BIOL335: Homology search

Homology Search
Paul Gardner
March 24, 2015
Paul Gardner Homology Search

News & Views reminder (20% of your course grade, due
March 26, Reviewed April 2 (5/20), Revisions April 28
(15/20))
Meredith et al. (2014) Evidence for a single loss of
mineralized teeth in the common avian ancestor. Science
Nunez et al. (2015) Integrase-mediated spacer acquisition
during CRISPR-Cas adaptive immunity. Nature

Homology search
In a huge collection of biological
sequences how can you locate
similar sequences?
by using heuristic, super fast,
sequence alignment methods

BLAST

BLAST
Identify all ’hits’ of at least W long
Find any hits on the same diagonal of an alignment matrix
Trigger a full alignment in that region
Basic idea: identify near-identical sub-sequences ﬁrst → align any
hits in full

What does that E-value (Expect) mean?
>gb|CP001191.1| Rhizobium leguminosarum bv. trifolii WSM2304, complete genome
Length=4537948
Features in this part of subject sequence:
cold-shock DNA-binding domain protein
Score = 57.2 bits (62), Expect = 2e-05
Identities = 78/106 (74%), Gaps = 6/106 (6%)
Strand=Plus/Plus
Query 1 CTTCGTCAGATTTCCTCTCAATATCGATCATACCGGACTGATATTCGTCCGG----GAAC
|| |||||||| ||||||||| |||||| | | | || |||| |||| ||||
Sbjct 828507 CTCCGTCAGATATCCTCTCAACATCGATACGGCTTGTCGGACATTCTTCCGCAGGCGAAC
Query 57 TCTAGCGATTGAAA-GGAAATCGTTATGAACTCAGGCACCGTAAAG
| | || |||||| ||| ||||||||||| |||||| ||| |||
Sbjct 828567 ACAA-CGGTTGAAAAGGAGATCGTTATGAATTCAGGCGTCGTCAAG

How can we evaluate the signiﬁcance of a score?
Note that a bit-score of 57.2 by itself is not that useful.
It depends on the sequence & database size & composition.
To counter this we can compute an Expect-value (E-value).
This is the expected number of hits with the observed score for
the given query and database sizes.
P-values can also be used
0 100 200 300 400 500 600 700
0
2000
4000
6000
8000
10000
Separating true from false hits
score (bits)
Num.matches
Random sequences/Negative controls
True homologs/Positive controls
Threshold
False negatives
True positives
False positives
True negatives

How can we evaluate the signiﬁcance of a score?
0 100 200 300 400 500 600 700
0
2000
4000
6000
8000
10000
Separating true from false hits
score (bits)
Num.matches
Random sequences/Negative controls
True homologs/Positive controls
Threshold
False negatives
True positives
False positives
True negatives
E = κMN2−λx
E: E-value
M&N: query &
database size
κ&λ: ﬁtting
parameters

BLAST is not the only, or best tool for the job!

Proﬁle-based homology search
Krogh, A. et al. (1994) Hidden Markov models in computational biology. Applications to protein modeling. J Mol
Biol.
Image provided by Eric Nawrocki.

Proﬁle-based homology search – scoring sequences
Image provided by Eric Nawrocki.

Proﬁle HMM are slightly more complicated
A tree-weighting scheme takes care of unbalanced
alignments
Dirichlet-mixture priors are used to incorporate information
about amino-acid biochemistry
Eﬀective sequence number is used to down-weight priors
when many sequences are available
Transition probabilities to Insert & Delete states are estimated
from the alignment

Why not just use BLAST?
ACCURACY!
Every benchmark of homology search tools has shown that
proﬁle methods are more accurate than single-sequence
methods.
Eddy (2011) Accelerated Proﬁle HMM Searches. PLoS
Computational Biology.

Why not just use BLAST?
SPEED! To search a single query vs a database of all proteins:
BLAST: searches 42 million UniProt sequences
HMMER: searches 15,000 Pfam proﬁles
The search space is ∼ 3, 000x smaller for proﬁles
Save Planet Earth, use HMMER3

Pfam
What is a Pfam-A Entry?
hmmsearch
hmmbuild
hmmalign
SEED
HMM
OUTOUT
ALIGNDESC
Slide borrowed from Rob Finn.

But, what about RNA?
5’
3’
0
Sequence conservation
1
A
G
U
K G
C
U
C
A
U
U
CA
C
C
K
W
Y U
U
A
U
G
W
YR
G
YCC
C
g
C
Y
V
U
U
H R G C G
G
A
A
K
A
Y
G
YG
C
U
W
C
A
U
A
A R
M
Y
A
Y
C
G
A
A
U
G
AY
G
C M
H
A
A
G
M
M
WG
G
U
G
C
C
U R
Y
C
G
U
C
C A M
C
U
W
A
a
C
Y
G
A
U
A
W Y
R
K
G
U
G
MRU
R
C
R
C
W
U
U
A
U
C
AA
V
C
A
Y
C
G
G
R
C
GA
M
A
C
G
UY
G
A G
U
K
A
G
G
C
A
C
CGC
C
U
W
5’
3’
0
1
A
A
Y
A
A
A
A
U
A
A
U
U
U
A
C
AUUCCA AG
G
A
C
C
G
G
UA
U
U
A
U
U
GU A
G
G
G
G
A
U
U
U
GU
G
AC
U
U
Y C
A
A
G
G
C
A
A
Y
G
U
C
C
U
C
U
C
U
A
C
AA
C
C
G
A
G
U
U
C R
A
G
A
A
U
A
A
R
Y
A
C
M
A
A
YG
G
C
U
C U
U
U
U
U
G
UU
A
U
U
C
G
A
A
A
G C
U
U
A
C
A
A
G
DU
V
Y
R
G
Y
R
U
M
U
U
C
U
R
U
A
U
R
C
U
C
W
C
Y
Uc
a
M
U
Y
A C
U
U
U
C
M
A
G
U
AC
U
U
C
A
C
A
C G
G
G
C
CWRACAK
M
U
5’ 3’
0
1
U
V
D
WHAUGA
U
G
A
G
Y
U
C
M
A
C
U
U
C
W
U
u
G
G
U
C
C
G
U
G U U U C U G A g a R
M
C
Y
M
R
U
G
A
U
M
U
B
W
R
U
G
a
S
A
A
a
G
U
UCUGAY
U
H
M

Covariance models
Nawrocki & Eddy (2007) Query-Dependent Banding (QDB) for Faster RNA Similarity Searches. PLOS
computational biology.

Benchmark
Freyhult, Bollback & Gardner (2007) Exploring genomic dark matter: A critical assessment of the performance of
homology search methods on noncoding RNA. Genome Research.

Rfam

Relevant reading
Reviews:
Eddy SR (2004) What is a hidden Markov model? Nature
Biotechnology.
Methods:
Altschul SF et al. (1997) Gapped BLAST and PSI-BLAST: a
new generation of protein database search programs. Nucleic
acids research.

The End

BIOL335: Homology search

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à BIOL335: Homology search

Similaire à BIOL335: Homology search (20)

Plus de Paul Gardner

Plus de Paul Gardner (20)

Dernier

Dernier (20)

BIOL335: Homology search