Protein Structure Alignment

1
Protein Structure Alignment

• William R. Taylor and Christine A. Orengo
• J Mol. Biol. (1989) 208. 1-22

Presented byJie BaoDept of Computer
Science Iowa State Universitybaojie_at_cs.iastate.ed
u http//www.cs.iastate.edu/baojie
2
Abstract

• Focus of this research find a good similarity
  score for two residues in the structures being
  compared , and then fo the matching
• The score is based on distance plot analysis
• Matching is done by dynamic programming
• Advantage
• Insensitive to insertions and deletions
• No initial seeding is needed
• Test it on samples
• Globins, Calcium-binding proteins ,Rhodanese,
  Immunoglobulin domains, Plastocyanin/azurin,
  Lysozyme

3
Outline

• Problem
• Methods
• Result
• Discussion

4
Whats the problem

• Align two sequence based on their structure
  position

5
Existing Methods

• Least-squares Matthews Rossmann 1985
• Limitation 1 equivalence of positions must be
  established before the superposition is preformed
• Limitation 2 the displacement of subdomain
  within one structure can result in a poor overall
  fit between topologically equivalent structures.
• Search in rotational space Rossmann etc.
  1973-1977
• Comparing all pairs of structural fragments
• Both of them are computationally demanding and
  the latter is sensitive to insertion/deletions

6
Outline

• Problem
• Methods
• Result
• Discussion

7
Dynamic Programming- Basic sequence alignment

• Sequence AlignmentAADADEFGHAADCDEAGH
• Identical pair 2insertion gap penalty g
  1Window 4
• Sij Dij max Si1,j1 max Sk,j1 g
  , kgti2 max Si1,l g, l gt j2
• Start from lower rightEnd with upper left
• The highest score is found and its inheritance
  path is traced back

-1
2
-1
-1
2
2
8
Dynamic Programming -Basic interatomic distance
matching(1)

• Basic interatomic distance matching method
• Consider only alpha-carbon atoms of two
  structures and compare distance between them
• Similarity score s a /(Adij-Bdklb) where
  Adij,Bdkl are interatomic distances a
  limits the maximum possible score b
  preventing division by zero
• Score between i in A and k in B Sik Sum-nmn
  a/ (Adi,Im-Bdk,kmb)
• Based on the score, standard dynamic programming
  algorithm can used for alignment of positions

9
Dynamic Programming -Basic interatomic distance
matching(2)

• Explanation of the position similarity score

B
A
j
dij
in
kn
i-n
k-n
k
i
The overall score is given by the sum of
individual distance comparisons
10
Dynamic Programming -Structural Environment
matching

• The basic method is adequate for matching local
  structures
• but it was disrupted if the range of comparison
  (-n to n) spanned an insertion/deletion
  discontinuity
• Do DP on Lower level of distance comparison to
  produce a best equivalence of position between
  two environments.Sik max a/ (Adij-Bdklb)
• a 50, b 5

11
Dynamic Programming -Alignment dependency
between levels

• Every comparison of the last step will produce an
  alignment of matched distances which is alignment
  of the two sequence too.
• The values along he trace-back path in the lower
  level matrix were accumulated in corresponding
  elements in the higher level matrix
• Apply cutoff on S to prevent the excessive
  accumulation of background noise sqrt200N
  , N is the length of shorter sequence numbers
  lt cutoff are ignored
• Advantage large contribution are made only to he
  upper matrix for regions that match well in the
  lower comparison.
• Weakness solely based on interatomic distances
  and has the limitation that similar distances
  achieve a high score even when these are between
  pairs of atoms that might be in completely
  different relative directions

12
Dynamic Programming -Alignment dependency
between levels (2)

• The method is used at 2 levels
• First to find the best equivalence of distances
  for the 2 residues being compares
• Then at a higher level to find the best
  equivalence of residues within 2 sequence
• Gap penalty of 5 is applied

