Transcript Assembling and Annotating the Draft Human Genome

The Genes, the Whole Genes,
and Nothing But the Genes
Jim Kent
University of California Santa Cruz
Ben Franklin - Childhood Hero
Hi Voltage Experiments
A Man of High Values
Early to bed
Early to rise
Rock Collection
Shell Collection
Bottlecap Collection
Bug Collection
Jim Kent
Genome Scientist
(not to be confused with Richard Stallman)
Modern Bug Collection
if (a = b)
if (string == “something”)
for (x=0; x<count; ++x);
process(x);
for (x=0; x<width; ++x)
for (y=0; y<height; ++x)
plot(x, y, data[x][y]);
Naive Biological Questions
•
•
•
•
Is an ant an individual or is it the hive?
Do dolphins talk with each other?
How do amphibians and worms regenerate?
How does an animal develop out of an egg?
From Egg to Adult in 3x109 Bases
• A single cell, the fertilized egg, eventually
differentiates into the ~300 different types of cells
that make up an adult body.
• With a few exceptions all of these cells contain the
full human genome, but express only a subset of
the genes.
• Gene expression patterns are determined largely
by the cell type, and vice versa.
From Totempotency to Senility
• Human cells become more and more
specialized during development
• An egg can become anything. (Initially
most of it will become placenta & amnion).
• Liver cells only become liver cells.
• Neurons can’t even reproduce.
Cell Type Determinants
During Development
• Cell type of parent cell.
• Interactions with other cells.
• Interactions with the extracellular
environment.
Primary Flows of Information and Substance in Cell
DNA
creation
regulation
mRNA
transcription
factors
splicing
factors
Environment
& other cells
Receptors
signaling
molecules
Enzymes
energy
structural
proteins
structural structural
sugars
lipids
An Extreme Case of
Dedifferentiation
• The cloning of Dolly the sheep showed that
a differentiated genome could be reset.
• An egg is huge compared to a normal cell.
• Putting a normal cell into an egg as Wilmut
et al did, swamps out the normal cell
transcription factor and receptors with egg
transcription factors and receptors.
• Cloning success rate sometimes improved
by passing a nucleus through multiple eggs.
Regeneration by Nature
• Among vertebrates only amphibians can
regenerate limbs.
• The process involves dedifferentiation,
repatterning, and growth.
• Not likely we’ll be able to engineer this soon.
• Simpler regenerations though may be tractable and
medically quite important.
Human Diseases Involving a
Small Population of Cells
• Parkinson’s - from the death of dopamine
producing neurons in the substantia negra.
• Macular degeneration - a leading cause of
blindness in the elderly.
• Type I Diabetes - from the death of insulinproducing cells in the pancreas.
Pancreas Differentiation Pathway
From Huang Tsai, J Biomed Sci 2000:7:27-34 and Jensen et al, Diabetes 2000:49 163-176
Flexibility of Stem Cells
• In many cases stem cells are flexible enough that
putting them into a particular tissue will cause
them to differentiate into the type of cells that
make up that tissue.
• At low levels bone transplanted bone marrow
(blood stem cells) develops into neurons in stroke
victims!
• Making this happen at high enough levels to be
useful will likely require some engineering.
To Understand the Body Need
•
•
•
•
•
•
•
The genome
A comprehensive list of genes
Gene expression data
Protein localization in cell
Protein/protein and protein/DNA interaction
information.
Ways to store, display and query masses of data
so human investigators can focus on relevant
bits.
Many human investigators.
Where are we now?
•
•
•
•
•
The genome >95% complete. 98% complete in
April.
A comprehensive list of genes - ~75% of coding
regions. <50% of transcription start sites.
Gene expression data - publically available on
~1/3 of genes.
Protein localization in cell - very spotty.
Computer predictions are about 75% accurate.
Protein/protein and protein/DNA interaction
information - just getting started.
The Genes
• Identifying genes is a prerequisite for a
great deal of other research.
– Expression microarrays
– In situ mRNA hybridization
– Producing proteins for cellular localization
experiments
– Etc.
