Practical Aspects of Structure Prediction Michael Feig, Michigan State University

1
Practical Aspects of Structure
PredictionMichael Feig, Michigan State
University
Theory and Computation in Molecular Biological
Physics Center for Theoretical Biological
Physics Research Workshop and Summer
School August 9-20, 2004 La Jolla, CA
2
What is protein structure prediction?
Amino Acid Sequence
3D Model of Biologically Active Conformation
Structural Analogy
Empirical Knowledge
Physical Theory
Experimental Constraints
3
Why structure prediction?

• Structural information leads to protein
  function
• Protein structures allow rational drug design
• Difficulties in experimental structure
  determination
• Can be fully automated

4
Protein Structure Elements
5
Secondary Structure Prediction
Helical? Extended? Random coil? Disordered?
Accuracy
PSIPRED http//bioinf.cs.ucl.ac.uk/psipred/ JPRED
http//www.compbio.dundee.ac.uk/www-jpred/ SA
M-T99 http//www.cse.ucsc.edu/research/compbio/HM
M-apps/T99-query.html
6
Tertiary Structure Prediction
How do secondary structure elements fold?
Prediction through homology, analogy, ab initio
7
Quaternary Structure Prediction
Domain organization? Oligomers? Complexes?
Prediction requires modeling of protein-protein
interactions
8
Sequence Homology
Human thioredoxin (1AUC)
SDKIIHLTDDSFDTDVLKADGAILVDFWAEWCGPCKMIAPILDEIADEYQ
GKLTVA . .. . .. ..
....... . . MVKQIESKTAFQEALDAAGDKLVVVDFSATWCGPCKM
IKPFFHSLSEKYSNVIFL- KLNIDQNPGTAPKYGIRGIPTLLLFKNG
EVAATKVGALSKGQLKEFLDAN---LA ..... . . ..
.. ... . ..
. EVDVDDCQDVASECEVKCMPTFQFFKKGQ----KVGEFS-GANKEKL
EATINELV
E. Coli thioredoxin (1THO)
9
Comparative Modeling
Assumption Proteins with similar sequence
have similar structure
E. Coli thioredoxin (1THO)
Human thioredoxin (1AUC)
10
Structural Templates from Homology

• Challenges
• Correct alignment
• Loop modeling
• Side chain rebuilding

PGTAPKYGIRGIPTLLLFKNGEVAATKVGALSKGQLKEFLDAN---LA
. . .. .. ... . ..
. QDVASECEVKCMPTFQFFKKGQ----KVGEFS-GANKEKLEATINEL
V
11
Accuracy of Predictions by Homology
automatic predictions by SWISS-MODEL web service
12
Prediction through Fold Recognition
Assumption Proteins with similar secondary
structure share fold
1N91
1JRM
13
Structural Templates from Fold Recognition

• Challenges
• Good template
• Correct alignment
• Fragment modeling
• Refinement

14
Ab initio Structure Prediction
Theoretical models provide energetic
description Sampling generates protein-like
conformations Scoring identifies
native-like structure with lowest free energy
SEALGDTIVKNA
15
Conformational Sampling

• Needs to generate protein-like conformations
• Needs to include near-native structures
• Needs to be computationally efficient
• Suitable methods
• gt Low resolution on/off-lattice models
  (SICHO)
• gt Assembly from short protein fragments
  (ROSETTA)
• gt Torsional space sampling of all-atom
  models

16
Reduced Protein Representations
all-atom
Ca side chains
lattice
17
SICHO Lattice Model
Monte Carlo simulations gt Attempt move gt
Compute DE gt Accept with probability p
Simulated annealing Constant Temperature Replica
Exchange Sampling MONSSTER program
Kolinski Skolnick Proteins 32, 475 (1998)
18
SICHO Energy FunctionKnowledge-Based Terms

