Thread by @realscientists

Ok so I had a rather hectic evening, but now let's talk about simulating proteins!

First of all, what's a protein?

Protein's are large biological molecules that are responsible for a myriad of different processes in living things. Feel hungry? In pain? Taste something sweet? Feeling an emotion? All of these things are controlled by a vast array of proteins!

Proteins are amazing! Each one is highly specialised to carry out a specific task, what's incredible is that all proteins are made from the same building blocks called amino acids (AAs).

en.wikipedia.org/wiki/Amino_acid

Let's use an analogy: We can think of AAs as a brick of LEGO 🧱

Different colours and shapes of brick correspond to different properties (e.g. if it has a +ve or -ve charge, loves or hates water etc).

We want to build a LEGO truck. This truck is specialised for a specific job (e.g. moving something from one place to another).

We can think of our truck as a protein that, perhaps, moves things in and out of cells.

Usually with a box of LEGO you have an instruction manual, you can see what the final 3D truck will look like. All you have to do is put the bricks together in the correct sequence.

Let's say you didn't know what the final truck looked like (someone had removed all of the images on the box).

That's ok, as long as you followed the sequence you would still be able to make it.

With AAs this isn't (always) the case.

Having the correct sequence of AAs is great but it's really hard to predict what the final 3D truck should look like using just that.

We really need a reference structure to work from. Perhaps a bus? Similar in size / shape and still moves things from one place to another!

Luckily for us some very clever scientists have the job of solving these structures. Sticking with our LEGO analogy, they can find related vehicles (like a bus) and from that infer what the structure of certain parts of our truck must be by looking at their sequence.

What I've described above is exactly what we do for proteins. We have a sequence of AAs and can look at solved (crystal) structures of related proteins to see where the similar parts are. This is helpful for solving parts of a proteins structure that we can't resolve very well.

(there was also some recent results from Google DeepMind where they used machine learning to predict unknown protein structures, but I won't go into that for this example!)

deepmind.com/blog/article/a…

The take home: AAs are ordered together in a sequence, the sequence dictates the protein's final 3D structure, and this 3D structure determines its function.

Very clever people solve these structures and if we have those then we can start simulations 🚀

Now we can take the resolved 3D structure and choose a size and shape of box to place it in. Then we add any salt that may be necessary (your body is a bag of salty water 🧂) and finally, energy minimise.

Sometimes our starting structure might have some bad clashes between atoms so we need to fix those before running our simulation. The energy minimisation I mentioned just jiggles the atoms around to a lower energy state.

Now we're good to go! All of the same principles as I described in my thread on MD apply and we have certain PEFs that have been parameterised specifically for biological systems.

https://twitter.com/realscientists/status/1381648800696123396?s=20

These have funny names like AMBER, CHARMM, or Martini* 🍸(to name a few)

* this is actually a coarse grain PEF and works by "zooming out" so we can simulate huge systems with a loss in atomic detail

cgmartini.nl

That seems like enough for now!

Tomorrow, I'll give a brief overview of what sort of biological questions I've been looking at and how we use simulations to answer them. I'll also cover some of the limitations we have with simulations and what techniques we use to help.

I'll also explain how you (yes YOU) can get involved and do your own protein simulations using your computer.

But right now I'm signing off! 💤

(sorry for any typos, it's late!)