Transcript Folie 1 - FLI

3D Structures of Biological Macromolecules
Part 5: Protein Structure Prediction
Jürgen Sühnel
[email protected]
Leibniz Institute for Age Research, Fritz Lipmann Institute,
Jena Centre for Bioinformatics
Jena / Germany
Supplementary Material: http://www.fli-leibniz.de/www_bioc/3D/
PDB Content Growth
698
4187
5421
7069
Start
1993:
2003:
2005:
2008:
= 1588 structures (~ 2 structures per day)
= 23674 structures (~ 11 structures per day)
= 34325 structures (~ 15 structures per day)
= 55063 structures (~ 19 structures per day)
(experimental structures only)
PDB Content Statistics
May 12, 2009
SwissProt/TrEMBL: Growth Rate
15-Jan-2008
Swiss-Prot/TrEMBL: Amino Acid Composition
Swiss-Prot
TrEMBL
15-Jan-2008
Structural Genomics
Structural genomics consists in the determination of the three dimensional structure of all proteins of a
given organism, by experimental methods such as X-ray crystallography, NMR spectroscopy
or computational approaches such as homology modelling.
As opposed to traditional structural biology, the determination of a protein structure through
a structural genomics effort often (but not always) comes before anything is known regarding
the protein function. This raises new challenges in structural bioinformatics, i.e. determining protein function
from its 3D structure.
One of the important aspects of structural genomics is the emphasis on high throughput determination of
protein structures. This is performed in dedicated centers of structural genomics.
While most structural biologists pursue structures of individual proteins or protein groups, specialists in
structural genomics pursue structures of proteins on a genome wide scale. This implies large scale
cloning, expression and purification. One main advantage of this approach is economy of scale.
On the other hand, the scientific value of some resultant structures is at times questioned.
en.wikipedia.org/wiki/Structural_genomics
Structural Genomics
Protein Structure Prediction
clickable map
http://speedy.embl-heidelberg.de/gtsp/flowchart2.html
Protein Structure Prediction
A Good Protein Structure
• Minimizes disallowed
torsion angles
• Maximizes number of
hydrogen bonds
• Minimizes interstitial
cavities or spaces
• Minimizes number of
“bad” contacts
• Minimizes number of
buried charges
Protein Structure Prediction – CAFASP Contest
http://www.cs.bgu.ac.il/~dfischer/CAFASP5/
Protein Structure Prediction – CASP Contest
http://predictioncenter.gc.ucdavis.edu/
Protein Structure Prediction
– Secondary structure
– 3D structure
• Modeling by homology (Comparative modeling)
• Fold recognition (Threading)
• Ab initio prediction
– Rule-based approaches
– Lattice models
– Simulating the time dependence of folding
• Refinement
• Exploring the effect of single amino acid substitutions
• Ligand effects on protein structure and dynamics
(induced fit)
Lysozyme
Lysozyme – 5lyz
Lysozyme – 5lyz
Lysozyme – 5lyz: Information from the JenaLib Atlas Page
Lysozyme – 5lyz: Information from the JenaLib Atlas Page
Lysozyme – 5lyz: Information from the JenaLib Atlas Page
Lysozyme – 5lyz: Information from the JenaLib Atlas Page
Lysozyme – 5lyz: PROSITE Signature
Lysozyme – 5lyz: PROSITE Signature
PROMOTIF Secondary Structure Analysis – 5lyz
.
.
Protein Backbone Torsion Angles
D. W. Mount: Bioinformatics, Cold Spring Harbor Laboratory Press, 2001.
Protein Backbone Torsion Angles
PROMOTIF Secondary Structure Analysis – 5lyz
PROMOTIF Secondary Structure Analysis – 5lyz
PROMOTIF Secondary Structure Analysis – 5lyz
Chou-Fasman Secondary Structure Prediction
Amino Acid Propensities
From a database of experimental 3D structures, calculate the
propensity for a given amino acid to adopt a certain type of
secondary structure

Example:
N(Ala)=2.000; N(tot)=20.000; N(Ala, helix)=568; N(helix)=4,000.
