Transcript Assembly, Comparison, and Annotation of Mammalian Genomes

Assembly, Comparison, and Annotation
of Mammalian Genomes
David Haussler
Howard Hughes Medical Institute
University of California, Santa Cruz
Bioinformatics of mammalian genomes
•Sequence Assembly
•Genome Browsers:
new computational
microscopes
•Computing Evolution’s Path:
key to understanding function
Assembling the human genome
• GigAssembler (Kent)
– Built first draft of the human genome from lowerlevel contigs produced by Phrap( P. Green)
Outgoing UCSC internet traffic (green) for year 2000. Main peak is
activity on July 7, 2000 when human sequence was first posted on the web
• Celera Assembler (Myers/Sutton)
– First mammalian whole genome shotgun assembler
Assembling other
mammalian genomes
• Arachne (Jaffe/Batzolou, Lander group at MIT)
– Built first draft of mouse genome, February 2002
– Mouse also assembled by Phusion assembler (Mullikin,
Sanger Centre)
• Atlas (Havlak/Chen/Durbin, Gibbs group at Baylor)
– Built first draft of rat genome, November 2002
Browsers as web-based
genome microscopes
• Ensembl Browser (Birney et al.)
• MapViewer (NCBI Mapviewer team)
• UCSC Genome Browser (Kent et al.)
http://genome.ucsc.edu, currently getting
more than 140,000 page requests per day
Browsers take you from early
maps of the genome . . .
. . . to a multi-resolution view . . .
. . . at the gene cluster level . . .
. . . the single gene level . . .
. . . the single exon level . . .
. . . and at the single base level
caggcggactcagtggatctggccagctgtgacttgacaag
caggcggactcagtggatctagccagctgtgacttgacaag
linking to functional information
In situ image from I. Dragatsis et al. 1998
Goal: the browser as a continuously-tuned
engine for discovery
• Multiple streams of high-throughput
genomics data generated asynchronously
• Data fed into nightly updates of browser
database, analysis and display
• Browser becomes a new kind of
microscope scanning the genome at ever
greater detail, dimension, and depth
Using evolution to find genes
and other functional elements
Mouse conservation pattern in the
IGFALS gene on human chr. 16 and a
known transcription factor binding site
R. Weber, L. Elnitski et. al.
At least half of the human genome consists
of relics of retrotransposons
DNA of genome
Retrotransposon
New copy of retrotransposon
1. Transcription
5. Insertion
of retrotransposon
DNA
2. Translation
RNA
3. Reverse transcription
of RNA to DNA
Reverse
transcriptase
4. Synthesis of second
DNA strand
Ancestral retrotransposons
• Retrotransposon relics from our common ancestor
with mouse and other placental mammals
• They cover 22% of the human genome
• “AR” sites can be used to study neutral evolution:
mutation without selection
• “AR” sites are similar to “4D” sites in genes
(four-fold degenerate sites in codons)
Estimated rate of neutral substitution from AR
and 4D sites co-varies along the chromosomes
R. Hardison, K. Roskin, S, Yang, A. Smit, et al.
By comparison to local neutral substitution
rates, it appears that about 5% of the human
genome may be under purifying selection.
K. Roskin, R. Weber, F. Chiaromonte
More species increases power to
detect conserved elements
Human
Chimp
Baboon
Cat
Dog
Pig
Cow
Rat
Mouse
Chicken
Zebrafish
Fugu
Tetraodon
BROWSER
SNAPSHOT
About 4% of CFTR region is under purifying selection
Data from Eric Green at NGHRI, alignments by Webb Miller
Models of molecular evolution
Branch length equals
average number of
substitutions per site
Models of molecular evolution
Branch length equals
average number of
substitutions per site
A
Models of molecular evolution
Branch length equals
average number of
substitutions per site
A
A
G
Models of molecular evolution
Branch length equals
average number of
substitutions per site
T
A
A
A
G
G
G
Models of molecular evolution
T
T
T
Branch length equals
average number of
substitutions per site
A
A
A
G
G
G
Models of molecular evolution
T
T
T
Branch length equals
average number of
substitutions per site
A
A
A
G
G
G
Continuous-time Markov models of
molecular evolution can be used to
calculate p-values for conservation
Conditional probability
distribution on each
branch has the form
P = eQt where t is the
time and Q is a 4 by 4
rate matrix.