The Whole Genes
• The full gene including the 5’ and 3’ UTRs
are critical for
– Avoiding misleading fragmentation/fusion
artifacts.
– Understanding mRNA targeting and stability
– Finding transcription factor binding sites
– Understanding the regulatory networks that
drive and maintain cell differentiation.
Nothing But the Genes
• Experimental analysis is expensive.
• Unreal genes can mislead:
– Analysis of multiple alignments to look for
active sites etc.
– Protein classification systems and phylogenies
• One bogus gene can lead to another as
much annotation is done via homology.
Methods of Identifying Genes
•
•
•
•
•
•
mRNA/cDNA sequencing
Microarrays covering entire genome
Genetics in model organisms
Cross species protein homology
Cross species genomic homology
HMMs and other computational
genefinding.
cDNA Sequencing
• Extract RNA from cells.
• Use reverse transcriptase and a poly-U primer to
convert to cDNA starting at poly-A tail.
• Insert cDNA into vectors that grow in E. coli
• Sequence a read from one or both sides of insert
using primers on vector
• If EST looks to be new sequence full cDNA.
• Artifacts and limitations are possible at each stage!
Common cDNA Problems & Solutions
• For rarely expressed genes little RNA is available.
– Normalize libraries. Use embryonic and exotic tissues
as mRNA source.
• Splicing is not instantanious, can get retained
introns.
– Spin out nuclei and just use cytoplasmic mRNA
– Align to genome and look for splicing
• Reverse transcriptase falls off before it’s finished
– Preferentially taking larger cDNAs.
– G-cap selected libraries (Sugano)
– Normalizing only on 5’ ends (Soares)
More cDNA Problems & Solutions
• Reverse transcriptase has a high error rate and is
prone to small deletions.
– Compare cDNA to genomic DNA
– Sequence multiple cDNA clones
• At a low level cell seems to tolerate a certain
degree of nonsense transcription and splicing.
Normalizing increases concentration of these as
well as of rare genes.
– Ignore everything that’s not coding (ouch)
– ???
cDNA Status & Summary
• ~10,000 cDNA sequence have been accumulated
over years by various labs working on gene
families and pathways.
• Riken project has ~33,000 unique cDNAs in
mouse. ~11,000 of these seem to have retained
introns. ~3,000 are noncoding antisense. ~70%
include initial ATG
• Mammalian Gene Collection (MGC) has ~15,000
human cDNAs with initial ATGs. Having to resort
to exotic libraries and RT-PCR to get more.
• Human refSeq has ~18,000 human cDNAs.
Whole Genome Microarrays
• Perlegen and Affymetrix are making microarrays
that cover entire non-RepeatMasker masked
genome. Results on chromosome 21 and 22
published.
• Based on 25-mers.
• Rarely expressed genes may not stand out above
background.
• Have to cope with cross-hybridization issues, GC
content, etc.
• Advantages - no homology required, can sense
lower concentrations of mRNA than random EST
sequencing.
Cross-hybridization at Work
Zoomed in on right side:
This turns out to be a pseudogene for TF2Ebp
Model Organism Genetics
• Zap hapless yeast, worms, flies, and mice
• Inbreed offspring and look for twisted ones.
• Advantages:
– Works at DNA level, so expression level doesn’t matter
– You get hints of function right away.
– Can look for gene interactions simply by breeding
mutants.
• Disadvantages:
– Finding which DNA is mutated can take a long time.
– Essential genes can be hard to find - all you see is
reduced fertility in the inbreeding stage.
– Genes only needed in certain environments and
duplicated genes may be missed in screens.
Comparative
Genomics
Cross Species Genome
Comparisons
• Mutations occur more or less randomly across
genome but
• Mutations in functional areas tend to be weeded
out by selection
• In comparing DNA across species, the functional
areas are more conserved than the nonfunctional
areas in general
• Caveat: ~5% of human genes don’t have clear
mouse orthologs (though most do have paralogs).
