Biological Databases for Protein Sequence Analysis

1
Biological Databasesfor Protein Sequence Analysis

• Terri Attwood
• School of Biological Sciences
• University of Manchester, Oxford Road
• Manchester M13 9PT, UK
• http//www.bioinf.man.ac.uk/dbbrowser/

2
Overview

• Introduction
• Web practical, science fact fiction
• the Twilight Zone, the Midnight Zone
• Biological databases
• primary, secondary pattern, composite, etc.
• Pattern recognition
• regular expressions, fingerprints, profiles,
  etc.
  
• Building a search protocol
• combining results, estimating significance

3
The practical - BioActivity

• BioActivity is intended to support the lectures
• you begin with a DNA sequence fragment
• try to find out what protein this codes for, the
  family to which it belongs, whether its function
  structure are known, etc.
  
• The practical is entirely Web-based
• largely uses local servers, but also links to
  external sites
  
• be patient mindful of traffic - don't waste
  time on slow links
  
• Most important of all
• please read the instructions!
• The Web is constantly evolving....
• please report dead links (otherwise theyll stay
  dead)!

4
(No Transcript)
5
The stuff you have to know

• Single- three-letter amino acid codes
• G Glycine Gly P Proline Pro
• A Alanine Ala V Valine Val
• L Leucine Leu I Isoleucine Ile
• M Methionine Met C Cysteine Cys
• F Phenylalanine Phe Y Tyrosine Tyr
• W Tryptophan Trp H Histidine His
• K Lysine Lys R Arginine Arg
• Q Glutamine Gln N Asparagine Asn
• E Glutamic Acid Glu D Aspartic Acid Asp
• S Serine Ser T Threonine Thr
• Additional codes
• B Asn/Asp Z Gln/Glu X Any amino acid

6
(No Transcript)
7
Basic definitions

• Primary structure
• the linear sequence of amino acids in a protein
• Secondary structure
• regions of local regularity
• i.e., a-helices, b-strands, -sheets -turns

8
Definitions contd.

• Super-secondary structure
• the packing of secondary structure elements into
  stable units
  
• e.g., b-barrels, bab units, Greek keys, etc..

9
Definitions contd.

• Tertiary structure
• the overall chain fold that results from packing
  of secondary structure elements

10
Definitions contd.

• Quaternary structure
• the arrangement of separate chains within a
  protein that has more than one subunit
  
• e.g., haemoglobin

11
Definitions contd.

• Quinternary structure
• the arrangement of separate molecules, such as in
  protein-protein or protein-nucleic acid
  interactions

12
Definitions contd.

• Bioinformatics
• broadly, Information Technology applied to
  biology
  
• this can mean anything from AI robotics to
  genome analysis!
  
• boundaries with computational biology now
  blurred
  
• originally coined in the 80s to mean
  bio-sequence analysis
  
• with increasing availability of protein
  structures, the term now also encompasses
  structure analysis
  
• but the scale of the problem here is vastly
  different.....

13
Importance of sequence analysis

• 694,000 sequences available in public databases
• millions more (including ESTs) in proprietary
  databases
  
• these s will snowball with completion of more
  genomes
  
• so what?
• Locked up in sequences is a huge amount of
  structural, functional evolutionary info
  
• they're a highly valuable resource
• By contrast, the of unique protein structures
  is 2000
  
• this represents a huge information deficit

14
Sequence-structure deficit

• Non-redundant growth of sequences during
  1988-1998 ( ) the corresponding growth in
  the number of structures ( ).

15
Challenges for bioinformatics

• Spurred on by the sequence/structure deficit, the
  challenges are to
  
• rationalise the mass of sequence data
• derive more efficient means of data storage
• design more incisive reliable analysis tools
• The imperative - to convert sequence information
  into biochemical biophysical knowledge
  
• to decipher the structural, functional
  evolutionary clues encoded in the language of
  biological sequences

16
The Holy Grail of bioinformatics

• ...to be able to understand the words in a
  sequence sentence that form a particular protein
  structure

17
The reality of sequence analysis

• ...isn't so glamorous....but means we can
  recognise words that form characteristic
  patterns, even if we don't know the precise
  syntax to build complete protein sentences

18
Pattern recognition prediction

• In investigating the meaning of sequences, 2
  distinct analytical approaches have emerged
  
• pattern recognition is used to detect similarity
  between sequences hence to infer related
  structures functions
  
• ab initio prediction is used to deduce structure,
  to infer function, directly from sequence
  
