An atomistic simulation of a polymer equilibrating. Each individual 'ball' represents a molecule.
This blog post has quite a different theme than the others. Rather than reviewing a published article, I will be speaking about some of my own work. Having recently returned from a summer research internship at MIT, the contents of this article are inspired by the work I did there.
During my time at MIT I worked in Dr. Alfredo Alexander-Katz's group where I used atomistic molecular dynamic simulations to study the nano-scale influence that anionic polymers have on ions in solution. Unlike the simulations I preform of protein hydrogels at my home institution, The University of Wisconsin - Milwaukee, the atomistic molecular dynamics used in this study are much higher-resolution. Atomistic molecular dynamics operate to model the trajectory and interactions of every single molecule in a system over a given time. So in a system of, say, 100,000 molecules, the computer calculates how each molecule interacts with every single other molecule each time step, and then derives from that a new trajectory that is fed into the next time step. This fundamentally differs from my model of protein hydrogels, in that, rather than modeling the interactions of individual molecules, I model the interactions of individual proteins.
To give you an idea of the scales involved here: the simulations I run of protein hydrogels on simple quad core I7's yield ~40 seconds of simulation time per 24 hours of computation time. Comparatively, the simulations I would run of my atomisitc systems, using a pre-built high performance and highly optimized software on a 64-core dedicated machine, would yield ~20 nanoseconds per 24 hours of computation time.
Right now you may be asking yourself, "Kirill, why would anyone use a such computationally intense method? Everyone could be two million times more efficient if they used your model!" The answer to this question is simple, my model is inherently limited in its application and accuracy. My model is only meant to simulate a specific subset of proteins, constrained to a specific set of conditions where these proteins form a hydrogel, and extract a limited amount of information about the system. Atomisitic molecular dynamics models are designed to simulate nearly anything that has molecules. All you need is to input the set of molecules contained in your system, how the molecules should interact with one another, and press go. From there, because you have the trajectory of every molecule in your system at any given time, you can extract nearly any information you can think of about your system. This all allows for a more accurate and versatile approach to molecular modeling.
Alternatively, one could then ask themselves, "Alright Kirill, if atomistic simulations are so much better--why not use them to simulate protein hydrogels?" The answer is: because atomisitc, and even coarser grain models, would be prohibitively slow to simulate my systems. The purpose of my model is to explain the mechanical response of protein hydrogels to force, such a task involves being able to simulate protein hydrogels for hundreds of seconds. Even some of the coarsest grain models, operating on the strongest super-computers and simulating systems orders of magnitude smaller than mine, can only achieve hundreds of microseconds per day. Luckily, I'm not interested in an atomistic description of protein hydrogels (right now at least); rather, I'm only looking to extract specific information about protein hydrogels. This allows me to restrict my scope, making simplifying approximations that significantly reduce computation time at the cost of overall accessible information.
Hopefully, if you were previously unfamiliar with some of the scales involved in molecular dynamics, this article has helped broadly shed some light on the different approaches and methods involved in molecular simulation.