A low p-value• A high B statistic (log-odds of D.E.)• A large negative moderated t statistic

(because this gene is down-regulated).

Linear Models in MicroarraysLinear Models in Microarrays

• What makes this a LINEAR model?• Let E{} be the expected value of .• We have :

E{M1} = (1) MKO-WT

E{M2} = (-1) MKO-WT

• A linear relationship.

• The design matrix is :

limmalimma and and limmaGUIlimmaGUI

• http://bioinf.wehi.edu.au/limma• Documentation is available after installing

the package, by typing “help.start()” in R, clicking on “packages” and then clicking on “limma”.

• http://bioinf.wehi.edu.au/limmaGUI• Documentation is available online.• Example data sets are available online.

Swirl Zebrafish ExampleSwirl Zebrafish Example

• http://bioinf.wehi.edu.au/limmaGUI/Doc/Swirl/