Discovery of higherorder functional domains in the human genome

1
Discovery of higher-order functional domains in
the human genome

• Bob Thurman
• Noble Lab, Department of Genome Sciences
• University of Washington
• Seattle
• 11 October, 2006
• ASHG 2006, New Orleans

2
Idea
ENm005
1.7 Mb

• Idea of functional domains has been around for a
  long time.
• Now there are a variety of functional datasets
  available in nearly continuous fashion across
  genome. (ENCODE)
• Apply modern computational techniques to these
  datasets to delineate large-scale functionally
  active and inactive regions of the genome.

RNA
H3K27me3
H3ac
TR50
A
A
I
3
The tools

• Use wavelets to smooth disparate datasets out to
  a common scale.

1.6kb scale
6.4kb scale
4
The tools

• Use Hidden Markov Models (HMMs) to segment
  regions based on data.

5
The procedure
6
Results single-track segmentations
A
A
A
A
I
I
I
I
H3ac segmentation, ENm005 (1.7Mb)
7
Concordance of segmentations

• TR50 generally concordant with everything except
  RNA
• H3ac generally concordant with everything except
  H3K27me3
• H3K27me3 not very concordant with anything except
  TR50

8
Simultaneous four-track segmentation
A
A
I
I
Concordance with single-track
9
Enrichment of annotated genomic features
All
LTR
strict
loose
moderate
SINE Alus
LINEs (L1)
LINEs (L2)
non-exonic
EST overlap
CpG Islands
Simple repeats
mRNA Tx Starts
DNA transposons
Gencode Tx Starts
Spliced EST Tx Starts
Repeats
Conserved Elements (ENCODE MSA)
10
Conserved non-coding sequence

• Against expectations, active domains are somewhat
  depleted in CNS (18 depleted over random
  expectation)
• Does adding CNS track add anything? No even in
  terms of CN elements, there is little difference.
  In addition, there is very poor single-track
  concordance with other data types.

11
Cell-type differences

• Data for two different cell types (GM06990 and
  HeLaS3) is available for H3ac and RNA expression.
• Single-track segmentation concordance between
  cell types

12
Future work

• Scale up!
• More data types
• More cell types
• Model organisms whole genome functional data
  already available

13
Acknowledgments

• John Stamatoyannopoulos
• Bill Noble and the Noble lab
• Nathan Day
• Andrew Hemmaplardh
• HMMseg, a Java program for multi-variate HMM
  segmentation with optional wavelet smoothing, to
  be released soon

14
FIN
15
The data

• ENCODE project identify all functional elements
  of the genome.
• Pilot phase looks at 1(30Mb) of genome
  comprising the ENCODE regions.
• Functional datasets collected in common cell line
  HeLaS3.
• Histone modifications H3ac (Sanger) and H3K27me3
  (UCSD)
• RNA transcription levels (Affymetrix)
• DNA replication timing TR50 (University of
  Virginia)

16
4-state segmenation stats
17
Gene Ontology (GO) analysis

• Any classes of genes over-represented in
  active/inactive states?
• over-representation of genes involved in signal
  transduction (particularly olfactory
  G-protein-coupled receptors) within repressed
  domains

18
Enrichment of annotated genomic features
Affy RNA
H3K27me3
TR50
19
Enrichment of annotated genomic features
A
A
A
A
I
I
I
I
H3ac
20
High-confidence regions

• Intersection of 4 individual segmentations.
• 10.3Mb total (over 1/3 of ENCODE), and almost all
  of that (except for 1.5kb) is concordant with the
  4-track segmentation.

21
Outline

• Continuous genome-wide data and scale
• Tools for analysis
• Wavelets and HMMs
• Results
• Definition of functional domains

22
Continuous genome-wide data and scale

• A wide variety of nearly continuous genomic data
  is now available in 30Mb (1) of the human
  genome comprising the ENCODE regions.
• We focus on the issue of scale. Use wavelets to
• normalize datasets collected on widely disparate
  scales and
• elucidate trends in single and combined datasets
  at large scales.

23
Segment regions using wavelets and Hidden Markov
Models (HMMs)
Approximate scale of 60kb gives minimum segment
size of 10kb.