Hierarchical Clustering in R

1
Hierarchical Clustering in R
2
Quick R Tips

• How to find out what packages are available
• library()
• How to find out what packages are actually
  installed locally
• (.packages())

3
Hierarchical Clustering

• A type of cluster analysis
• There is both divisive and agglomerative
  HCagglomerative is most commonly used
• Group objects that are close to one another
  based on some distance/similarity metric
• Clusters are created and linked based on a metric
  that evaluates the cluster-to-cluster distance
• Results are displayed as a dendrogram

4
(No Transcript)
5
Step 1 Data matrix

• First you need a numeric matrix
• Typical array data set will have samples as
  columns and genes as rows
• We want to be sure our data are in the form of an
  expression matrix
• Use Biobase library/package
• See http//www.bioconductor.org/packages/2.2/bioc/
  vignettes/Biobase/inst/doc/ExpressionSetIntroducti
  on.pdf
• gt exprslt-as.matrix(data, headerTRUE, sep"\t",
  row.names1, as.isTRUE)

6
Step 2 Calculate Distance Matrix

• Default dist() method in R uses rows as the
  vectors..but we want the distance between
  samples.i.e., the columns of our matrix.
• There is a handy package to help us at MD
  Anderson called oompaBase
• source("http//bioinformatics.mdanderson.org/OOMPA
  /oompaLite.R")
• oompaLite()
• oompainstall(groupName"all")
• Once installed, be sure to locally activate the
  libraries
• library(oompaBase)
• library(ClassDiscovery)
• library(ClassComparison)
• oompaBase also requires the mclust and cobs
  packagesdownload these from CRAN

7

• Use the function distanceMatrix() to create a
  distance matrix of your samples.
• Uses the expression set created in Step 1 as
  input
• Remember that there are many different types of
  distance metrics to choose from!
• See help(distanceMatrix)
• xlt- distanceMatrix(exprs,'pearson')

8
Step 3 Cluster

• Use the hclust() function to create a
  hierarchical cluster based on your distance
  matrix, x, created in Step 2.
• gt ylt-hclust(x,method"complete")
• gt plot(y)

9
Testing for Differential Gene Expression with the
T-test
10

• Get the multtest package from CRAN
• Package contains data from the Golub leukemia
  microarray data set (ALL v AML)
• 38 arrays
• 27 from lymphoblastic
• 11 from myeloid

http//people.cryst.bbk.ac.uk/wernisch/macourse/
11

• library(multtest)
• data(golub)
• golub.cl
• Generate the T statistic
• teststat lt-mt.teststat(golub, golub.cl)
• Convert into P-values
• rawp0 lt-2pt(abs(teststat),lower.tailF, df38-2)
• Correct for multiple testing and show the ten
  most significant genes
• procs lt-c(Bonferroni, BH)
• reslt-mt.rawp2adjp((rawp0), procs)
• resadjp110,

http//people.cryst.bbk.ac.uk/wernisch/macourse/