• These methods are different shouldnt be
  confused
  
• Sequence- structure-based pattern recognition
  methods demand that some characteristic has been
  seen before housed in a db
  
• Prediction methods remove the need for template
  dbs because deductions are made directly from
  sequence

19
Science fact fiction

• Sequence pattern recognition is easier to
  achieve, is much more reliable, than fold
  recognition
  
• which is 40-50 reliable even in expert hands
• Prediction is still not possible
• is unlikely to be so for decades to come (if
  ever)
  
• Structural genomics will yield representative
  structures for more proteins in future
  
• structures of new sequences will be determined by
  modelling
  
• prediction will become an academic exercise
• But, to debunk a popular myth, knowing structure
  alone does not inherently tell us function

20
A reality check

• What is the function of this structure?

• What is the function of this sequence?

• What is the function of this motif?
• the fold provides a scaffold, which can be
  decorated in different ways by different
  sequences to confer different functions - knowing
  the fold function allows us to rationalise how
  the structure effects its function at the
  molecular level

21
A test case for structural genomics
Structure-based assignment of the biochemical
function of hypothetical protein mj0577
(Zarembinski et al., PNAS 95 1998)
Although the structure co-crystallised with ATP,
the biochemical function of the protein is
unknown

22
The Twilight Zone

• Prediction methods dont work because we dont
  fully understand the Folding Problem
  
• we cant read the language sequences use to
  create their folds
  
• But, with sequence analysis techniques, we can
  try to find similarities between new sequences
  those in dbs
  
• whose structures functions we hope have been
  elucidated
  
• This is straightforward at high levels of
  identity, but below 50 it is difficult to
  establish relationships reliably
  
• Analyses can be pursued with decreasing certainty
  towards the Twilight Zone
  
• 20 identity, where results may look plausible
  to the eye, but are no longer statistically
  significant

23
Application areas of analysis tools

• The scale indicates identity between aligned
  sequences
  
• Alignment of 2 random seqs can produce 20
  identity
  
• less than 20 does not constitute a significant
  alignment
  
• around this threshold is the Twilight Zone,
  where alignments may appear plausible to the eye,
  but cant be proved by conventional methods

24
Homology analogy

• The term homology is confounded abused in the
  literature!
  
• sequences are homologous if theyre related by
  divergence from a common ancestor
  
• analogy relates to the acquisition of common
  features from unrelated ancestors via convergent
  evolution
  
• e.g., b-barrels occur in soluble serine proteases
  integral membrane porins chymotrypsin
  subtilisin share groups of catalytic residues,
  with near identical spatial geometries, but no
  other similarities
• Homology is not a measure of similarity is not
  quantifiable
  
• it is an absolute statement that sequences have a
  divergent rather than a convergent relationship
  
• the phrases "the level of homology is high" or
  "the sequences show 50 homology", or any like
  them, are strictly meaningless!
  
• This is not just a semantic issue
• loose use muddies thinking about evolutionary
  relationships

25
A terminology muddle

• In comparing 3D structures, exactly the same
  arguments apply
  
• structures may be similar, as denoted by RMS
  positional deviation between compared atomic
  positions
  
• common evolutionary origin remains a hypothesis,
  until supported by other evidence
  
• homology among similar structures is a
  hypothesis
  
• This may be correct or mistaken, but their
  similarity is a fact, no matter how it is
  interpreted
  
• Similarity of sequence or structure is just that
  - similarity
  
• Homology connotes a common evolutionary origin
• Reeck, G.R., de Haen, C., Teller, D.C.,
  Doolittle, R.F., Fitch, W.M., Dickerson, R.E.,
  Chambon, P., McLachlan, A.D., Margoliash, E.,
  Jukes, T.H. Zuckerkandl, E. (1987) Homology
  in proteins and nucleic acids a terminology
  muddle and a way out of it. Cell, 50, 667.

26
Orthology paralogy

• Among homologous sequences we can distinguish
• orthologues - largely perform the same function
  in different species
  
• paralogues - perform different but related
  functions in one organism
  
• Studying orthologues opens the way to molecular
  palaeontology
  
• e.g., using phylogenetic trees to show
  cross-species relationships
  
• Paralogues shed light on underlying evolutionary
  mechanisms
  
• paralogous proteins are thought to have arisen
  from single genes via successive duplication
  events
  
• duplicated genes follow separate evolutionary
  pathways new specificities evolve through
  variation adaptation
  
