 Hello! Today I want to talk about chromium densities and molecular dynamic simulations and how to put these two together. It is how do we incorporate additional experimental information in molecular dynamic simulations? How do we steer MD simulations by adding extra forces that represent chromium data? Now, what is chromium data? Actually, what do we want to do with it? There are a bunch of things we could do. We use cryo electron microscopy to try to solve protein structures. Essentially, the first result we get from such a cryo electron microscopy experiment is a huge stack of images, two-dimensional images from which we then reconstruct three-dimensional densities that in some way or the other represent our proteins, which are shown in the top left corner. Now, MD comes in from the other side and models, proteins, behavior under dynamic conditions, so giving things temperature, having a look at ensembles and looking at how things would move. Now we want to bring one thing to the other, so this is what we like to do here in the first step. You could imagine lots of different challenges to assess and here we start with the simplest one, so to say, and that is just reconciling the structure you built above the density and so-called fitting. What you could also do is of course ask not only what's the signal structure but what would be all possible structures that would match such a density and something we can also do up to a certain extent with MD, and then of course in a future way, and maybe as a possible inspiration for future research, what we might want to do is also just maybe just skip densities directly and just jump directly from images to from the raw data to the densities, something we cannot do as of yet. So we'll just focus on densities and what are these densities then? These densities are three dimensional fields, there's a regular grid where each grid point has a certain value assigned to it, so often you see aquarium data represented as an iso-surfaces, but it's more like an x-ray image where each and every point in space gets a value assigned to it and maybe this rendering makes it a bit more clear, and when we now want to combine MD simulations with this type of data, as you see here, for example to fit such a protein into the density, we apply the additional forces that are based on the density and drive the system into the density here, and there's one choice to be made and this is where I'll drill on a bit more in this talk now, and that is a force constant for the forces for the cryo-m, so how much do you trust your molecular dynamics force field versus how much input do you want to give to the cryo-m data, how much weight do you want to give on that input, and there's different choices you could make, you could choose a very small force constant here and what will happen then is that your protein will see the density slightly affected by it, but not very much drawn into the density, so if you choose too small of a force constant, even the protein will be able to just float around in water and there'll be little effect at all of the cryo-m data, and then on the other end of the spectrum you could apply quite large forces and what you will see then is that your protein structure which is crammed into the density and will be made to fit at all cost, which really might not be what you want, consider that there's approximately some parts of extra density where you will really force the protein to fill in these cavities and extra parts of densities and consider trade-off between steer chemistry and following the data here, so what we want to do is balancing these forces from molecular dynamics and cryo-m just so that we happen to find the golden middle way between no fit and no structural information that is choosing the right force constant k here and for this we use some adaptive protocol and the guiding principle here is that we want to play as little force as possible to push the structure more towards the cryo-m density to make it more similar with the cryo-m density and the similarity here in this example just goes from left to right as you go from left to right you increase similarity with the density and this is just an example schematic figure so we want to go up this hill and we start by pushing a little bit and now there's two different things that could happen we could apply the vising forces from the cryo-m density and we see that similarity increases that is forces have been high enough and we can just keep going that way or we see another scenario where our little push here was not enough and we move downhill again and now as a response to that we can either scale up the forces just to push over that hill again or if we see that our forces are high enough actually what we want to do is scale them down a little bit so that we adaptively reduce the force and are sure that we don't distort the structure completely and by using such an adaptive protocol we ensure that we keep just enough force to keep moving from left to right without applying all too much force all the way and you can imagine even once you're over that hill then this ball will get rolling that is a we will just match the density by itself so we don't even need that extra push and then we can even scale down the fitting forces even further so that's a protocol we developed and here you see it applied in a simple case we we just fit a very small helix against a very high resolution artificial density and as we keep going in the simulation we see that we become more similar to the density but since we enforce still more and more similarity what will happen anyway even we try to be as gentle as possible but if we want to keep going to the right as we had in this picture we have to increase the force constant over time so what happens initially is just a little bit of movement of the helix will just drastically increase similarity but then as we go further in the simulation forces will have to be larger and larger and at some point they'll be so large that they overrule the force fields so the question is how to decide when to stop even with that adapter force protocol and for this we can quantify things a bit more we can have a look at the average potential energy that reflects what this structure is so that's a running average of the potential energy and on the other hand we can have a look at some measure of how well do we fit in this case using some fsc average you could choose and think of different measures and what we want to do here is stop just right when the moving average of the potential energy is just about to explode and that means that we're cramming in the structure that we are seeing something that is most likely not biologically relevant however this is something we can determine depending on the situation and what type of system you want to look at and as you see with the balance we start the simulation we we just apply very little force and seen no structural distortion easy increase in fitting and at some point we will have to force and push the system very hard now that is a kind of made up example does this work when we go to largest systems and biologically relevant systems and how challenging can things be to do this we have a look at an alderlice structure which is very well defined system still on need high resolution density of around 2.1 ocean we see here applied and what we did to make the refinement a bit more challenging as we distorted alderlice structure by just heating it up in a dynamic simulation simulated annealing simulation and now have a look at how well can we reconcile and we find back the initial structure and as you can see during the melting process the structure expanded quite a bit and also shifted around a bit and now with this adaptive force skating we can also see that in for such a case we get back to our structure which should give you a good example for example for a homology modeling applications when your initial starting structure is even a bit remote from what you want to refine and fit so here you see that we run just normal MD simulations and if you want to do that yourself for your application it's really just add the density to a normal MD simulation and all this adding is done in the mdp file but anything prefix density guided simulation and if you really are going for the smallest and most lazy possible setup is really just saying yes to density guided simulations by this parameter and adding of course your density information as some usual MRC map ccp for whatever you like in this case and then you can run if you see or if you're interested in more information we put one tutorial that just shows a small test case for density guided simulations on the virtual machines where you can try things for yourself play around a bit with the options and try things at home we also have a bike sale webinar on exactly that type of simulations which you will find on youtube on the bike sale channel and there i'll go much more to depth and details of all the methods and just have a look at the chromics manual where there's an extensive section on all the technicalities of that method and all the math and formulas explain yeah thanks a lot for listening and i hope you enjoy that do try out the tutorial if you're interested in that type of method or if you just want to play around and yeah i'll thank all the people helping the work especially eric dindal and group and people from my phd time is long gone gutting it that helped start up the whole thing so thank you