advanced sampling techniques when molecular dynamics is just not enough
TRANSCRIPT
Advanced Sampling Techniques
When molecular dynamics is just not enough.
Why do we need enhanced sampling?
MDNN
Trying to fold a peptide from two different initial conditions, gives two different free energy surfaces !
Rates depend on barriers and T
What can we do to speed things up?
• It’s basically a problem of rates, or kinetics– Rates are exponentially dependent on temperature – higher T = higher KE = ability to cross higher barriers
• Can we just simulate at a higher T?
• Yes! Folding is faster at higher T.
Why is convergence so slow?• We want the average of observable properties- so we
need to include in that weighted average all of the important values that it might have
• Structural averages can be slow to converge- you need to wait for the structure to change, and do it many, many times.
• Sampling an event one time doesn’t tell you the correct probability (or weight) for that event
– What is the proper weight? Boltzmann’s equation
What can we do to speed things up?
• It’s basically a problem of rates, or kinetics– Rates are exponentially dependent on temperature – higher T = higher KE = ability to cross higher barriers
• Can we just simulate at a higher T?
• Yes! Folding is faster at higher T.
but…we mess up the thermodynamics. – Equilibrium constants (weights) also depend on T
Populations depend on ΔE and T
Free energies along a defined reaction coordinate via Umbrella Sampling
Umbrella SamplingHow to force barrier crossings without compromising thermodynamic properties?
Very slow transitions
Umbrella Sampling
trapped, bad ΔG
good for ΔG
One could just run dynamics and wait until all space has been sampled.
Then, if one extracts P(xk) from the trajectory, the PMF can be written as:
However, it takes forever to properly sample all conformations, and to jump over the barrier. The solution is to bias the system towards whatever value of the coordinate we want.
This is called unbiased sampling
AMBER WS 2012CECAM Lausanne
Umbrella Sampling
True PMF
Ideal BiasingPotential
No barrier,perfect sampling
We could BIAS the simulation, but we do not really know how to do it exactly.
AMBER WS 2012CECAM Lausanne
Umbrella Sampling
True PMF
Windows: 1 2 3 4 5 choose i, k and xi system-dependent
Introduce biasing potentials along the reaction coordinate
AMBER WS 2012CECAM Lausanne
Adding a quadratic biasing potential
Check for sufficient overlap
Histograms from neighboring windows should overlap strongly, all points on the RC must be sampled suffciently.
AMBER WS 2012CECAM Lausanne
Umbrella Sampling
Simulation Window Histogram Part of PMF
Final computed PMF from many windows
Solved iteratively using e.g. the WHAM program by Alan Grossfield
Constructing the PMF
AMBER WS 2012CECAM Lausanne
Umbrella Sampling
Histograms from neighboring windows should overlap strongly,
all points on the RC must be sampled suffciently.
Solved iteratively using e.g. the WHAM program by Alan Grossfield (http://membrane.urmc.rochester.edu/content/wham)
Check for sufficient overlap between sampled regions
Histograms & free energy profiles
• Umbrella run needs many simulations
• Do NOT need to sample full range in 1 simulation
G= -RTlnP/P0
Comparing 2 conformations
Song, Hornak, de los Santos, Grollman and Simmerling, Biochemistry 2006
It will take much too long to get precise populations for these 2 minima just by running MD.
8OG binding mode in complex: dihedral umbrella sampling
synanti
Song, Hornak, de los Santos, Grollman and Simmerling, Biochemistry 2006
Simulations reveal how the energy profile changes if a mutation is made
syn
anti
Song, Hornak, de los Santos, Grollman and Simmerling, Biochemistry 2006
Effect of mutations
4 Conformational space has many local minima 4 Barriers hinder rapid escape from high-energy minima
4 How can we search and score all conformations?
Replica Exchange
Coordinate
Ene
rgy
Early tries:
Run MD at Tlow, once in a while heat up to Thigh, run some more.Change to Tlow
Repeat…
The problem is that the statistical weights of the first and second Tlow are unknown and unknowable.
Metropolis Monte Carlo, one of the most important algorithms of the 20th century.