• Such complexity presents real challenges for
  sequence analysis

27
Challenges for sequence analysis

• Much of the challenge is in getting the biology
  right
  
• complicated by orthology vs paralogy
• Following a db search, it may be unclear how much
  functional annotation can be legitimately
  inherited by a query
  
• source of numerous annotation errors in dbs
• propagation could lead to an error catastrophe
• Further complications result from the modular
  nature of proteins
  
• modules are autonomous folding units, used as
  protein building blocks - like Lego bricks, they
  can confer a variety of functions on the parent
  protein, either by multiple combinations of the
  same module, or via different modules to form
  mosaics
• Automatic systems dont distinguish orthologues
  from paralogues dont consider the modular
  nature of proteins

28
(No Transcript)
29

• Monkeys are exploited in different Goldberg
  machines, where they perform different functions
  - here, we couldnt predict a monkey in that
  spot, even with total knowledge of the rest of
  the machine
• Similarity searches are just like this
• identifying the presence of a module tells little
  of the function of the complete system
  
• knowing most components of a mosaic, we cant
  predict a missing one
  
• modules (monkeys) in different proteins dont
  always perform exactly the same function

30
The Midnight Zone

• Identifying evolutionary links between sequences
  is useful
  
• this often implies a shared function
• Arguably, prediction of function from sequence is
  of more immediate value than the prediction of
  structure
  
• However, between distantly-related proteins,
  structure is more conserved than the underlying
  sequences
  
• thus, some relationships are only apparent at the
  structural level
  
• Such relationships can't be detected by even the
  most sensitive sequence comparison methods
  
• the region of identity where sequence comparisons
  fail completely to detect structural similarity
  is the Midnight Zone - there is thus a
  theoretical limit to the effectiveness of
  sequence analysis methods

31
Ground rules for bioinformatics

• Don't always believe what programs tell you
• they're often misleading sometimes wrong!
• Don't always believe what databases tell you
• they're often misleading sometimes wrong!
• Don't always believe what lecturers tell you
• they're often misleading sometimes wrong!
• In short, don't be a naive user
• when computers are applied to biology, it is
  vital to understand the difference between
  mathematical biological significance
  
• computers dont do biology
• they do sums
• quickly!

32
Significance

• Appreciating that mathematical biological
  significance are different is crucial
  
• It is especially important in understanding the
  limitations of
  
• database search algorithms
• multiple sequence alignment algorithms
• pattern recognition techniques
• functional site structure prediction tools
• Contrary to popular opinion, there is currently
  still
  
• no biologically-reliable automatic multiple
  alignment algorithm
  
• no infallible pattern-recognition technique
• no reliable gene, function or structure
  prediction algorithm

33
(No Transcript)
34
(No Transcript)
35
Biological Databases

• Overview
• Primary data sources
• GenBank, SWISS-PROT TrEMBL
• Composite sequence databases
• NRDB, OWL, SPTrEMBL
• Secondary pattern databases
• PROSITE, PRINTS, Profiles, Pfam, BLOCKS,
  IDENTIFY
  
• Composite pattern databases
• BLOCKS, InterPro

36
Primary sequence databases

• In the '80s, when sequences started to
  accumulate, several labs saw advantages to
  establishing central repositories
  
• trouble is, many labs thought the same made
  their own
  
• Nucleic Protein
• EMBL SWISS-PROT
• GenBank PIR
• DDBJ MIPS
• TrEMBL
• NRL-3D
• The proliferation of dbs causes problems
• do they have the same format? Which is the most
  accurate? The most up-to-date? The most
  comprehensive? Which should we use?

37
Composite sequence databases

• A solution to proliferating dbs is to compile a
  composite
  
• these render searches very efficient, especially
  if non-redundant
  
• Trouble is, there are now several composites,
  each with their own format redundancy criteria
  
• NRDB OWL SPTrEMBL
• PDB SWISS-PROT SWISS-PROT
• SWISS-PROT PIR TrEMBL
• PIR GenBank
• GenPept NRL-3D
• GenPept updates
• NRDB SPTrEMBL are non-identical, not
  non-redundant
  
• but which is best? Which the most comprehensive?
  The most up-to-date? Which should we use?

38
Secondary pattern databases

• As well as 1' resources, there are also many 2'
  pattern dbs derived from them
  
• trouble is, they use different 1' sources
  different analysis methods, all have different
  formats!
  
