1 of 14

The problem with standard simulations

Chong, Saglam, and Zuckerman, Curr. Opin. Struct. Biol., 2017.

2 of 14

Generates unbiased, continuous pathways and rates with orders of magnitude less computing than standard MD.

Efficiency scales exponentially with the barrier.

von Neumann & Ulam “splitting strategy” (1950s)

Huber & Kim, Biophys. Journal (1996)

Zuckerman & Chong, Ann. Rev. Biophys. (2017)

Weighted ensemble MD focuses on transitions

Resampling procedure

3 of 14

How does resampling work?

time = 0 ps

probability =

Run MD

τ = 100 ps

 

progress coordinate

4 of 14

time = τ

τ = 100 ps

probability =

flux = 0

progress coordinate

5 of 14

Run MD

τ = 100 ps

probability =

flux = 0

progress coordinate

time = τ

6 of 14

probability =

time = 2τ

τ = 100 ps

flux = 0

progress coordinate

7 of 14

τ = 200 p

Run MD

time = 2τ

τ = 100 ps

probability =

flux = 0

progress coordinate

8 of 14

τ = 300 ps

time = 3τ

τ = 100 ps

probability =

flux = 0

progress coordinate

9 of 14

τ = 300 ps

Run MD

time = 3τ

τ = 100 ps

probability =

flux = 0

progress coordinate

10 of 14

t = 3τ

Probability =

τ = 400 ps

time = 4τ

flux = 0.00125 ps-1

τ = 100 ps

probability =

progress coordinate

11 of 14

τ = 400 ps

time = 4τ

τ = 100 ps

probability =

flux = 0.00125 ps-1

progress coordinate

12 of 14

τ = 400 ps

Run MD

time = 4τ

Flux = 0.5

τ = 100 ps

probability =

flux = 0.00125 ps-1

progress coordinate

13 of 14

No free lunch!

May miss slow motions orthogonal to the progress coordinate.

Worse case is “brute force” sampling of orthogonal motions.

However, we can switch progress coordinates “on-the-fly.”

14 of 14

Weighted ensemble MD greatly extends accessible timescales

10-15

(fs)

10-12

(ps)

10-9

(ns)

10-6

(μs)

10-3

(ms)

100

(s)

Surface

side chain

rotation

Bond vibration

Hinge bending

Protein folding

Loop motions

Standard MD (Anton)

Weighted ensemble MD

Protein binding

Large-scale conformational transitions