BiologyDriven Clustering of Microarray Data

1
Biology-Driven Clustering of Microarray Data

• Applications to the NCI60 Data Set

K.R. Coombes, K.A. Baggerly, D.N. Stivers, J.
Wang, D. Gold, H.G. Sung, and S.J. Lee
2
Introduction

• Microarray data is more than a large,
  unstructured matrix.
• We already know many genes important for studying
  cancer through their involvement in specific
  biological processes
• We also know that reproducible chromosomal
  abnormalities play an important role in cancer
• Need analytical methods that use biological
  information early

3
Methods

• First, updated the annotations of the genes on
  the microarray
• Performed separate analyses
• using genes on individual chromosomes
• using genes involved in different biological
  processes
• Developed ways to assess how well each set of
  genes classified samples

4
Quality of Annotations

• Problem
• I.M.A.G.E. clone IDs and GenBank accession
  numbers are archival
• UniGene clusters, gene names, descriptions,
  functions, etc., are changeable
• Solution
• Download latest UniGene (build 137) and LocusLink
  to update annotations

5
How many genes on the array have good annotations?
Only trust the 7478 spots where the UniGene
clusters match.
6
Where are the genes located?
7
How do we determine the functions of genes?

• UniGene -gt LocusLink -gt GeneOntology
• GeneOntology is a structured, hierarchical
  vocabulary to describe gene functions in three
  broad areas
• biological process (why)
• molecular function (what)
• cellular component (where)

8
What kinds of genes are on the microarray?
9
Data Preprocessing

• Remove spots with poor annotations and spots with
  median intensity below the 97th percentile of
  empty spots.
• Normalize each array so median log ratio between
  channels is one
• Center each gene so mean log ratio across
  experiments is zero
• Use (1-correlation)/2 as distance metric

10
How well does a set of genes distinguish types of
cancer?

• Three methods for assessment
• Qualitative (PCA, MDS)
• Quantitative (PCA ANOVA)
• Semi-quantitative (Grading Dendrograms)

11
Multidimensional Scaling
12
PCANOVA
13
How good is a dendrogram?

• A cluster contains all and only one kind of
  cancer
• B all, with extras
• C all except one
• D all except one, with extras
• E all except two
• F all except two, with extras

14
Can cancers be distinguished by genes on one
chromosome?
15
Heterogeneity of different types of cancer

• Some cancers (colon, leukemia) are fairly easy to
  distinguish from others
• Some (breast, lung) are so heterogeneous as to be
  almost impossible to distinguish
• Some chromosomes (1, 2, 6, 7, 9, 12, 17) can
  distinguish many cancers.
• Some (16, 21) are essentially random

16
(No Transcript)
17
(No Transcript)
18
Can cancers be distinguished by genes of one
function?

• Table for functional categories looks a lot like
  the table for chromosomes
• Some biological process categories (signal
  transduction, cell proliferation, cell cycle,
  protein metabolism) can distinguish many types of
  cancer
• Others (apoptosis, energy pathways) cannot

19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
Conclusions (I)

• Multiple views into the data provide substantial
  insight into differences in cancer types and gene
  sets.
• Cancer types differ greatly in their degree of
  heterogeneity, ranging from homogeneous (colon,
  leukemia) through moderately heterogeneous
  (renal, melanoma) to extremely heterogeneous
  (breast and lung).

24
Conclusions (II)

• Homogeneous cancers exhibit strong identifying
  signals across most views of the data.
• There are large difference in the ability of
  genes of different chromosomes or involved in
  different biological processes to distinguish
  cancer types.

25
Supplementary Material

• Complete results of each analysis by chromosome
  and by function are available no our web site
• http//www.mdanderson.org
• /depts/cancergenomics