• But it isn't all bad - SWISS-PROT is emerging as
  a standard, most of the 2' dbs use it as their
  basis
  
• PROSITE SWISS-PROT Regular expressions
  (patterns)
  
• PRINTS SWISS-PROT/TrEMBL Aligned motifs
  (fingerprints)
  
• Pfam SWISS-PROT/TrEMBL Hidden Markov Models
  (HMMs)
  
• Profiles SWISS-PROT Weight matrices (profiles)
• BLOCKS PRINTS/InterPro/Domo Weighted motifs
  (blocks)
  
• IDENTIFY PRINTS/InterPro Permissive regular
  expressions

39
Why create pattern databases?

• Arise from the need to make more specific
  functional diagnoses than are possible by just
  searching the 1's
  
• Theyre built on the principle that homologous
  sequences may be gathered into alignments, within
  which are regions (motifs) that show little
  variation
• these usually reflect vital structural or
  functional roles
  
• Motifs are exploited in different ways to build
  diagnostic patterns for protein families
  
• new sequences can be searched against dbs of such
  patterns to see if they can be assigned to known
  families
  
• hence they offer a fast track to the inference of
  function

40
What's in a sequence?
41
Single motif methods
Fuzzy regex (IDENTIFY)
Exact regex (PROSITE)
Full domain alignment methods
Profiles (PROFILE LIBRARY)
HMMs (Pfam)
Identity matrices (PRINTS)
Multiple motif methods
Weight matrices (BLOCKS)
42
The challenge of family analysis
43
Know your family
44
The problem with domains
45
PROSITE

• This was the first pattern database
• protein families characterised by single motifs
• Sequence information in motifs is reduced to
  consensus or regular expressions the seed
  pattern used to search SP
  
• results are checked by hand to determine true
  false matches
  
• noisy patterns are revised to achieve optimal
  results
  
• Some families cant be characterised by single
  motifs
  
• here, additional patterns are created refined
  until an optimal set of patterns is achieved that
  capture most or all of the family
  
• results are then manually annotated for inclusion
  in the db

46
(No Transcript)
47
(No Transcript)
48
PRINTS

• Most protein families are characterised by 1
  motif
  
• it is sensible to use them all to build a
  diagnostic signature
  
• This is the principle of fingerprints
• these offer improved diagnostic reliability by
  virtue of the biological context provided by
  motif neighbours
  
• Motifs are excised from alignments by hand
  encoded as ungapped, unweighted local alignments
  
• residue information is augmented via iterative
  searches
  
• sequences matching all motifs that weren't in the
  original alignment are added to the motifs, the
  db searched again
  
• The process is repeated until convergence
• results are manually annotated prior to inclusion
  in the db

49
(No Transcript)
50
SUMMARY INFORMATION 37 codes involving 8 el
ements 0 codes involving 7 elements
0 codes involving 6 elements
0 codes involving 5 elements
0 codes involving 4 elements
1 codes involving 3 elements
0 codes involving 2 elements
COMPOSITE FINGERPRINT INDEX 8 37 37
37 37 37 37 37 37
7 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0
3 1 0 0 0 1 1 0 0
2 0 0 0 0 0 0 0 0
------------------------------------------
1 2 3 4 5 6 7 8
True positives.. PRIO_COLGU PRIO_MACFA PRIO_C
EREL PRIO_ODOHE PRIO_GORGO PRIO_PANTR PRIO_
HUMAN O46648 PRIO_SHEEP PRIO_CALJA PRIO_BOV
IN PRP2_BOVIN PRIO_ATEPA PRIO_SAISC PRIO_PR
EFR PRIO_PONPY O75942 PRIO_CAPHI PRIO_
CEBAP PRIO_CAMDR PRIO_FELCA PRP1_TRAST PRI
O_RABIT PRP2_TRAST PRIO_PIG PRIO_CANFA P
RIO_CRIGR PRIO_CRIMI Q15216 PRIO_RAT
PRIO_CERAE PRIO_MUSPF PRIO_MUSVI PRIO_MESAU
PRIO_MOUSE O46593 PRIO_TRIVU Subfamily Co
des involving 3 elements Subfamily True positive
s.. PRIO_CHICK
51
(No Transcript)
52
Profiles Pfam

• An alternative to motif-based methods exploits
  regions between motifs, which contain valuable
  information
  
• the full alignment effectively becomes the
  discriminator
  
• A complex scoring scheme allowing for
  substitutions INDELs is used to create
  family-specific profiles
  
• These profiles can be used to detect distant
  relation-ships, where only few residues are
  conserved
  
• this is the basis of the Profile library
• In an extension of this approach, alignments are
  encoded as probabilistic models termed HMMs
  