P(Ala,helix) = [N(Ala,helix)/N(helix)] / [N(Ala)/N(tot)]
P(Ala,helix) = [568/4.000]/[2.000/20.000] = 1.42
Used in Chou-Fasman algorithm
Chou-Fasman Secondary Structure Prediction
• Assign all of the residues in the peptide the appropriate set of parameters.
• Scan through the peptide and identify regions where 4 out of 6 contiguous residues have P(a-helix) > 100.
• That region is declared an alpha-helix. Extend the helix in both directions until a set of four contiguous
residues that have an average P(a-helix) < 100 is reached. That is declared the end of the helix.
If the segment defined by this procedure is longer than 5 residues and the average
P(a-helix) > P(b-sheet) for that segment, the segment can be assigned as a helix.
• Repeat this procedure to locate all of the helical regions in the sequence.
• Scan through the peptide and identify a region where 3 out of 5 of the residues have a value of
P(b-sheet) > 100. That region is declared as a beta-sheet. Extend the sheet in both directions
until a set of four contiguous residues that have an average P(b-sheet) < 100 is reached.
That is declared the end of the beta-sheet. Any segment of the region located by this procedure
is assigned as a beta-sheet if the average P(b-sheet) > 105 and the average P(b-sheet) > P(a-helix)
for that region.
• Any region containing overlapping alpha-helical and beta-sheet assignments are taken to be helical if the
average P(a-helix) > P(b-sheet) for that region. It is a beta sheet if the average
P(b-sheet) > P(a-helix) for that region.
•To identify a bend at residue number j, calculate the following value
p(t) = f(j)f(j+1)f(j+2)f(j+3)
where the f(j+1) value for the j+1 residue is used, the f(j+2) value for the j+2 residue is used and
the f(j+3) value for the j+3 residue is used. If: (1) p(t) > 0.000075; (2) the average value for
P(turn) > 1.00 in the tetrapeptide; and (3) the averages for the tetrapeptide obey the inequality
P(a-helix) < P(turn) > P(b-sheet), then a beta-turn is predicted at that location.
Lysozyme – 5lyz: Chou-Fasman Secondary Structure Prediction
http://fasta.bioch.virginia.edu/fasta_www/chofas.htm
Lysozyme – 5lyz: Chou-Fasman Secondary Structure Prediction
GRCE
RCEL
CELA
ELAA
(0.57|0.98|0.70|1.39)
0.91
(0.98|0.70|1.39|1.41)
1.12
(0.70|1.39|1.41|1.42)
(1.39|1.41|1.42|1.42)
1.23
1.41
http://fasta.bioch.virginia.edu/fasta_www/chofas.htm
Lysozyme – 5lyz: PhD/PROF Structure Prediction
PROF_sec:
Rel_sec
SUB_sec
O3_acc
P3_acc
Rel_acc
SUB_acc
PROF predicted secondary structure: H=helix, E=extended (sheet), blank=other (loop)
PROF = PROF: Profile network prediction Heidelberg
reliability index for PROF_sec prediction (0=low to 9=high)
subset of the PROFsec prediction, for all residues with an expected average accuracy > 82% (tables in header)
NOTE: for this subset the following symbols are used:
L: is loop (for which above ' ' is used)
.: means that no prediction is made for this residue, as the reliability is: Rel < 5
observed relative solvent accessibility (acc) in 3 states: b = 0-9%, i = 9-36%, e = 36-100%.
PROF predicted relative solvent accessibility (acc) in 3 states: b = 0-9%, i = 9-36%, e = 36-100%.
