Comment by comicjk

All of the loops and swirls are summary representations of known atomic positions: really, knowing a protein structure means knowing the position of every atomic nucleus, relative to the nuclei, down to some small resolution, and assuming a low temperature.

The atoms do wiggle around a bit at room temperature (and even more at body temperature), which means that simulating them usefully typically requires sampling from a probability distribution defined by the protein structure and some prior knowledge about how atoms move (often a potential energy surface fitted to match quantum mechanics).

There are many applications of these simulations. One of the most important is drug design: knowing the structure of the protein, you can zoom in on a binding pocket and design a set of drug molecules which might disable it. Within the computer simulation, you can mutate a known molecule into each of your test molecules and measure the change in binding affinity, which tells you pretty accurately which ones will work. Each of these simulations requires tens of millions of samples from the atomic probability distribution, which typically takes a few hours on a GPU given a good molecular dynamics program.