• this is the basis of Pfam

53
BLOCKS IDENTIFY

• There are advantages to storing motifs in a raw
  form
  
• no information is lost
• different scoring schemes may be used to confer
  different diagnostic potentials on the same data
  
• Additional pattern databases have arisen in this
  way
  
• BLOCKS - processed PROSITE families automatically
  (BLOCKS includes many other sources)
  
• BLOCKS-format PRINTS - PRINTS motifs with BLOCKS
  scoring
  
• IDENTIFY - creates fuzzy expressions from PRINTS
  InterPro
  
• These databases are derived fully automatically,
  hence offer
  
• no family annotation (they link back to PRINTS
  InterPro)
  
• no further family coverage

54
Composite pattern databases

• To simplify sequence analysis, the pattern
  databases are being integrated to create a
  unified protein family resource - InterPro
  
• this is a central annotation resource (derived
  from PRINTS PROSITE documentation), with
  pointers to its satellite databases
  
• release 3.0 contains 3591 entries
• current partners are PRINTS, PROSITE, Profiles,
  Pfam ProDom
  
• future partners will include SMART, TigrFam
  hopefully others (BLOCKS, MetaFam, etc.)
  
• lags behind its sources

55
(No Transcript)
56
(No Transcript)
57
Pattern Recognition

• Overview
• Pattern recognition methods
• regular expressions, fingerprints, blocks,
  profiles HMMs
  
• Which method is best?

58
Pattern recognition methods

• These methods classify proteins into families
• the basis of the methods is multiple sequence
  alignment
  
• They depend on developing a representation of
  conserved elements of alignments that may be
  diagnostic of structure or function, whether
  from
• homologous sequence families
• sequences that share some structural/functional
  domains
  

59
Determining significance of database matches

• When searching a db, the challenge for analysis
  methods is to determine if matches are related
  (true-positive) or unrelated (true-negative)
  
• At a given scoring threshold, it is likely that
  unrelated sequences will be matched erroneously
  (false-positives) some correct matches will be
  missed (false-negative)
• The aim is to improve the resolution between the
  curves - in the overlap, it is difficult or
  impossible to establish if matches are
  significant
• Different methods tackle this problem in
  different ways

60
Regular expressions/patterns

• These are derived from single conserved regions,
  which are reduced to consensus expressions for db
  searches
  
• they are minimal expressions, so sequence
  information is lost
  
• the more divergent the sequences used, the more
  fuzzy poorly discriminating the pattern
  becomes
  
• Alignment Pattern
• GAVDFIALCDRYF
• GPIDFVCFCERFY G-X-IV-DE-F-IVL-X2-C-DE-R-
  FY2
  
• GRVEFLNRCDRYY
• Patterns do not tolerate similarity
• sequences either match or not, regardless of how
  similar they are
  
• matching is a binary on-off event frequently
  misses true matches
  
• single-motif methods are very hit-or-miss - how
  do you know if you've encoded the best region?

61
In the beginning was PROSITE

• G_PROTEIN_RECEPTOR PATTERN
• PS00237
• G-protein coupled receptor signature
• GSTALIVMYWC-GSTANCPDE-EDPKRH-X(2)-LIVMNQGA
  -
  
• X(2)-LIVMFT-GSTANC-LIVMFYWSTAC-DENH-R
• /TOTAL919(919)/POS869(869)/FALSE_POS50(50)/F
  ALSE_NEG70
  
• /PARTIAL49 UNKNOWN0(0)

• This represents an apparent 18 error rate
• the actual rate is probably higher
• Thus, a match to a pattern is not necessarily
  true
  
• a mis-match is not necessarily false!
• False-negatives are a fundamental limitation to
  this type of pattern matching
  
• if you don't know what you're looking for, you'll
  never know you missed it!

62
R-Y-x-DT-W-x-LIVM-ST-T-P-LIVM(3)
63
(No Transcript)
64
Regular expressions/rules

• Regular expression patterns are most effective
  when applied to highly-conserved, family-specific
  motifs
  
• It is often possible to identify, shorter generic
  patterns that are characteristic of common
  functional sites
  
• Functional site Rule
• N-glycosylation N-P-ST-P
• Protein kinase C phosphorylation ST-X-RK
• Casein kinase II phosphorylation ST-X2-DE
• Such features result from convergence to a common
  property
  
• glycosylation sites, phosphorylation sites, etc.
• They cannot be used for family diagnosis don't
  discriminate
  
• they can only be used to suggest whether a
  certain functional site might exist (which must
  then be tested by experiment)
  