reliability index for PROFacc prediction (0=low to 9=high)
subset of the PROFacc prediction, for all residues with an expected average correlation > 0.69 (tables in header)
NOTE: for this subset the following symbols are used:
I: is intermediate (for which above ' ' is used)
.: means that no prediction is made for this residue, as the reliability is: Rel < 4
http://cubic.bioc.columbia.edu/predictprotein/submit_def.html#top
Lysozyme – 5lyz: PhD/PROF Structure Prediction, BLAST
http://cubic.bioc.columbia.edu/predictprotein/submit_def.html#top
Lysozyme – 5lyz: PhD/PROF Structure Prediction, BLAST
http://cubic.bioc.columbia.edu/predictprotein/submit_def.html#top
Lysozyme – 5lyz: PhD/PROF Structure Prediction
•
•
•
•
•
•
•
Perform BLAST search to find local alignments
Remove alignments that are “too close”
Perform multiple alignments of sequences
Construct a profile (PSSM) of amino-acid frequencies at each residue
Use this profile as input to the neural network
A second network performs “smoothing”
The third level computes jury decision of several different instantiations of
the first two levels.
http://cubic.bioc.columbia.edu/predictprotein/submit_def.html#top
Lysozyme – 5lyz: PsiPred Structure Prediction
http://bioinf.cs.ucl.ac.uk/psipred/psiform.html
PsiPred
PSIPRED is a simple and reliable secondary structure prediction method, incorporating
two feed-forward neural networks which perform an analysis on output obtained from
PSI-BLAST (Position Specific Iterated - BLAST).
Version 2.0 of PSIPRED includes a new algorithm which averages the output from up to
4 separate neural networks in the prediction process to further increase
prediction accuracy.
Using a very stringent cross validation method to evaluate the method's performance,
PSIPRED 2.0 is capable of achieving an average Q3 score of nearly 78%.
Predictions produced by PSIPRED were also submitted to the CASP4 server and
assessed during the CASP4 meeting, which took place in December 2000 at Asilomar.
PSIPRED 2.0 achieved an average Q3 score of 80.6% across all 40 submitted target
domains with no obvious sequence similarity to structures present in PDB,
which placed PSIPRED in first place out of 20 evaluated methods
(an earlier version of PSIPRED was also ranked first in CASP3 held in 1998).
http://bioinf.cs.ucl.ac.uk/psipred/psiform.html
PSI-BLAST
Position specific iterative BLAST (PSI-BLAST) refers to a feature of BLAST 2.0
in which a profile (or position specific scoring matrix, PSSM) is constructed
(automatically) from a multiple alignment of the highest scoring hits in an initial
BLAST search.
The PSSM is generated by calculating position-specific scores for each position in
the alignment. Highly conserved positions receive high scores and weakly conserved
positions receive scores near zero.
The profile is used to perform a second (etc.) BLAST search and the results of each
"iteration" are used to refine the profile.
This iterative searching strategy results in increased sensitivity.
Comparing Secondary Structure Prediction Results
PsiPred
Chou-Fasman
Phd/PROF
Comparing Secondary Structure Prediction Results
Protein Secondary Structure Prediction - Summary
1st Generation - 1970s
• Chou & Fasman, Q3 = 50-55%
2nd Generation -1980s
• Qian & Sejnowski, Q3 = 60-65%
3rd Generation - 1990s
• PHD, PSI-PRED, Q3 = 70-80%
Features of the new methods:
• Taking into account evolutionary information
• Neural networks
Failures:
• Nonlocal sequence interactions
• Wrong prediction at the ends of H/E
Q3 – Percentage of correctly assigned amino acids in a test set
Protein Structure Prediction
http://speedy.embl-heidelberg.de/gtsp/flowchart2.html
Modeling by Homology (Comparative Modeling)
http://salilab.org/modeller/
Modeling by Homology (Comparative Modeling)
http://modbase.compbio.ucsf.edu/modbase-cgi-new/search_form.cgi
Modeling by Homology (Comparative Modeling)
http://modbase.compbio.ucsf.edu/modbase-cgi-new/search_form.cgi
Modeling by Homology (Comparative Modeling)
http://modbase.compbio.ucsf.edu/modbase-cgi-new/search_form.cgi
Modeling by Homology (Comparative Modeling)
http://swissmodel.expasy.org/
Modeling by Homology (Comparative Modeling)
Comparative modeling predicts the three-dimensional structure of a given
protein sequence (target) based primarily on its alignment to one or more proteins
of known structure (templates).