13
Dynamic Programming -Vector comparison method

• Solution compare interatomic vector rather than
  simple distances
• The vector s is defined in the local frame of
  reference for every residue.
• The Similarity score is changed to s a
  /((AVij-BVkl)2b)

• Prepare the X-Y-Z
• X-axis was defined by the n-c vector
• A tentative Y-axis by the Cbeta-H vector
• Z-axis was their mutual perpendicular vector
• Y-axis was redefined as perpendicular to X and Z.

j
dij
in
i
i-n
Vij
14
Dynamic Programming -the final method

• The vector-based method uses 3-d distances
• Higher dimensions also could incorporate any data
  that can be defined at the residue level.
• Nature of the amino acid can be usedSik max
  (wDRiRka)/ (Adij-Bdklb) DXY is the value
  in the Dayhoff matrix for the exchange of Amino
  Acids of type X and Y w is he weight, default
  it 1,0 a 40, b 2

15
Implementation

• Implemented in SSAP(Structure and squence
  alignment program)Written in C , run on VAX-II
  under VMSSeparate FROTRAN program was used to
  prepare dataData from PDB

CPU Time of run(mintues) SeqLen, Win Time 50
20 5.2 30 10.3 100
30 36.0 60 121.0150
20 36.2 40 130.4
60 269.6
16
Outline

• Problem
• Methods
• Result
• Discussion

17
Data Set
PDB ID used in the paper/ Current ID
18
Results GlobinsMBN, HHB
This Work
4HHB haemoglobinAlpha-chain Beta-chain
5MBN myoglobin
LC

• Compared with Lesk Chothia 1980, conventional
  superposition
• Result
• alpha and beta chains of 4HHB are the most
  similar
• s(MBN, beta) gt s(MBN, alpha)

Structural comparison of the globins
19
Results Calcium-binding proteinsParvalbumin,
CIB

• Calcium-binding proteins contains two motifs
  (helices) C D and (eponymous) E F
• Parvalbumin (helices) also contains A B
• The algorithm aligned the correct ion binding
  motifs in both structures, ignoring the redundant
  motif in parvalbumin (AB)compared with
  Cariepy Hodges 1983

20
Results - Rhodanese

• Align the two halves of rhodanese
• Compared with Ploegman 1978 least sequence, the
  result is identical in all but minor aspects
• Graphic observation reveals this work is more
  plausible

1RHD, with two similar alternatingbeta/alpha-type
domains
21
Results Immunoglobulin domains
3FABAntigen-binding fragment
1FC1 Constant fragment
22
Results Immunoglobulin domains
Structure of Immunoglobulin

• Compares each domain against every other
• Produce correct alignment in 9 of 15 comparsions
• Compared with Lesk Chothia 1982

Immunoglobulin heavy H Immunoglobulin Light L
include 6 all-beta domains of two types
constant (C) and variable (V)
23
Results - Lysozyme

• Lysozyme (hen egg-white) 6LYZ
• Lysozyme (T4) 1LZM/2LZM
• Compared with Rossmann Argos 1976, 1977 Matthews
  1981
• Found the first common helix to be displaced by
  one residue form both puvlished comparsions
• Aligement of the initial beta-strand agrees with
  that of R A 1976
• The reminder of the alignment is the same for
  both methods, except for trivial displacements at
  the fringes of equivalent blocks

6LYZ
2LZM
24
Results Plastocyanin/azurin

• Plastocyanin 3PCY
• Azurin 1AZU
• Compared with Chothia Lesk 1982 and Adman 1985
• Except for some minor insertions and deletions on
  the fringes, the alignments agrees with others.

3PCY
1AZU
25
Outline

• Problem
• Methods
• Result
• Discussion

26
Discussion

• The results demonstrated that this method produce
  is equal in quality , and in some cased superior
  , to those reported.
• This method is insensitive to the displacement of
  equivalent substructures
• Future work develop statistical criteria for
  evaluating the significance of structural
  comparisons.
• And it can be extended beyond residue level , eg.
  Secondary structure