Parameterizations of Q: JC, …, HKY, REV, UNR
Calculation of p-values
• p-value is probability of getting a given parsimony score
or better, using a cont. time Markov model of evolution
• p-values are calculated recursively for the two subtrees,
for all possible values of parsimony score and ancestral
bases for each subtree
• data for subtrees is combines to produce p-value at root
Method developed by Mathieu Blanchette and Martin Tompa
Calculation of p-values
• p-value is probability of getting a given parsimony score
or better, using a cont. time Markov model of evolution
• p-values are calculated recursively for the two subtrees,
for all possible values of parsimony score and ancestral
bases for each subtree
• data for subtrees is combines to produce p-value at root
Method developed by Mathieu Blanchette and Martin Tompa
Calculation of p-values
• p-value is probability of getting a given parsimony score
or better, using a cont. time Markov model of evolution
• p-values are calculated recursively for the two subtrees,
for all possible values of parsimony score and ancestral
bases for each subtree
• data for subtrees is combines to produce p-value at root
Method developed by Mathieu Blanchette and Martin Tompa
Examples of conserved regions
Analysis of CFTR region by Mathieu Blanchette
Regulatory modules
Mathieu Blanchette
Conserved RNA structure in a 3’ UTR
Mathieu Blanchette
Intronic RNA structural element
73kb to ST7 1st exon
73kb to ST7 2nd exon
~90 bp conserved stem
Mathieu Blanchette
Modeling different modes of
substitution
We want to pay attention to how
elements are conserved, not just
that they are conserved
Context matters
substitution rate matrix for non-coding dinucleotides
Adam Siepel
Dinucleotide and trinucleotide
models fit substitution data from
neutral regions much better
Improvement in log likelihood on AR sites
for higher order models of base substitution
Adam Siepel
Method also produces improved
models of codon evolution
Adam Siepel
Phylogenetic HMMs
human
baboon
mouse
dog
cat
cow
pig
chicken
TAATGGTA…CCAGTTA…GCAGAGT…
TAATGGTA…CCGGTTA…ACAGAGT…
CGATGGTG…CCGGTCG…ACAGAGC…
CTATGGTC…CCTGTTA…TCAGAGC…
GTATGGTC…CCTGTCG…TCAGAGC…
CCATGGTT…CCCGTAG…CCAGAGT…
CCATGGTT…CCCGTAG…CCAGAGT…
TTATGGTA…CCTGTTA…ACAGAGT…
Adam Siepel
Comparative cDNA analysis finds
alternatively spliced genes
Human splice variants of ZNF278 conserved in mouse
Chuck Sugnet
Molecular evolution is more
than base substitutions
• Insertions
• Deletions
• Duplications
• Inversions
• Rearrangements
Genome-wide human-mouse
alignments reveal a host of multibase
evolutionary events
A 15,000 base inversion on human
chromosome 7 containing two genes
J. Kent, W. Miller, R. Baertsch
Hot spots for rearrangements?
At finer resolution, many thousands of syntenic
blocks between human and mouse are found,
and short blocks are clustered in clumps
J. Kent, W. Miller, R. Baertsch
Grand challenge of human
molecular evolution
Reconstruct the evolutionary history
of each base in the human genome
Credits
Thanks to Jim Kent, Terry Furey, Mathieu Blanchette, Adam
Siepel, Chuck Sugnet, Ryan Weber, Krishna Roskin, Mark
Diekhans, Robert Baertsch, Matt Schwartz, Angie Hinrichs,
Donna Karolchik, Heather Trumbower, Yontao Lu, Fan Hsu,
Daryl Thomas, Jorge Garcia, Patrick Gavin and Paul Tatarsky
at UCSC
Francis Collins, Bob Waterston, Eric Lander, Richard Gibbs,
Eric Green, Elliot Margulies, David Kulp, Alan Williams, Ray
Wheeler, Webb Miller, Ross Hardison, Scott Schwartz,
Francesca Chiaromonte, Thomas Pringle, Greg Schuler,
Deanna Church, Steve Sherry, Ewan Birney, Michelle Clamp,
David Jaffe, Asif Chinwalla, Jim Mullikin,Tim Hubbard,
Arian Smit, Nick Goldman, Barbara Trask, Ian Dunham, Sean
Eddy, Evan Eichler, David Cox, Carol Bult, and many other
outside collaborators