The prediction process consists of
• fold assignment,
• target template alignment,
• model building, and
• model evaluation and refinement.
The number of protein sequences that can be modeled and the accuracy of
the predictions are increasing steadily because of the growth in the number of
known protein structures and because of the improvements in the modeling
software.
Further advances are necessary in recognizing weak sequence structure
similarities, aligning sequences with structures, modeling of rigid body shifts,
distortions, loops and side chains, as well as detecting errors in a model.
Despite these problems, it is currently possible to model with useful accuracy
significant parts of approximately one third of all known protein sequences.
http://salilab.org/modeller/
Fold Recognition (Threading)
Methods of protein fold recognition attempt to detect similarities between
protein 3D structure that are not accompanied by any significant sequence similarity.
The unifying theme of these appraoches is to try and find folds that are
compatible with a particular sequence. Unlike sequence-only comparison,
these methods take advantage of the extra information made available by
3D structure information.
Rather than predicting how a sequence will fold, they predict how well a fold
will fit a sequence.
Fold Recognition (Threading) – Why ?
• Secondary structure is more conserved than
primary structure
• Tertiary structure is more conserved than
secondary structure
• Therefore very remote relationships can be
better detected through 2o or 3o structural
homology instead of sequence homology
Fold Recognition (Threading)
Fold Recognition (Threading) – 2 Kinds
• 2D Threading or Prediction Based Methods
(PBM)
– Predict secondary structure (SS) or ASA of query
– Evaluate on basis of SS and/or ASA matches
• 3D Threading or Distance Based Methods
(DBM)
– Create a 3D model of the structure
– Evaluate using a distance-based “hydrophobicity” or
pseudo-thermodynamic potential
Fold Recognition
• Database of 3D structures and sequences
– Protein Data Bank (or non-redundant subset)
• Query sequence
– Sequence < 25% identity to known structures
• Alignment protocol
– Dynamic programming
• Evaluation protocol
– Distance-based potential or secondary structure
• Ranking protocol
Fold Recognition
http://www.sbg.bio.ic.ac.uk/~3dpssm/index2.html
Ab Initio Prediction
• Predicting the 3D structure without any “prior knowledge”
• Used when homology modelling or threading have failed
(no homologues are evident)
• Equivalent to solving the “Protein Folding Problem”
• Still a research problem
Ab Initio Prediction
http://rosettadesign.med.unc.edu/
Ab Initio Prediction
Simons, Strauss, Baker. J. Mol. Biol. 2001, 306, 1191-1199.
Ab Initio Prediction – Lysozyme (5lyz)
http://rosettadesign.med.unc.edu/
Combining Prediction Procedures
http://robetta.bakerlab.org/
Molecular Mechanics (Force Field)
http://cmm.info.nih.gov/modeling/guide_documents/molecular_mechanics_document.html
How Do We Get the Parameters ?
Experimental Data
(Examples: Geometrical Parameters)
Quantum-chemical Calculations
(Examples: Charges)
Geometry Optimization
Optimization Methods – Steepest Descent
Steepest descent
Optimization Methods – Conjugate Gradients Method
Optimization Methods – Newton-Raphson Methods
g -. gradient
h - Hessian
FLI Computing Facilities
IBM Linux Cluster
SGI Altix
Cluster vs. Grid Computing
Clusters
are made up of dedicated components and all
components in a cluster are exclusively owned and
managed as part of the cluster. All resources are known,
fixed and usually uniform in configuration. It is a
static environment.
Grids
differ from clusters because grids share
resources from and among independent system owners.
Grids are configured from computer systems that are
individually managed and used both as independent
systems and as part of the grid. Thus, individual
components are not 'fixed' in the grid and the overall
configuration of the grid changes over time. This
results in a dynamic system that continually assesses
and optimises its utilisation of resources.
EUROGRID - BioGRID
www.eurogrid.org/wp1.html
Simulation of Protein Folding
Simulation of Protein Folding
Thousand trillon FLOPs
IBM Blue Gene Project | System-on-a-Chip Approach
~ 65.000 processors
teraflop – a trillion floating point operations
per second