Protein Structure Prediction

1
Protein Structure Prediction

• Shandar Ahmad
• Kyushu Institute of Technology,
• Iizuka 820 8502,
• Fukuoka-ken, Japan
• shandar_at_bse.kyutech.ac.jp

2
Secondary structure The basic unit of protein
structure

• Protein structures are stabilized by Hydrogen
  bonds between atoms of the amino acid sequence.
• There is a choice of hydrogen bond partner for
  each residue.
• The pattern of hydrogen bond pairing determines
  secondary structure.

3
Types of secondary structure

• Eight types of secondary structures have been
  defined by Kabsch and Sander in DSSP (Dictionary
  of secondary structures in proteins). They are
• Alpha helix (H) 5. Pi-helix (I)
• Isolated beta bridge (B) 6. Turn (T)
• Extended Beta (E) 7. Bend (S)
• 3-10 helix (G) 8. Coil (C)

4
Alpha helixHydrogen bond is formed between nth
and (n4)th residues
5
Beta strand (E), part of beta ladder
6
Beta strand (E) cont..
7
Turn structure (T)
8
Other helices
9
Bend Conformation

• Bend is the caused by interactions with other
  parts of protein.
• Proline introduces bend due to conformational
  constraints.
• Water molecules cause bend to maximize CO
  exposure to water.

10
Some structural domains in proteins

• Movies data

11
Methods to get secondary structure from
experimentally known structures

• DSSP is the most commonly used program to
  calculate secondary structure of proteins.
• DSSP also provides a database to get Sec
  structure by searching their PDB codes.
• Database and programs can be accessed at
• http//www.cmbi.kun.nl/gv/dssp/
• Program can be downloaded for local calculations.
• PDB files also contain secondary structures in
  their headers, but only the broad details.

12
Prediction of secondary structure

• Older methods
• Chou and fasman method (1974)
• The Chou-Fasman method of secondary structure
  prediction depends on assigning a set of
  prediction values to a residue and then applying
  a simple algorithm to those numbers.
• For example
• p(t) f(j)f(j1)f(j2)f(j3) See next table
• Online predictions http//fasta.bioch.virginia.ed
  u/fasta_www/chofas.htm
• Typical success rate of prediction is of 50

13
Name P(a) P(b) P(turn) f(i) f(i1) f(i2)
f(i3) Alanine 142 83 66 0.06 0.076 0.035
0.058 Arginine 98 93 95 0.070 0.106 0.099
0.085 Aspartic Acid 101 54 146 0.147
0.110 0.179 0.081 Asparagine 67 89 156
0.161 0.083 0.191 0.091 Cysteine 70 119
119 0.149 0.050 0.117 0.128 Glutamic Acid
151 37 74 0.056 0.060 0.077 0.064
Glutamine 111 110 98 0.074 0.098 0.037
0.098 Glycine 57 75 156 0.102 0.085 0.190
0.152 Histidine 100 87 95 0.140 0.047
0.093 0.054 Isoleucine 108 160 47 0.043
0.034 0.013 0.056 Leucine 121 130 59
0.061 0.025 0.036 0.070 Lysine 114 74 101
0.055 0.115 0.072 0.095 Methionine 145 105
60 0.068 0.082 0.014 0.055 Phenylalanine
113 138 60 0.059 0.041 0.065 0.065 Proline
57 55 152 0.102 0.301 0.034 0.068 Serine
77 75 143 0.120 0.139 0.125 0.106
Threonine 83 119 96 0.086 0.108 0.065
0.079 Tryptophan 108 137 96 0.077 0.013
0.064 0.167 Tyrosine 69 147 114 0.082
0.065 0.114 0.125 Valine 106 170 50 0.062
0.048 0.028 0.053
Download
Source http//prowl.rockefeller.edu/aainfo/chou.h
tm
14
Further improvements

• 1978 Garnier improved the method by using
  statistically significant pair-wise interactions
  as a determinant of the statistical significance.
  This improved the success rate to 62
• 1993 Levin improved the prediction level by using
  multiple sequence alignments.
• The reasoning is as follows.
• Conserved regions in a multiple sequence
  alignment provides a strong evolutionary
  indicator of a role in the function of the
  protein.
• Those regions are also likely to have conserved
  structure, including secondary structure and
  strengthen the prediction by their joint
  propensities.
• This improved the success rate to 69.

15
Neural network based methods

• In 1993, Qian Sejnowski and Holey and Karplus
  introduced first neural network based method.
• Sequence information is sent to the neural
  network and the output is classified as helix,
  beta, or other secondary structures
• See next figure

16
(No Transcript)
17
Encoding the amino acids

• Amino acid residues are coded as 21 bit binary
  vectors.
• For predicting secondary structure of a residue,
  this information about residue and its neighbour
  is sent to the neural network.
• For known structural data, network is trained and
  validated.

18
Other advanced methods of prediction

• PHD Predict Protein
• 1994 Rost and Sander combined neural networks
  with multiple sequence alignments. The success
  rate is 72.
• http//www.embl-heidelberg.de/predictprotein/predi
  ctprotein.html
• Jpred (Cuff and Barton)
• http//www.compbio.dundee.ac.uk/www-jpred/
• Predator (Frishman D, Argos P )
• http//npsa-pbil.ibcp.fr/cgi-bin/npsa_automat.pl?p
  age/NPSA/npsa_preda.html
• PSIPRED (DT Jones)
• http//bioinf.cs.ucl.ac.uk/psipred/

19
Solvent accessibility of amino acid residues

• This is another important property of amino acids
  in proteins, which we want to predict.
• Solvent accessibility is defined as the Area
  around the surface of a residue, which is exposed
  to water (or any solvent).
• Higher solvent accessibility or accessible
  surface area (ASA) indicates greater chance of
  interactions with DNA, Ligands etc. and being in
  the active sites.