MANIAC computer.Los Alamos national Lab1943-1944
80 Kbytes memory11Khz Clock
When Bugs were real (Mark II computer, 1945)
Coordinate
Ene
rgy
We want to sample all conformations in a way that when we are done, the probability of finding a particular conformation is
Moreover, in equilibrium, we require that the rates from minimum A to B be exactly equal to the rate from minimum B to A. This is called Microscopic Reversibility.
Coordinate
Ene
rgy
Temperature Replica Exchange(Parallel tempering)
375K
350K
325K
300K
REMD
Hansmann, U., CPL 1997
Sugita & Okamoto, CPL 1999
[ ] [ ]min(1,exp{( )( ( ) ( )})i jm n E q E q
vnew vold
Tnew
Told
4 Impose desired weighting (limiting distributionon exchange calculation
375K
350K
325K
300K
4 Impose reversibility/detailed balance
Rewrite using only potential energies, by rescaling kinetic energy for both replicas.
Exchange?
Temperature Replica Exchange4 MD runs over range of T
4 Periodically swap structures
4 Faster convergence than regular MD
4 Populations as a function of T
4 Drawback: ΔE must be small small ΔT for large systems
[ ] [ ]min(1,exp{( )( ( ) ( )})i jm n E q E q
Number of replicas ~ N1/2
Does it really converge faster than regular MD?
Yes !
But not really as much as one would think (or hope, or dream)
5kBT / bG k Tk e Δ
For a barrier of ~5kBTlow, and a Thigh=2Tlow , the barrier crossing rate is x12
Better methods use replicas in Hamiltonian space. Different biasing for each replica
Do you have an MD code? Build yourself a REMD code in minutes ! Just write a wrapper !
START REMD
Spawn replicas and run MD at different temperatures
Stop MDsDecide to exchange or not
Advanced methods can help
MD
Replica exchange
NN
Roe, Hornak and Simmerling, J. Mol. Biol. 2006
What if you don’t know the pathway for the change?
in silico structure interconversions
?
“endpoint method”: • link many simulations • timescale independent
Bergonzo, Campbell, Walker & Simmerling, Intl J Quantum Chem 2009
NEB: use many simulations to study slow dynamics
R
P
Co
ord
inat
e 2
F||
F┴
Coordinate 1
Final path
Standard forces
Spring forces
Reaction Path Calculations: NEB
kcal/mol
•NEB requires coupling of multiple simulations using “springs”
A B
How can we get from A to B?
Other techniques
Accelerated MD
Hamiltonian Replica Exchange
Multi Dimensional Replica Exchange
Constant pH ( + REMD)
Steered Molecular Dynamics
Accelerated Molecular Dynamics
de Oliveira C.A.F., Hamelberg, D., McCammon, J.A., On the Application of Accelerated Molecular Dynamics to Liquid Water Simulations. J. Phys. Chem. B 2006.Hamelberg, D., de Oliveira C.A.F., McCammon J.A. , Sampling of slow diffusive conformational transitions with accelerated molecular dynamics. The Journal of chemical physics, 2007. Grant, B.J., Gorfe, A.A., and McCammon, J.A. , Ras conformational switching: simulating nucleotide-dependent conformational transitions with accelerated molecular dynamics. PLoS computational biology, 2009. de Oliveira, C.A.F., et al., Large-Scale Conformational Changes of Trypanosoma cruzi Proline Racemase Predicted by Accelerated Molecular Dynamics Simulation. PLoS computational biology, 2011.
AMD effect on the Potential
Principal Component Analysis• Build PC space based on the 500ns aMD simulation• Project X-ray, Shaw structures, 500ns cMD, and 1ms cMD
Cost: several million dollars Energy consumption ~116.5KW Code mostly hardwired to the hardware
Cost: $2000 Energy consumption ~0.9KW AMBER 12
Routine Access to Millisecond Time Scale Events with Accelerated Molecular Dynamics. Levi Pierce, Romelia Salomon-Ferrer, Cesar Augusto de Oliveira, J. Andrew McCammon and Ross C. Walker. J. Chem. Theory Comput., 2012, 8 (9), pp 2997–3002 DOI: 10.1021/ct300284c