• such patterns are termed rules

65
Diagnostic limitations of short motifs

• Consider the sequence motif Asp-Ala-Val-Ile-Asp
  (DAVID)
  
• results of db searching for such a sequence will
  differ, depending on whether we search for exact
  or permissive fuzzy matches
  
• Pattern Matches
• D-A-V-I-D 71 (99)
• D-A-V-I-DEQN 252
• DEQN-A-V-I-DEQN 925
• DEQN-A-VLI-I-DEQN 2,739
• DEQN-AG-VLI-VLI-DEQN 51,506
• D-A-V-E 1,088 (1,493)
• (number of matches in OWL29.6 ( OWL31.1))
• Use of fuzzy regular expressions has the
  potential advantage of being able to recognise
  more distant relationships
  
• the inherent disadvantage that more matches
  will be made by chance, making it difficult to
  separate out true matches from noise

66
Residue groups for fuzzy patterns

• It is possible to assign residues to groups
  corresponding to various biochemical properties -
  e.g., charge size
  
• using such groups to create fuzzy expressions
  theoretically ensures that resulting motifs have
  sensible biochemical interpretations
  
• small Ala, Gly
• small hydroxyl Ser, Thr
• basic His, Lys, Arg
• aromatic Phe, Tyr, Trp
• aliphatic Val, Leu, Ile, Met
• acidic/amide Asp, Glu, Asn, Gln
• small/polar Ala, Gly, Ser, Thr, Pro
• This is more flexible than exact regular
  expression matching
  
• but the inherent permissiveness of the fuzzy
  approach brings an inevitable signal-to-noise
  trade-off

67
Fingerprints

• Fingerprints are groups of motifs excised from
  alignments used for iterative db searching
  
• no weighting scheme is used
• searches depend only on residue frequencies
• resulting scoring matrices are thus sparse
• Each motif trawls the database independently
• search results are correlated to determine which
  sequences match all the motifs which match only
  partially
  
• no information is thrown away
• Iteration refines the fingerprint increases its
  potency
  
• fingerprints are diagnostically more powerful
  than regular expressions

68
TM domain
TM domain
69
loop region
70
A fingerprinting overview
71

• T C A G N S P F L Y H Q V K D E
  I W R M B X Z
  
• 0 0 2 0 0 0 0 0 0 7 0 0 1 0 0 0
  0 2 0 0 0 0 0
  
• 0 0 3 0 0 0 0 0 2 0 0 0 4 0 0 0
  3 0 0 0 0 0 0
  
• 6 0 0 0 0 0 0 6 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0
  
• 1 0 1 1 0 3 0 0 1 0 0 0 3 0 0 0
  0 0 0 2 0 0 0
  
• 2 0 1 1 0 0 0 0 0 0 0 3 0 4 0 0
  0 0 1 0 0 0 0
  
• 4 0 1 0 0 1 0 2 0 1 3 0 0 0 0 0
  0 0 0 0 0 0 0
  
• 0 0 0 0 0 0 0 0 0 0 0 0 0 12 0 0
  0 0 0 0 0 0 0
  
• 0 0 0 0 0 6 0 0 0 0 0 0 0 6 0 0
  0 0 0 0 0 0 0
  
• 0 0 0 0 0 0 0 0 12 0 0 0 0 0 0 0
  0 0 0 0 0 0 0
  
• 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0
  0 0 10 0 0 0 0
  
• 9 0 0 0 0 0 0 0 0 0 2 1 0 0 0 0
  0 0 0 0 0 0 0
  
• 0 0 0 0 0 0 12 0 0 0 0 0 0 0 0 0
  0 0 0 0 0 0 0
  
• 0 0 4 0 0 2 0 0 6 0 0 0 0 0 0 0
  0 0 0 0 0 0 0
  
• (b)
• T C A G N S P F L Y H Q V K D E
  I W R M B X Z
  
• 0 0 4 0 0 0 0 8 4 34 0 0 15 0 0 0
  1 7 0 0 0 0 0
  
• 0 4 15 0 0 0 0 0 7 0 0 0 37 0 0 0
  10 0 0 0 0 0 0
  
• 50 0 0 0 0 3 0 18 0 0 0 0 0 0 0 0
  0 0 0 2 0 0 0
  

• YVTVQHKKLRTPL
• YVTVQHKKLRTPL
• YVTVQHKKLRTPL
• AATMKFKKLRHPL
• AATMKFKKLRHPL
• YIFATTKSLRTPA
• VATLRYKKLRQPL
• YIFGGTKSLRTPA
• WVFSAAKSLRTPS
• WIFSTSKSLRTPS
• YLFSKTKSLQTPA
• YLFTKTKSLQTPA
• (a)
• Key
• (a) motif, with 3 conserved positions
• (b) corresponding frequency matrix
• (c) same matrix, but after 3 iterations
• (d) same matrix, with PAM250 weighting

