Download TP+FP

C
E
N
T
R
E
F
O
R
I
N
T
E
G
R
A
T
I
V
E
B
I
O
I
N
F
O
R
M
A
T
I
C
S
V
U
Master Course
Sequence Alignment
Lecture 9
Database searching (3)
Dot-plots
a simple way to
visualise sequence
similarity
Can be a bit messy, though...
Filter:
6/10 residues have to match...
Dot-plots, what about...
• Insertions/deletions -- DNA and proteins
• Duplications (e.g. tandem repeats) – DNA
and proteins
Dot plots are
• Inversions -- DNA
calculated using a
diagonal window of
preset length that is
slid through the
search matrix -typically the central
cell holds the window
score (e.g. sum,
average)
Dot-plots, self-comparison
Direct repeat
Tandem repeat
Inverted repeat
charge
(cysteine bridge)
Protein structure hierarchical levels
PRIMARY STRUCTURE (amino acid sequence)
SECONDARY STRUCTURE (helices, strands)
VHLTPEEKSAVTALWGKVNVDE
VGGEALGRLLVVYPWTQRFFE
SFGDLSTPDAVMGNPKVKAHG
KKVLGAFSDGLAHLDNLKGTFA
TLSELHCDKLHVDPENFRLLGN
VLVCVLAHHFGKEFTPPVQAAY
QKVVAGVANALAHKYH
QUATERNARY STRUCTURE (oligomers)
TERTIARY STRUCTURE (fold)
Globin fold
 protein
myoglobin
PDB: 1MBN
Helices are
labelled ‘A’
(blue) to ‘H’
(red). D helix
can be missing
in some globins:
what happens
with the
alignment?
 sandwich
 protein
immunoglobulin
PDB: 7FAB
TIM barrel
 /  protein
Triose
phosphate
IsoMerase
PDB: 1TIM
Pyruvate kinase
Phosphotransferase
 barrel regulatory domain
/ barrel catalytic substrate binding
domain
/ nucleotide binding domain
What does this mean for
alignment?
• Alignments need to be able to skip
secondary structural elements to complete
domains (i.e. putting gaps opposite these
motifs in the shorter sequence).
• Depending on gap penalties chosen, the
algorithm might have difficulty with making
such long gaps (for example when using
high affine gap penalties), resulting in
incorrect alignment.
What does this mean for
homology searching?
• Database searching algorithms just need to
decide if the alignment score is good
enough for inferring homology
• Sometimes, alignments can be incorrect but
the score can be close enough for the
database searching method to correctly
identify the DB sequence as a homolog (or
not)
• However, for distant hits alignments
become crucial
Sequence Analysis/Database Searching
Finding relationships between genes and gene products of different species,
including those at large evolutionary distances
Compared to the preceding plot, RMSD is better able to pin-point relationships between
more divergent sequences (RMSD stays relatively small for a longer time as compared to
PAM distance) – Structure more conserved than sequence. Note that the spread around
RMSD is larger
Structural superpositioning
RMSD: how
far are
equivalenced
Cα atoms
separated on
average?
Two superposed protein
structures with two wellsuperposed helices
Red: well
superposed
Blue: low match
quality
C5 anaphylatoxin -- human (PDB code 1kjs) and pig
(1c5a)) proteins are superposed
How to assess homology search
methods
• We need an annotated database, so we know
which sequences belong to what homologous
(super)families
• Examples of databases of homologous
families are PFAM, Homstrad or Astral
• The idea is to take a protein sequence from a
given homologous family, then run the search
method, and then assess how well the
method has carried out the search
• This should be repeated for many query
sequences and then the overall performance
can be measured
C; family: zinc finger -- CCHH-type
C; class: small C; reordered by kitschorder 1.0a
C; reordered by kitschorder 1.0a
C; last update 7/9/98
>P1;1zaa1 structureX:1zaa: 3 :C: 33 :C:zinc-finger (ZIF268, domain 1):Mus musculus:2.10:18.20
------RPYACPVESCDRRFSRSDELTRHI-RI-HTGQK*
>P1;1zaa2 structureX:1zaa: 34 :C: 61 :C:zinc-finger (ZIF268, domain 2):Mus musculus:2.10:18.20
-------PFQCRI--CMRNFSRSDHLTTHI-RT-HTGEK*
>P1;1zaa3 structureX:1zaa: 62 :C: 87 :C:zinc-finger (ZIF268, domain 3):Mus musculus:2.10:18.20
-------PFACDI--CGRKFARSDERKRHT-KI-HLR--*
>P1;1ard structureN:1ard: 102 : : 130 : :zinc-finger (transcription factor ADR1):Saccharomyces cerevisiae:-1.00:-1.00
------RSFVCEV--CTRAFARQEHLKRHY-RS-HTNEK*
>P1;1znf structureN:1znf: 1 : : 25 : :zinc-finger (XFIN, 31st domain):Xenopus laevis:-1.00:-1.00
--------YKCGL--CERSFVEKSALSRHQ-RV-HKN--*
>P1;2drp2 structureX:2drp: 137 :A: 165:A:zinc-finger (tramtrack, domain 2):Drosophila melanogaster:2.80:19.30
----NVKVYPCPF--CFKEFTRKDNMTAHV-KIIHK---*
>P1;3znf structureN:3znf: 1 : : 30 : :zinc-finger (enhancer binding protein):Homo sapiens:-1.00:-1.00
------RPYHCSY--CNFSFKTKGNLTKHMKSKAHSKK-*
>P1;5znf structureN:5znf: 1 : : 30 : :zinc-finger (ZFY-6T):Homo sapiens:-1.00:-1.00
------KTYQCQY--CEYRSADSSNLKTHIKTK-HSKEK*
Example
You can
also look at
superposed
structures..
