Protein Structure Prediction

1
Protein Structure Prediction

• Luonan Chen

2
Protein Sequence Analysis

• Molecular properties (pH, mol. wt. isoelectric
  point, hydrophobicity)
• Secondary Structure
• Super-secondary (signal peptide, coiled-coil,
  trans-membrane, etc.)
• 3-D prediction, Threading (tertiary structure)
• Domains, motifs, etc.
• Subunit (quaternary structure)

3
Self-assembly

• Proteins self-assemble in solution
• All of the information necessary to determine the
  complex 3-D structure is in the amino acid
  sequences
• Structure determines function
• lock key model of enzyme function
• Know the sequence, know the function?
• Nearly infinite complexity

4
Structure of Peptide
N-terminal
C-terminal
Peptide
Backbone C-N-Ca-C Dihedral Angle or
torsional angle (F,?)
Instead of 9 variables, use 2 variables (F,?) for
each AA ?180 (CO and N-H)

(Stable
resonance) .
5
Structure prediction

• Protein Structure prediction is the Holy Grail
  of bioinformatics
• Since structure function, then structure
  prediction should allow protein design, design of
  inhibitors, etc.
• Huge amounts of genome data - what are the
  functions of all of these proteins?

6
Chemical Properties of Proteins

• Proteins are linear polymers of 20 amino acids
• Chemical properties of the protein are determined
  by its amino acids
• Molecular wt., pH, isoelectric point are simple
  calculations from amino acid composition
• Hydrophobicity is a property of groups of amino
  acids - best examined as a graph

7
(Increase local flexibility)
(Increase stability)
8
Terminology

• Active site, Blocks, Core, Fold
• Domain, Motif
• Family, superfamily
• Module
• Class
• Primary, Secondary, Tertiary, Quaternary

9
Secondary Structure

• Protein 2ndary structure takes one of three
  forms
• a helix
• ß sheet
• Turn, coil or loop
• 2ndary structure are tightly packed in the
  protein core in a hydrophobic environment
• 2ndary structure is predicted within a small
  window
• Many different algorithms, not highly accurate
• Better predictions from a multiple alignment
• Methods neural networks, nearest-neighbor
  method, HMM,

10
3-D Structure of Protein
Right-hand turn (most), 3.6 residues per turn,
F600, ?400 on average
Turn or coil
Antiparallele and parallel
Alpha-helix
Beta-sheet
Loop
Loop and Turn
11
Neural Networks for 2ndary
12
Protein Structure Classification

• Class a a bundle of a helices connected by loops
  on the surface of protein
• Class ß antiparallel ßsheets
• Class a/ß mainly parallel ßsheets with
  interveninga helices
• Class aß mainly segregated a helices and
  antiparallel ß sheets
• Multidomain proteins comprise domains
  representing more than one of the above 4 classes
• Membrane and cell-surface proteins a helices
  (hydrophobic) with a particular length range,
  traversing a membrane

13
Class ß
Class a
Class aß
Class a/ß
membrane
Membrane proteins
14
Structure Prediction on the Web

• Secondary Structural Content Prediction (SSCP)
  EMBL, Heidelberg
• http//www.bork.embl-heidelberg.de/SSCP/sscp_seq.h
  tml
• BCM Search Launcher Protein Secondary Structure
  Prediction Baylor College of Medicine
• http//dot.imgen.bcm.tmc.edu9331/seq-search/struc
  -predict.html
• PREDATOR EMBL, Heidelberg
• http//www.embl-heidelberg.de/cgi/predator_serv.pl
  
• UCLA-DOE Protein Fold Recognition Server
• http//www.doe-mbi.ucla.edu/people/fischer/TEST/ge
  tsequence.html

15
Super-secondary Structure

• Common structural motifs
• Membrane spanning
• Signal peptide
• Coiled coil
• Helix-turn-helix

16
Hydrophobicity Profile for 2ndary(positions of
turns between 2ndary structure, exposed and
buried residues, membrane-spanning segments,
antigenic sites)
17
3-D Structure

• Cannot be accurately predicted from sequence
  alone (known as ab initio)
• Levinthals paradox a 100 aa protein has 3200
  possible backbone configurations - many orders of
  magnitude beyond the capacity of the fastest
  computers
• There are perhaps only a few hundred basic
  structures, but we dont yet have this vocabulary
  or the ability to recognize variants on a theme
• Methods HMM, structure profile method, contact
  potential method, threading method,
  conformational energy (monte Carlo Algorithm)

18
Procedure of Prediction
No
Database similarity search
Align Known structure
sequence
Family analysis
Yes
Relationship to Know structure
Predict 3D structure
3D comparative modeling
Yes
No
3D structural Analysis in Lab
19
Hidden Markov Models for 2D and 3D

• Hidden Markov Models (HMMs) are a more
  sophisticated form of profile analysis.
• Rather than build a table of amino acid
  frequencies at each position, they model the
  transition from one amino acid to the next.

20
(No Transcript)
21
Homology Modeling If two
proteins show sufficient sequence similarity, it
essentially guarantees that they adopt the same
structure. Safe thresholds gt50 identity over
25 residues gt30 identity over 50 residues gt25
identity over 80 residues or more If one of the
two similar proteins has a known structure, can
build a rough model of the protein of unknown
structure. Quality of the model diminishes with
lower sequence identity.
22
Steps in Homology Modeling
1. Do sequence alignment with protein of known
structure
Known Structure ksedemkase- - - -
dlkkhgatvltalg
Unknown Structure
kseddmrrseafgctytcdlrkhgntvltalg
3. Rebuild loops where there are gaps in the
aligment
2. Replace any side chains that are different in
the homolog (green side chains)

• Adjust side-chains to accommodate the new
  residues and loops
• Energy Minimize

23
Structure 3D Profile Method (or 3D-1D method)
36 environments
Data from known library
(AA Residues)
24
Threading Protein Structures

• Best bet is to compare with similar sequences
  that have known structures gtgt Threading
• Only works for proteins with gt25 sequence
  similarity to a protein with known structure
• Current state of the art requires many days of
  computing on a dedicated workstation
• Some websites offer quick approximations
• Will improve as more 3-D structures are described
• Another aspect of the Genome Project

25
Monte Carlo Algorithm for 3D

• X set of atomic coordinates or
  mainchain-sidechain torsion angles of a protein.
  
• E(x) conformation energy
• k is Boltzmanns constant T
  is an effective temperature
• Metropolis
  Algorithm
• 1. generate a random state x, calculate E(x)
• 2. perturb x x?x, to generate a neighbouring
  conformation
• 3. calculate E(x)
• 4. If E(x) gt E(x), accept x as a new state.
  (downhill). Otherwise accept x with a
  probability exp(-(E(x)-E(x))/kT). (uphill)
• 5. return to 2