72
Fingerprint visualisation

• Full potency of fingerprinting is gained from the
  mutual context provided by motif neighbours
  
• Important, as it inherently implies a biological
  context to motifs matched in the correct order,
  with appropriate distances between them
  
• results are thus biologically more meaningful
  than those from single motifs
  
• Allows sequence identification even when parts of
  the fingerprint are absent
  
• such matches are best visualised graphically

73
(No Transcript)
74
(No Transcript)
75
Blocks

• Blocks are groups of motifs derived automatically
  from families identified in PRINTS InterPro
  
• sequences are aligned automatically motifs are
  automatically identified by searching for spaced
  residue triplets (e.g., AxxxVxxC)
  
• a block score is calculated using the BLOSUM62
  matrix
  
• validity of blocks is confirmed with a 2nd
  motif-finding algorithm
  
• blocks found by both methods are considered
  reliable
  
• Sequences within motifs are clustered to reduce
  contributions to residue frequencies from sets of
  closely-related sequences
  
• each cluster is treated as a single sequence
  given a score that gives a measure of its
  relatedness
  
• the higher the weight, the more dissimilar the
  segment from others in the block, the most
  distant being given a score of 100
  
• segments

76
(No Transcript)
77
(No Transcript)
78
Profiles

• Profiles are scoring tables derived from full
  alignments
  
• these define which residues are allowed at given
  positions
  
• which positions are conserved which degenerate
• which positions, or regions, can tolerate
  insertions
  
• the scoring system is intricate, may include
  evolutionary weights, results from structural
  studies, data implicit in the alignment
  
• variable penalties are specified to weight
  against INDELs occurring in core 2' structure
  elements
  
• Within a profile, the I M fields contain
  position-specific scores for insert match
  positions
  
• in conserved regions, INDELs aren't totally
  forbidden, but are strongly impeded by large
  penalties defined in the DEFAULT field
  
• these are superseded by more permissive values in
  gapped regions
  
• the inherent complexity of profiles renders them
  highly potent discriminators, but they are
  time-consuming to derive

79
(No Transcript)
80
(No Transcript)
81
Hidden Markov Models

• HMMs are similar in concept to profiles
• they are probabilistic models consisting of
  inter-connecting states
  
• essentially, linear chains of match, delete or
  insert states
  
• Match states are assigned to conserved columns in
  an alignment
  
• insert states allow for insertions relative to
  match states
  
• delete states allow match positions to be
  skipped
  
• thus, building an HMM requires each position in
  an alignment to be assigned to match, delete or
  insert states
  
• HMMs usually perform well, but can be
  over-trained
  
• they may also suffer if created from automatic
  iterative processes
  
• if it once accepts a false match, an HMM becomes
  corrupt

82
An HMM
C
L
Y
E
C
L
W
D
83
Which method is best?

• The range of methods available leads to familiar
  problems
  
• which should we use?
• which is the most reliable?
• which is the most comprehensive?
• None of the pattern-recognition techniques is
  infallible
  
• each has its optimum area of application
• None of the resulting pattern databases is
  complete
  
• none is the best
• bearing in mind the diagnostic strengths
  weaknesses of the different approaches, keeping
  biological significance in mind, the best
  strategy is to use them all

84
Current status of pattern databases

• PROSITE (SIB) - 1034 entries
• single motifs (regexs) - best with small highly
  conserved sites
  
• Profile library (ISREC) - 300 entries
• weight matrices - good with divergent domains
  superfamilies
  
• PRINTS (Manchester) - 1500 entries
• multiple motifs (fingerprints) - best for
  families and sub-families
  
• Pfam (Sanger Centre) - 2727 entries
• HMMs - good with divergent domains
  superfamilies
  
• InterPro (EBI) - 3591 entries
• derived from PRINTS, PROSITE, Profiles, Pfam,
  ProDom, etc.
  
• BLOCKS (FHCRC) - 2433 entries
• multiple motifs (derived from PRINTS, InterPro,
  Domo etc.)
  