Sequence searching
QUERY
DATABASE
True Positive
True Positive
True Negative
POSITIVES
T
False Positive
NEGATIVES
True Negative
False Negative
So what have we got
Predicted
Observed
P
N
P TP
FP
N
TN
FN
Sensitivity and Specificity – medical world
+
+
T
e
s
t
-
9990
True
Positive
(TP)
990
False
Positive
(FP)
10
False
Negative
(FN)
989,010
True
Negative
(TN)
All with
Disease
10,000
All without
Disease
999,000
Sensitivity=
TP/(TP+
FN)
9990/(99
90+10)
Specificity=
TN/(FP+TN)
989,010/
(989,010+99
0)
All with Positive
Test
TP+FP
Positive Predictive
Value=
TP/(TP+FP)
9990/(9990+990)
=91%
All with Negative
Test
FN+TN
Negative Predictive
Value=
TN/(FN+TN)
989,010/(10+989,0
10)
=99.999%
Everyone=
TP+FP+FN+TN
Pre-Test Probability=
(TP+FN)/(TP+FP+FN+TN)
(in this case = prevalence)
10,000/1,000,000 = 1%
Receiver Operator Curve
(ROC)
• Plot Sensitivity (TP/(TP+FN)) against 1Specificity (1 - TN/(FP+TN)), where the
latter is called error
Sensitivity
Sensitivity is also
called Coverage
Error = 1 - specificity
Database Search Algorithms:
Sensitivity, Selectivity
•
Sensitivity – the ability to detect weak similarities between sequences
(often due to long evolutionary separation). Increasing sensitivity reduces
false negatives, i.e. those database sequences similar to the query, but
rejected.
Sensitivity (or Coverage) = TP / (TP+FN)
•
Selectivity – the ability to screen out similarities due to chance. Increasing
selectivity reduces false positives, those sequences recognized as similar
when they are not.
Selectivity (or Positive Prediction Value) = TP / (TP + FP)
•
Specificity also describes the ability of the method to select proper hits
Specificity = TN / (TN + FP)
Sensitivity
Selectivity, Specificity
Courtesy of Gary Benson (ISSCB 2003)
COG – Cluster of Orthologous
Groups
•Orthologues found using bidirectional best hit searching with
PSI-BLAST
•All COG family members are
supposed to have the same function
•Searching with an unknown
sequence only needs to hit a single
member of a COG family, annotation
can then be transferred
http://www.ncbi.nlm.nih.gov/COG/
COG2813
Structure-based function prediction
• SCOP (http://scop.berkeley.edu/) is a protein structure
classification database where proteins are grouped into a
hierarchy of families, superfamilies, folds and classes, based on
their structural and functional similarities
Structure-based function prediction
• SCOP hierarchy – the top level: 11 classes
Structure-based function prediction
All-alpha protein
membrane protein
All-beta protein
Alpha-beta protein
Coiled-coil protein
Structure-based function prediction
• SCOP hierarchy – the second level: 800 folds
Structure-based function prediction
• SCOP hierarchy - third level: 1294 superfamilies
Structure-based function prediction
• SCOP hierarchy - third level: 2327 families
Structure-based function prediction
• Using sequence-structure alignment method, one can predict a
protein belongs to a
– SCOP family, superfamily or fold
folds
superfamilies
families
•
•
•
Proteins predicted to be in the same SCOP family are orthologous
Proteins predicted to be in the same SCOP superfamily are homologous
Proteins predicted to be in the same SCOP fold are structurally
analogous
Profile wander
A
B
B
C
C
D
Multi-domain Proteins (cont.)
