Introduction -if you already know about MD simulations, move on to the next title
Molecular dynamics simulations (MD for short) is a series of computational methods to simulate the behavior of matter at a given scale. The most relevant type of MD simulation for this story is that where a molecular system is described by all its atoms, taking each atom as a sphere connected with springs to other atoms and experiencing various kinds of interactions with all other atoms of the system. For simulations of proteins, the “molecular system” is most likely a protein surrounded by water, but it can also include other molecule such as nucleic acids, drugs that bind the protein, even pieces of biological membranes or membrane mimics.
Summarizing what could be a whole physics course in itself, in MD simulations the positions of the atoms are evolved over time following a potential (“forcefield”) that determines the strengths and equilibrium positions of all possible interactions, under the shaking of a temperature bath and a given pressure. This evolution of the atoms’ positions produces a kind of movie, called trajectory, that hopefully represents the real kinds of motions that the molecular system can experience.
The trajectory, or movie of the molecular motions, builds up very slowly, because at each timepoint the program needs to compute the forces on all the atoms and then propagate them over time, but since some forces are quite strong then the time interval (“time step”) during which the motion is integrated is very, very small. To put some numbers, the time step is typically in the order of 1–2 femtoseconds (1E–15 s) but we are interested in motions that happen in nanoseconds, microseconds and even milliseconds timescales. This means that we need to integrate the steps (i.e. calculate forces and propagate the atoms’ positions accordingly) millions to billions of times.
As you can imagine then, MD simulations run faster and faster as more powerful processors come out over the years, yet today normal MD simulations with all-atom description reach at most around 10–100 microseconds. Already for around 10 years, GPUs provide most of the computer power used to run these MD simulations. Although it’s not extremely difficult to run them, at least not in a research setting, it is good that this team of researchers from California developed a Colab Notebook to run MD simulations on Google’s GPUs, because in many underdeveloped countries one GPU…
Continue reading: https://towardsdatascience.com/new-preprint-describes-google-colab-notebook-to-efficiently-run-molecular-dynamics-simulations-of-9b317f0e428c?source=rss—-7f60cf5620c9—4