Choice of integrator and temperature for protein dynamics

[Dan Ziemianowicz]

This is probably a simple question, and I've been searching literature, forums and lecture notes online, but what I've found is either incomplete or too technical for my understanding - I'm a layman in the computational and MD field.

My goal is to probe solvent accessibility of proteins to verify my wet lab experiments. However, I don't want just the "outside" of the protein but also to capture any dynamics protein structure where solvent can enter. I could use smaller radii of probe spheres in my calculations but that doesn't help when I want to observe particularly flexible regions of protein structure e.g. transient pockets.

Also, some of my protein systems are seemingly unstable because they explode/reach very high temperatures very quickly if I don't employ a thermostat (e.g. Verlet integrator without separate thermostat) and minimize the energy sufficiently, while others are quite stable.

My questions are thus:
1) how does temperature in the simulation relate to temperature in the "real world"? I've read in some places that it's an approximate estimate at best, but that was in potentially outdated literature. If I set temperature to 300K, is that pretty close to the dynamics of the protein in the real world 300K?
2) for my purposes, does it matter which integrators I use? I know there are differences in how calculations are made but my calculations are relatively coarse grained and I'm mainly looking at the dynamics of individual residues. Are there practical rules of thumb for choice of integrator and temperature?
3) is ~100ns sufficient for relatively slow motions such as the dynamics of residues and secondary structure?

Any resources to a laymans/beginners guide would also be helpful. The documentation for OpenMM is great but it assumes a base knowledge which I do not have (yet).

[Peter Eastman]

If I set temperature to 300K, is that pretty close to the dynamics of the protein in the real world 300K?

Yes, it means exactly the same thing.

for my purposes, does it matter which integrators I use?

I suggest using LangevinIntegrator. That has the temperature control built in and should work fine for your purposes. 1 ps^-1 is a reasonable default value for the friction coefficient.

is ~100ns sufficient for relatively slow motions such as the dynamics of residues and secondary structure?

That depends on what you're simulating. Proteins have motions that occur on many different time scales from ps to minutes. If you want to study something that happens on a time scale of 1 ns, then 100 ns of simulation is plenty. If it's something that happens on a time scale of 1 ms, that won't be nearly enough.

A big focus of the field is developing accelerated sampling techniques that let you simulate motions without needing as much simulation time. There are lots of them with different advantages and disadvantages: replica exchange, simulated tempering, metadynamics, aMD, etc.

[Dan Ziemianowicz]

Follow up question:

If I set temperature to 300K, is that pretty close to the dynamics of the protein in the real world 300K?

Yes, it means exactly the same thing.

I've done some simulations and I'm finding that at temperatures of 400-500K my protein still retains a substantial amount of structure (via visual check), despite being quite flexible. I ran the simulations for ~100ns and found that the solvent accessibility of residues remains constant which means, I think, that the timescale is sufficient. Moreover, I find that I don't get a fully denatured protein until 1000K+. I'm running the sim with a solvated protein.

This suggests that either the protein is ultra-stable (unlikely) or the temperature is not quite 1:1 with reality. Or, that I'm doing something wrong / interpreting it wrong.

Any insights?

[Peter Eastman]

100 ns is a very short time for a protein. Just because it hasn't completely unfolded in that amount of time, you shouldn't assume it isn't going to. And since you say it "still retains a substantial amount of structure", I take that to mean that it has already partially unfolded?

Of course, it's also possible the protein really does have some structure. Proteins aren't static. They're constantly folding and unfolding. As the temperature increases, the free energy landscape shifts so that less ordered states become more common and highly ordered ones become less common. At 400K it's probably above the point where the native state is the most populated one. But it could still spend lots of time in states that have some structure.