• A common conserved protein domain such as the
tyrosine kinase domain can obscure weak but
relevant matches to other domain types (e.g. only
appearing after 5000 kinase hits)
• Sequences containing low-complexity regions, such
as coiled coils and transmembrane regions, can
cause an explosion of the search rather than
convergence because of the absence of any strong
sequence signals.
• Conversely, some searches may lead to premature
convergence; this occurs when the PSSM is too strict
only allowing matches to very similar proteins, i.e.,
sequences with the same domain organization as the
query are detected but no homologues with different
domain combinations.
Multi-domain Proteins - DOMAINATION
Iterate PSI-BLAST
searches and domain
delineation
DOMAINATION
uses sequence signals
to identify domain
boundaries
George R.A. and Heringa J. (2002)
Protein domain identification and
improved sequence similarity
searching using PSI-BLAST, Proteins:
Struct. Func. Gen. 48, 672-681.
Multi-domain Proteins – DOMAINATION
method
P(boundary)
query
Strategy: Combine C- and N-termini of local
alignments to delineate domain boundaries
Count start and stops of alignments
DOMAINATION: Identifying domain boundaries
Sum N- and C-termini of
gapped local alignments
True N- and C- termini are
counted twice (within 10 residues)
Boundaries are smoothed using two
windows (15 residues long)
Combine scores using biased
protocol:
if Ni x Ci = 0
then Si = Ni + Ci
else Si = Ni + Ci +(Ni x Ci)/(Ni + Ci)
DOMAINATION: identifying domain
deletions
• Deletions in the query (or insertion in the
DB sequences) are identified by
– two adjacent segments in the query align to the
same DB sequences (>70% overlap), which
have a region of >35 residues not aligned to the
query.
(remove N- and C- termini)
DB
Query
DOMAINATION: identifying domain
permutations
• A domain shuffling event is declared
– when two local alignments (>35 residues)
within a single DB sequence match two
separate segments in the query (>70% overlap),
but have a different sequential order.
b
a
a
b
DB
Query
DOMAINATION: identifying continuous and
discontinuous domains
•Each segment is assigned an independence score (In).
If In>10% the segment is assigned as a continuous domain.
•An association score is calculated between non-adjacent
fragments by assessing the shared sequence hits to the
segments. If score > 50% then segments are considered as
discontinuous domains and joined.
Low Complexity segments
• A sequence of L residues of N types can have L!/N na!
different sequences of that same composition, where the
composition vector = (n1,.., na,.., N) and
N na! = n1! * n2! * .. * nN!
• If Rc is a vector of length N, where the vector numbers
correspond to the number of residues with a given frequency
(e.g. there are 5 amino acid types with 0 abundance, 3
amino acid types with abundance 1, etc., in the sequence),
then the total number of distinct sequences corresponding to
a particular complexity state-vector is
(L! / N na!) * (N! / L rc!), where L rc! = r0! * r1! * .. * rL-1! * rL!
• Based on this, the final complexity score calculated by the
SEG program is
PSEG = (1/NL) * (L! / N na!) * (N! / L rc!)
DOMAINATION: Post-processing low
complexity regions in database sequences
Remove local fragments with > 15% LC
Conserved hypotheticals
>P00001 Conserved hypothetical
A substantial fraction of genes in sequenced genomes encodes 'conserved hypothetical'
proteins, i.e. those that are found in organisms from several phylogenetic lineages but have not
been functionally characterized.
Profile wander (or matrix migration)
• Permissive iterative searching user higher E-values
can lead to incorrect hits (false positives) that
become included into the profile. More incorrect hits
can then be added in subsequent iterations, and true
homologues can be lost. Also, the search can
explode, leading to large numbers of spurious hits.
• A further loss of information can be incurred with
PSIBLAST, because PSI-BLAST PSSMs are trimmed
to only use the highest scoring region in a search,
ignoring less conserved regions
Sequence identity scoring zones
• >25-30%: homology zone
• 15-25%: twilight zone
• <15%: midnight zone (Rost, 1999)
Is midnight zone properly definable?