Title: Efficient Energy Computation for Monte Carlo Simulation of Proteins
1Efficient Energy Computation for Monte Carlo
Simulation of Proteins
- Itay Lotan
- Fabian Schwarzer
- Jean-Claude Latombe
Stanford University
2Monte Carlo Simulation (MCS)
Popular method for studying the conformation
space of proteins
- Estimation of thermodynamic quantities over the
space - Search for low-energy conformations, in
particular the native (folded) state
3Preview of Whats to Come
- Method for speeding up MCS of proteins
- Exploits the fact that a protein backbone is a
kinematic chain - Avoids the combinatorial explosion of atomic
interactions - Gives as much as 12X speed-up for proteins we
tested
4MCS What It Is
- Random walk through the conformation space of a
protein that samples conformations on its path. - Converges to the underlying distribution of
conformations after enough time.
5MCS How It Works
- Propose random change in conformation
- Compute energy E of new conformation
- Accept new conformation with probability
6Energy Function
- Bonded terms
- Bond length, Bond angle, etc..
- Non-bonded terms
- Van der Waals, Electrostatic and heuristic
Non-bonded terms depend on distances between
pairs of atoms ? O(n2), expensive to compute
7Pairwise Interactions
- Use cutoff distance (6 - 12Å)
- Only O(n) interactions (Halperin Overmars 98)
- O(1) interactions per atom
Find interacting pairs without enumerating all
pairs!
8Reusing Energy Terms
- Only few DOFs are changed at each step
1)
2)
- Large sub-chains remain rigid between steps
- Many energy terms unaffected by change
9Our Goal
- Improve computational efficiency of MCS by
reducing average time to accept/reject a new
conformation - Independent of
- Energy function
- Step generator
- Acceptance criterion
Exploiting protein backbone is kinematic chain
10Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
11Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
12Grid Method
- Subdivide space into cubic cells
- Compute cell that contains each atom center
- Store results in hash table
dcutoff
13Grid Method cont.
- T(n) time to recompute
- O(1) time to find interactions for each atom
- T(n) to find all interactions in all cases
- No way of detecting unchanged interactions
Asymptotically optimal in worst-case!
14Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
15The ChainTree
TNO TJKTKL
TJK
TKL
16Updating the ChainTree
- Update path to root
- Recompute transforms that shortcut change
- Recompute BVs that contain change
17Finding Interacting Pairs
Test the ChainTree against itself
18Finding Interacting Pairs
- Do not search inside rigid sub-chains (unmarked
nodes) - Do not test two nodes with no marked node in
between
19Finding Interacting Pairs
20Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
21Summing the Interactions
- At each step need to sum contribution of
- New interactions
- Changed interactions
- Unchanged interactions
(1) (2) are found by ChainTree search
How to retrieve (3) efficiently?
22The EnergyTree
A caching scheme for partial energy sums
- Efficient to update
- Efficient to query
23Using the EnergyTree
24Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
25Test Setup
- Energy function
- Van der Waals
- Electrostatic
- Attraction between native contacts
- Cutoff at 12Å
- 300,000 steps MCS
- Early rejection for large vdW terms
26Results 1-DOF change
27Results 5-DOF change
28Outline
- Related work
- The ChainTree
- Energy maintenance
- Tests
- Conclusion
29Conclusion
- Novel method to reduce average time per step in
MCS of proteins - Exploits kinematic chain nature of protein
- Significant speed-up for small number of
simultaneous DOF changes - Better for larger proteins
30MCS Software
- EEF1 force field (Lazaridis Karplus 99)
- Backbone DOFs (F,?) and fixed rotamers for
side-chains (Dunbrack Cohen 97) - Classical MCS with simple move-set
- Download and customize
http//robotics.stanford.edu/itayl/mcs