• Excluded volume

• Side chain burial propensity
• follows Kyte-Doolittle scale

• Centrosymmetric bias
• rg 2.2 Nres0.38
  

19
SICHO Energy FunctionPMF-based Statistical
Terms

• Short range interactions
• PMF based on statistical
• analysis of PDB

helix
extended
20
Conformational Sampling with SICHO
Protein A
21
Scoring Functions

• Knowledge-based/statistical
• derived from known protein structures
• limited by training data
• usually fast -gt useful as filters
• Force field based
• model physical energy landscape
• more robust and transferable
• often expensive -gt useful for final
  scoring

22
Desirable Scoring Function
23
MMPB(GB)/SA Scoring Function

• Solute internal energy

DGMM DUbonded DUvdW DUelec

• Solute-solvent interactions

DGsolv DGPB/GB DGhyd DGsolv,vdW

• Total relative free energy

DGMMPB/SA DGMM DGsolv TDSprotein
24
Scoring of Lattice ConformationsProtein A
25
Performance of Force Field Based
ScoringEvaluation of CASP4 Predictions
M. Feig, C. L. Brooks III Proteins 2002, 49,
232-245
26
Global Distance Test
GDT(r) How many overlapping continuous
residues are within r Å?
GDT(TS) (GDT(8) GDT(4) GDT(2) GDT(1))/8
27
GDT vs. RMSD
28
Ab initio Structure Prediction Protocol
Secondary Structure HM/FR Templates
Lattice Model Sampling Replica Exchange Monte
Carlo Sampling
-gt 20000 _at_ 5-25 Å RMSD
Reconstruction
All-Atom Clustering Scoring MMGB/SA
-gt 10-100 _at_ 5-10 Å RMSD
Refinement Replica Exchange Molecular Dynamics
-gt 1-5 _at_ 2-4 Å RMSD
29
Sampling ScoringAb initio predictions of DNase
fragmentation factor (1KOY)
30
Sampling, Scoring, ClusteringAb initio
predictions of DNase fragmentation factor (1KOY)
31
Ab initio Predictions DNase fragmentation factor
Best-scoring Prediction 7.4 Å RMSD
NMR structure 1KOY
32
Ab initio Sampling in Template-based Structure
Prediction

• Template provides
• known protein
• structure
• Ab initio sampling of
• unknown fragments
• in the context of
• template

33
Template Restraints Near Flexible Part
Restraint potential
0.1
0.0
0.2
0.4
1.0
0.7
34
Loop Sampling
35
Sampling with Restraints

• Secondary structure bias
• Secondary structure prediction
• NMR shift data
• Distance restraints
• Experimental restraints
• Side chain contacts from analogous
  structures
• Template restraints
• Homologous/Analogous Structures (PDB)

36
Structure Prediction Restrainedby Sparse
Experimental Data

• Secondary structure information
• NMR chemical shifts, circular dichroism
• Distance restraints (atom-atom,
  residue-residue)
• NMR NOEs, EPR, cross-linking, TRP
  flourescence
• Relative orientation of structural elements
• NMR dipolar coupling
• Surface residue distribution
• Antibody epitope scanning
• Molecular shape envelope
• SAXS, electron microscopy

37
Structure Refinement
?
predicted
native (NMR)
38
Sampling Towards the Native Basin
E
states
q
native basin
39
Energy Landscape Towards NativeT0167
native
40
Multi-Scale Sampling
Low Resolution
All-Atom
41
Multi-Scale Sampling Scheme
Low Resolution Sampling
All-Atom Reconstruction
All-Atom Energy Evaluation
Metropolis
reject
Save Conformation
42
Refined Structure Prediction Protocol
Homology Template
Low-Resolution Sampling Long simulated
annealing/replica exchange runs
All-Atom Scoring Clustering, Minimization,
MMGB/SA
5-10 Å RMSD
Low-Resolution Sampling Very short simulated
annealing runs
All-Atom Scoring Clustering, 1-10 ps MD, MMGB/SA
average
2-4 Å RMSD
gt20 Cycles
All-Atom Replica Exchange Sampling gt1 ns per
replica
1-2 Å RMSD
experimental-like native structure?
43
Structure Prediction Summary