Transcript PowerPoint 1.5MB - The Biomolecular Modeling & Computational

Molecular modelling /
structure prediction
(A computational approach to
protein structure)
Thomas Huber
Department of Mathematics
Room 724, Priestley building
[email protected]
Today:
• Why bother about proteins/prediction
• Concepts of molecular modelling
– The physicist’s approach
– The biologist’s approach
• Get a feel for usefulness/uselessness
• Where is the future going?
Why do we care about
Protein Structures/
Prediction?
• Academic curiosity?
– Understanding how nature works
• Drug & Ligand design
– Need protein structure to design molecules
which inhibit/excite
• cure all sorts of diseases
• Protein design
– making better proteins
• sensor proteins
• industrial catalysts (washing powder, synthetic
reactions, …)
• Urgency of prediction
– 104 structures are determined
• insignificant compared to all proteins
– sequencing = fast & cheap
– structure determination = hard & expensive
Three basic choices in
molecular modelling
• Representation
– Which degrees of freedom are treated
explicitly
• Scoring
– Which scoring function (force field)
• Searching
– Which method to search or sample
conformational space
The physicist’s approach:
Folding by 1st principles
Concept: Doing what nature does
• Representation:
atomic level
• Scoring: physical force field
• Searching: Newton’s equations of
motion
Naïve idea?
• Levinthal’s paradox (1968)
– 3 possible rotamers per dihedral angle
astronomical number of conformations
• Golf course scenario
Levinthal’s paradox
is irrelevant
• Folding is not a random process
 Bumpy bowl
scenario
Why are folding
simulations still unsuccessful?
•
•
•
•
Simulations computational expensive
Force fields are not good
Gross approximations in simulations
Nature uses tricks
• Posttranslational processing
• Chaperones
• Environment change
Is a physical
approach useless?
• No!
• Very useful aid to
structure determination / refinement
– Experimentally observed structural data
very incomplete
• NMR: only distances < 6Å
• Xtallography: only 50% of data can
be measured (phase information
missing)
– Physico-chemical information and
complement experimental data
• Give dynamical picture of structure
Biologist’s approach:
Prediction by induction
Concept: Homologous sequences fold
into similar structures
• Representation: amino acid sequence
• Scoring: sequence similarity (identity)
• Searching: optimal string matching
(with gaps and insertions)
Validation of concept
(Rost, 1999)
• >106 sequence alignments between protein
pairs
• Optimal discrimination between similar
and dis-similar structure
Is it useful?
• PDB statistics:
– 104 protein structures determined
– <103 protein folds
Template recognition
Alignment
Alignment correction
Backbone generation
Loop building
Side chain generation
Overall model refinement
Model verification
Force field
•
•
•
•
•
•
•
•
Sequence score
8 Modelling steps
– Comparison with Experimental results
– Steric overlap
– Ramachandran plot
Limiting factors
How good are
homology models?
• G.V. Vried 1998:
34 homologous protein pairs
What about side chains?
• Biology happens in side chains
• Packing side chains in protein core is
not a trivial problem
– Many alternative arrangements
– High energy barriers
Accuracy of modelled
side chains
• Dunbrack SCWRL results
– 299 monomeric proteins
– 40263 side chains
The Next Step:
Computational Proteomics
• Mass scale homology modelling of
entire genomes
– Lots of sequence data
– First pick the easy cases
– Computers are cheap and work 7-24
Prediction of Protein
Structure
How to detect remote
homologues
• Fold recognition using threading
– Combine concepts of physicist and
biologists
• Predicting secondary structure
• More about that in BIOL3004
– Structural biology elective
• Tue 8/5 10am
• Thu 10/5 10am
– Database mining elective
• L10
Take home messages
• Computational approaches are
– Not perfect
– Yet indispensable
• Molecular modelling has huge
potential in structural biology
– Currently 104 structures in PDB
– For every sequence in the Swissprot
database with homology to a structure
in the PDB models are available!!
– Vast amount of data still to come
• Levinthal paradox
– Is true
– BUT not relevant
• Different aims need different
approaches (3 choices of MM!)
– modelling enzyme reactions
– modelling protein folding
– weather forecast
Clever approaches more
important than bigger
computers