• IDENTIFY (Stanford)
• permissive regexs (derived from PRINTS InterPro)

85
Tools for predicting protein function from
sequence
86
Building a search protocol

• Overview
• The usual starting point
• searching the primary data sources
• NRDB, SPTR, etc.
• Pattern recognition methods
• searching the secondary sources
• patterns, profiles, blocks, fingerprints HMMs
• Estimating significance
• when do we believe a result?

87
A practical approach

• A central goal is to predict protein function
  from sequence
  
• Given a newly-determined sequence, we want to
  know
  
• what is my protein?
• to what family does it belong?
• what is its function?
• how can we explain its function in structural
  terms?
  
• By searching pattern dbs fold libraries, we may
  recognise patterns that allow us to infer
  relationships with previously-characterised
  families folds
• Given the variety of dbs to search, how do we use
  them to build a sensible search protocol?

88

• Protein sequence
  database identity search
  
• e.g., for short fragments, pinpoints
  identical matches
  
• to probe - may identify correct reading
  frame
  
• Protein sequence database similarity search
• e.g., nrdb, OWL, SPSPTrEMBL - identifies
• homologues to
  probe
  
• Protein pattern database search
• e.g., PROSITE, profiles, PRINTS, BLOCKS,
• Pfam - identifies
  family relationships or pin-
  
• points key
  structural or functional sites
  
• Known structure No known
  structure
  
• Structure classification database query
  Protein fold pattern library search
  
• e.g., scop, CATH, FSSP - provides details
  e.g., threading - identifies compatible
  
• of structural class, secondary structure
  folds for the probe sequence
  
• information, ligand-binding, etc.

89
Similarity searching

• Whether or not an identity search finds a match,
  the next step is to look for similar sequences
  
• e.g., you may wish to know if a wider family
  exists
  
• The most rapid option is to use BLAST (Best Local
  Alignment Search Tool), flavours of it, or
  FastA
  
• In BLAST output, look for
• high scores with low P-values (unlikely to be
  random)
  
• clusters of high scores at the top of the hitlist
  (a family?)
  
• trends in the type of sequences matched
• To ensure a comprehensive search, identity
  similarity searches are best performed on
  composite databases
  
• e.g., NRDB, SPSP-TrEMBL

90
Ideal results show high scores low E-values
91
Why bother with pattern searches?

• Primary searches won't always allow outright
  diagnosis
  
• BLAST FASTA are not infallible
• often can't assign mathematically significant
  scores
  
• results may be complicated by modules, domains or
  compositionally-biased regions
  
• annotations of retrieved hits may be incorrect
• Pattern databases contain potent descriptors
• so, distant relationships missed by BLAST may be
  captured by one or more of the family or
  functional site distillations

92
(No Transcript)
93
(No Transcript)
94
(No Transcript)
95
(No Transcript)
96
(No Transcript)
97
(No Transcript)
98
(No Transcript)
99
(No Transcript)
100
Structural functional interpretation

• Running db searches often does little more than
  identify a protein family
  
• this only scratches the surface - we still want
  to know what our protein does what it might
  look like
  
• The first step is to examine the detailed family
  documentations in PROSITE, PRINTS InterPro
  
• these should help to elucidate the function of
  the protein
  
• The next step is to examine the fold
  classification structure summary resources
  
• e.g., SCOP, CATH PDBsum (assuming the structure
  is known)

101
(No Transcript)
102
(No Transcript)
103
(No Transcript)
104
(No Transcript)
105
(No Transcript)
106
Estimating significance

• When do we believe a result?
• A real example.....

107
(No Transcript)
108
(No Transcript)
109
(No Transcript)
110
(No Transcript)
111
(No Transcript)
112
(No Transcript)
113
(No Transcript)
114
(No Transcript)
115
(No Transcript)
116
(No Transcript)
117
(No Transcript)
118
(No Transcript)
119
Conclusions

• Gene prediction, structure function prediction
  are non-trivial
  
• structure function prediction tools are, at
  best, 70 accurate
  
• What are the lessons for sequence analysis?
• when searching for distant homologues, several
  dbs should be searched
  
• different methods provide different perspectives
• dbs arent complete their contents dont fully
  overlap
  
• The more dbs searched, the more difficult it can
  be to interpret results
  
• The more computers are involved in automating
  genome annotation, the greater the need for
  collaboration with biologists
  
• The more data we have to handle, the more
  rigorous we must be in our thinking ( writing)
  if we are to make sense of the complexities
  
• We are still a long way from having reliable
  tools for deducing protein function from
  sequence
  
• but with the right approach, there is hope

120
The right approach risks being obscured by other
issues...