Transcript Conservation scores, updated

Conservation Scores
BNFO 602/691
Biological Sequence Analysis
Mark Reimers, VIPBG
Conservation and Function: what kinds
of DNA regions get conserved?
• Core coding regions are usually conserved
across hundreds of millions of years (Myr)
• Active sites of enzymes and crucial structural
elements of proteins are highly conserved
• Untranslated regions of genes are conserved
over tens but not over hundreds of Myr
• Some regulatory regions evolve ‘quickly’ –
over a time scale of tens of Myr
Conservation and Function: what kinds
of DNA regions get conserved?
• Many splice sites and splice regulators are
conserved between mouse and human
• Most promoters (70%) conserved between
mouse and human
• Majority (~70%) of enhancers not conserved,
but a significant minority are highly conserved
Approaches to Scoring Conservation
•
•
•
•
Base-wise: PhyloP, GERP
Small regions: PhastCons
Small regions, tracking bias: SiPhy
Regulatory conservation within exons may be
detected by any of these methods
• Key regulatory regions are harder to see
DEMO:
UCSC Alignment & Conservation Tracks
Genomic Alignment
• Alignment is crucial (and not trivial)
– Common alignment algorithms may misplace
ambiguous bases, leading to artifactual gaps
– Inversions are often badly handled
• Issue: incomplete alignments are not reflected in
scores of any current algorithm
– Conservation scores computed on aligned genomes
only
• Alignments of 46 placental mammals to human
genome in MultiZ format at UCSC
– Subset of primate alignments also
Alignment Issues
• When studying protein-coding regions,
substitutions are most common
• Most genome evolution happens through
insertions or deletions
– Human chimp alignable genome is 97% identical
– Only 91% of genome is alignable
• Regions may acquire regulatory function in
some lineages but have no function in most
UCSC Alignment Symbols
• Single line ‘-’: No bases in the aligned species.
– May reflect insertion in the human genome or
deletion in the aligning species.
• Double line ‘=‘: Aligning species has unalignable
bases in the gap region.
– Many mutations or independent indels in between
the aligned blocks in both species.
• Pale yellow coloring: Aligning species has Ns in
the gap region.
– Sequencing problems in aligning species
Conservation Across Mammals Differs
from Conservation Across Primates
• Many regions conserved
across mammals are also
conserved across
primates
– a few appear not to be
• Some regions appear to
be conserved (insofar as
can be measured) in
primates but not across
all mammals
• What is the diagonal?
Are these regions
conserved?
How to Assess Conservation?
• If all bases in one position are identical, while
others around it vary over all possibilities
• Over what lineage?
• How to improve power with modest chance of
variation at any one site?
– Look to neighboring sites’ conservation
• How to identify constraint, if not complete
identity?
Genomic Evolutionary Rate Profiling
(GERP) Measures Base Conservation
• Estimates neutral evolution rate as mean number of
substitutions in each aligned genome
• Original score (Cooper, 2005) is “rejected substitutions”:
number of substitutions expected under ‘neutrality’ minus
number of substitutions observed at each aligned position
• New scores based on ML fit of substitution rate at base
• Positive scores (fewer than expected) indicate that a site is
under evolutionary constraint.
– Negative scores may be weak evidence of accelerated rates of
evolution
PhyloP Assigns Conservation P-values
• Estimates mean number of substitutions in each
aligned genome to estimate neutral evolution rate
estimated from non-coding data (conservative)
• Computes probability of observed substitutions
under hypothesis of neutral evolutionary rate
• Scores reflect either conservation (positive scores) or
selection (negative scores)
• Score defined as –log10(P) where P is p-value for test
of number of substitutions following (uniform)
neutral rate inferred from all sites in alignment
NB PhyloP also refers to a
suite implementing four
related methods (Pollard
et al, Gen Res 2010)
PhastCons Fits a Hidden Markov Model
• PhastCons fits HMM with
states ‘conserved’ and
‘not conserved’
• Neutral substitution rates
estimated from data as
for PhyloP
• Tunable parameter m
represents inverse of
expected length of
‘conserved’ regions
• Parameter n sets
proportion of conserved
Siepel et al. Genome Res.
regions
2005;15:1034-1050
PhastCons Fits a Hidden Markov Model
• Scaling parameter ρ (0 ≤ ρ ≤ 1) represents the
average rate of substitution in conserved regions
relative to average rate in non-conserved regions
and is estimated from data
• Originally developed to detect moderate-sized
sequences such as non-coding RNA
• Can be adapted to shorter sequences but not as
powerful
• Not designed for disconnected conserved regions
–e. g. binding sites for multi-finger TF
SiPhy is Sensitive to Biased Substitution
• SiPhy models the pattern of substitutions,
rather than just the rate, as do most others.
– Biased substitutions (e.g. conserved lysine:
AAA <-> AAG only) will be identified as constrained
– Some TFBS have similar degeneracy in evolution
– This is a more refined approach than rate models,
but requires a fairly deep (or wide) phylogeny
• SiPhy uses a Bayesian approach and needs
two parameters (like PhastCons):
– the fraction of sequence conserved
– typical length of a conserved region.
Two Versions of SiPhy: w and p
• SiPhy-w estimates a global bias pattern R
• SiPhy-p estimates each bias pattern
• Generally done with short regions (e.g. 12 nt)
SiPhy Applied to Mammalian Genomes
Identification of four NRSF-binding sites in NPAS4.
K Lindblad-Toh et al. Nature (2011)
Comparison of Methods
• PhyloP and GERP give fairly similar results over
deep phylogenies (e.g. vertebrates)
• Differ substantially over bushes (e.g. primates)
• PhastCons is faster to run than SiPhy
• SiPhy is more sensitive over moderately deep
phylogenies (e.g. mammals)
– Cannot be implemented for primates because of
insufficient substitutions
Issues With Conservation Scores
• Most scores are misleading about gaps in
alignments: they don’t distinguish between
contig gaps (incomplete genomes) and
inserted or deleted regions
– This information is often available, but
inconvenient
– Older genomes had many gaps
– Modern model organism genomes are fairly
complete
– Alignment is still an issue
Issues With Conservation Scores
• Each model was devised with a particular kind
of conserved element in mind, and may not
be adaptable to all kinds of elements
– Short constrained sequences vs. exons
– Multi-finger TF binding sites are not done well
• No method tests for constraint over a specific
lineage