20
Accessible surface area or solvent accessibility
21
Relative solvent accessibility

• Total ASA of an amino acid is normalised to
  percentage scale.
• Scaling is different for 20 types of amino acids.
• ASA of extended state (Gly-X-Gly or Ala-X-Ala)
  are used for scaling.
• Sometimes this relative ASA is used to say if a
  residue is exposed or buried. E.g. if ASA is more
  than 25, it may be called exposed, and if less
  than 25 it may be called buried.
• Different values of threshold (other than 25)
  are used by different people.

22
Solvent accessibility prediction methods

• PHD server described above gives ASA predictions
  also. It devided residues into buried and exposed
  categories at 16 threshold and gives a
  prediction.
• Real value prediction method based on neural
  network was developed by us (Ahmad and Sarai
  2003), which can make a prediction upto 18 mean
  absolute error (better than any other prediction
  method available).
• http//gibk26.bse.kyutech.ac.jp/shandar/netasa/rv
  p-net/
• This is the only server which also provides
  graphical outputs.

23
A graphical prediction of Solvent accessibility
by RVP-Net. Shandar Ahmad and Akinori Sarai, 2003
24
Measuring prediction accuracy

• Different scales of prediction are used.
• Single residue accuracy or Qindex
• (Qhelix, Qstrand, Qcoil, Q3) gives percentage of
  residues predicted correctly as helix, strand,
  coil or for all three conformational states. The
  definition of Qindex is as follows.
• For a single conformational state
• number of residues correctly predicted in
  state i
• Qi --------------------------------------------
  ----- ----------- 100,
• number of residues observed in state
  i
• where i is either helix, strand or coil.

25
Other scores

• R Sxy / Sxx Syy Where Sxy ? (x xo)
  (y-yo)
• Sxx ? ? (x xo)2
• Syy ? ? (y yo)2
• Subscript o represents mean value of the
  corresponding variable.
• Sensitivity TP/ (TPFN)
• Specificity TN/(TNFP) (T-True, F-False,
  P-Positive, N-Negative)

26
Segment overlap SOV score

• SOV Segment OVerlap quantity measure for a
  single conformational state
• 1 SUM MINOV(S1S2)
  DELTA(S1S2)
• SOV(i) --- SUM -------------------------
  -- LEN(S1)
• N(i) SUM MAXOV(S1S2)
• S(i)
• Where
• S1 and S2 are the observed and predicted
  secondary structure segments (in state i, which
  can be either H, E or C) LEN(S1) is the number
  of residues in the segments
• S1 MINOV(S1S2) is the length of actual overlap
  of S1 and S2, i.e. the extent for which both
  segments have residues in state i, for example H
  
• MAXOV(S1S2) is the length of the total extent
  for which either of the segments S1 or S2 has a
  residue in state i DELTA(S1S2) is the integer
  value defined as being equal to the
  MIN(MAXOV(S1S2)- MINOV(S1S2)) MINOV(S1S2)
  INT(LEN(S1)/2) INT(LEN(S2)/2)
• THE SUM is taken over S, all the pairs of
  segments S1S2, where S1 and S2 have at least
  one residue in state i in common N(i) is the
  number of residues in state i

27
Higher level predictions of protein structure

• This includes prediction of
• Structure class as defined by SCOP or CATH
• Folds, as defined by SCOP and CATH
• Complete three dimensional structure.

28
Prediction of protein structure calss

• Some secondary structure prediction servers also
  predict classe e.g.
• http//www.cmpharm.ucsf.edu/jmc/pred2ary/
• (Chandonia and Karplus)
• All helix, all beta classes are easier to predict
  than a/b and ab structural classes.

29
Protein fold prediction

• Approaches to fold prediction may be classified
  into two categories
• Sequence to sequence prediction Based on getting
  the best alignments with known structures and
  predicting fold.
• Sequence to structure methods Structure is
  encoded as a sequence of residue environments.
  Score is assigned to each residue and finally
  score is added to detect the probability of a
  given fold.

30
Best fold predictors

• CASP is a biannual meeting for evaluating
  structure prediction. Following methods were
  found to be the best in 2002.
• Krzysztof Ginalski and Leszek Rychlewski
• Nucleic Acids Research, 2003, Vol. 31, No. 13
  3291-3292
• http//BioInfo.PL/Meta
• This is a metserver working on 3D-Jury method. It
  collects predictions from many servers and
  develops a consensus model based prediction.

31
(No Transcript)
32
Pcons Consensus predictor

• Earlier version of Meta server, but with slight
  difference in building the final prediction
• http//www.sbc.su.se/arne/pmodeller/

33
ROSETA predictor by Baker

• This is an ab-initio method of structure
  prediction.
• When there is no significant alignment available,
  this is the only way to predict.
• Performs better than all other predictors.
• No online predictions, but group website is here
• http//depts.washington.edu/bakerpg/highlights1.ht
  ml

34
Three dimensional structure prediction

• Methods are based on
• Ab-intio, Molecular dynamics and Monte Carlo
  methods of energy minimization.
• Comparative modelling using sequence alignments
  with known structures.
• Combination of the above two methods.

35
Some prediction methods/ predicted model databases

• Modeller
• http//www.salilab.org/modeller/modeller.html
• SwissModel
• http//www.expasy.org/swissmod/
• FAMSBASE
• http//famsbase.bio.nagoya-u.ac.jp/famsbase/
• GenTHREADER and PSIPRED
• http//bioinf.cs.ucl.ac.uk/psipred/
• UCLA/DOE Fold Server (Includes DASEY, in case no
  alignments are found).
• http//fold.doe-mbi.ucla.edu/

36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)