Comparative Genomics at BMP10
Conservation of Gene Features
100%
95%
90%
85%
80%
75%
70%
65%
60%
Intron Mid CDS Intron
55%
50%
Up 200
5’ UTR
1st CDS
aligning
identity
last CDS 3’ UTR Down 200
Conservation pattern across 3165 mappings of human
RefSeq mRNAs to the genome. A program sampled 200
evenly spaced bases in introns, coding exons, and UTRs,
and 200 bases upstream of transcription, and downstream
of poly-adenlyation. There are peaks of conservation at the
transition from one region to another.
Detail Near Translation Start
100%
95%
90%
85%
80%
75%
70%
65%
60%
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15
Note the relatively conserved base 3 before translation
Start (constrained to be a G or an A by the Kozak
Consensus sequence, and the first three translated bases
(ATG).
Normalized eScores
Computational Gene Finding
10 Different Gene Finders over a region containing two genes.
One gene has a refSeq mRNA and is ‘Known.’ There is
also a full length mRNA for an ortholog of the gene at the right.
Basic Techniques
• Bacteria - look for open reading frames - long
stretches between start and stop codons.
• Eukaryotes - introns are challenging
– Look for coding exons (bounded by AG / GT)
– HMMs can model coding regions and splice sites
simultaniously
– Generalized HMMs (genscan) can string together
probable exons
– Homology based ones (GeneWise) can map proteins to
genome allowing for some evolutionary divergence and
sequencing error.
Limitations of Basic Approach
• Introns are vast, GT/AG splice signals are small.
• Coding signal is stronger than start/stop signal. As
a result gene exons are often correct but genes are
fused and split.
• Pseudo genes, processed and otherwise, mimic
coding regions.
• Pure HMM approaches tend to overpredict
• Pure homology approaches only can tell us about
what we already know.
Composite Approaches
• Use protein homology info on top of HMMs
(fgenesh++, GenomeScan, Ensembl)
• Use EST info to constrain HMMs (Genie)
• Use cross species genomic alignments on
top of HMMs (twinscan, fgenesh2, SLAM,
SGP)
Computational Gene Finding
RefSeq
ESTs
protein
protein
X genome
X genome
X genome
protein
Ab initio
Ab initio
Computational + Wet
• Even the best computational genefinding is not
good enough.
• Mouse/human genomic homology reduces false
positive exons a fair amount typically for a given
level of sensitivity.
• Getting cDNA data on a gene is much easier if you
have a piece of it to start with. You can use the
piece to probe cDNA libraries rather than doing
random sequencing.
• On chromosome 22 Sanger probed ~60 libraries
with ~every genscan prediction, spliced EST, and
pufferfish homology.
Conclusions
• Human genome sequencing essentially complete.
• Currently have identified full coding region of
~18000 of ~25000 human protein coding genes.
• The remaining genes are generally harder to
identify than the ones we have identified already.
• The remaining genes will require a careful mix of
computational and wet work.
Acknowledgements
Individuals
Institutions
David Haussler, Angie Hinrichs,
Chuck Sugnet, Matt Schwartz,
Robert Baertsch, Donna Karolchik,
NHGRI, The Wellcome Trust,
HHMI, NCI, Taxpayers in the
US and worldwide.
Francis Collins, Bob Waterston, Eric
Lander, John Sulston, Richard Gibbs
Roderic Guigo, Michael Brent, Chris
Burges, Olivier Jaillon, David Kulp,
Victor Solovyev, Ewan Birney, Greg
Schuler, Deanna Church, Asif
Chinwalla, the Gene Cats.
Everyone else!
Whitehead, Sanger, Wash U,
Baylor, Stanford, DOE, and
the international sequencing
centers.
UCSC, Mouse Sequencing
Consortium, NCBI, Ensembl,
Genoscope, MGC, Softberry,
Affymetrix.
THE END
Parasol and Kilo Cluster
• UCSC cluster has 1000 CPUs
running Linux
• 1,000,000 BLASTZ jobs in 25
hours for mouse/human
alignment
• We wrote Parasol job
scheduler to keep up.
– Very fast and free.
– Jobs are organized into batches.
– Error checking at job and at
batch level.