 Hello everyone and welcome to this bioxial webinar. This is number 65. The topic is going to be QMM simulation of both fluorescent proteins and proton dynamics. So today we have two speakers, Dmitri Morozov from the University of Javascula and Mirko Pavlikat from Forschung Zentrum Julich. I am Arna Prueber based at the University of Edinburgh and with me is also my co-host Alessandro Avila from KTH Royal Institute of Technology in Stockholm. To introduce today's speakers, we have, first of all, we will start with Dmitri Morozov. Dmitri is based at the Department of Chemistry and Nanoscience Centre at the University of Javascula in Finland. His PhD research is focused on the development application of hybrid QMM methods to study biological system properties. Since 2013 he's been working in the group of Professor Gerard Grunhoff, also at the University of Javascula as a post-doctoral researcher. From 2019 he's been working within the bioxial consortium on implementing multi-scale methods to perform simulations of chemical and biological systems. Dmitri will be presenting on computational screening and properties evaluation of fluorescent proteins. And second, Mirko Pavlikat is based at the Institute for Advanced Stimulation and Institute of Neuroscience and Medicine at Forschung Zentrum Julich in Germany. In his PhD he investigated spectral properties and mechanisms of thiamine diphosphate-dependent enzymes. And then in 2020 he joined Professor Carloni's computational biomedicine group at Julich as a post-doctoral researcher. Now his research focuses on molecular simulations of biologically systems like using high-performance computing QMM approaches. And from January of 2021 he's been working also within bioxial consortium on applying the QMM software interfaces developed in the consortium to study proton dynamics with biomolecules in the gas phase. So with that I will hand over to our first speaker Dmitri. Thanks Anna. Okay, so welcome to our webinar. So my part will be devoted to the computational screening of fluorescent protein mutants, subprojected by Excel, which we're handling in the University of Javascula along with my colleagues. So, basically, first of all, of course I want to present, maybe not of you are familiar with what is fluorescent protein, so I will talk a bit about this. So the fluorescent proteins, which actually discovery of that have been awarded with a Nobel Prize in 2020 in 2008. So it's proteins which are under certain conditions when you eliminate them in its light, they also start fluorescing. The first type of this protein was found in 50s, I guess, and it was found in the Aquaria Victoria jellyfish. And when it was isolated and finally measured structure it appeared to be a bit of a structure with a chromophore inside the chromophore is typically looking like something like that. And actually if you check it's made after catalytically out of three amino acids. So basically styrosine, glycine, and certain acid can vary from type depending on the type of the protein. And of course, one of the features of that protein is that they're not just flores, but they can be also made reversibly switchable so it means that they can reverse the stitch between fluorescent and non fluorescent states, not all of them, but many of them can. So applications of course there is a large variety of applications nowadays in the modern biology, because they're used as a biological markers for like for for imaging the link cells even so here are some examples. Yeah, and that's of course because they're genetically encoded, they're very, very, very large application so now it's even you can dance do something with that is a cat. And of course, funny enough, that technique which is also called super resolution microscopy, yeah, which I use is actually for us the proteins have been also are the Nobel Prize in 2014. So, maybe you know that even six years you have a two Nobel prizes for from the same a field that this field is very important. So, but why we want to study this fluorescent product as well because they actually can be a large variety of them, and their fluorescence can span from the blue light up to the far right here. There is a large variety. So, the idea is so basically, sometimes you need a very specific protein for your very specific tasks. And that's why you want sometimes even to design yourself a protein. So, and that's why where's our computations to come into play. So, and we want to, we want to have a way how to design will develop a way how to design fluorescent proteins. So, for that we developed using by Excel, flu procad workflow, which is fluorescent protein computerized design, of course, and this workflow consists of several packages which are working together. So, for that we need, as the input we need some initial structure, we need to know which mutations we want to insert to that structure so basically, in some crystal structure, we need mutations, and we need to know which properties want to calculate. So, we then we do automatically add the only synchronized atoms, we can do that mutation using PMX, and then we have several types of properties which we can go with that workflow. So, apparently we can do a structure, we can do a structure determination of the product of the mutant, we can do a thermodynamic properties of the mutant, and we can do a photochemical properties of the mutant. So, I will start first with the application one which is predicting the structures. And for that we're using, which is this part of the workflow. And for that I will use, we're using a, of course, clusterization. So, basically what we do we first generate from the crystal structure and no mutations with generating the mutant structure, initial mutant structure, then we relax it. We do a large scale, we do a large molecular dynamical trajectory, here it's exactly 100 nanoseconds, but typically it is more than 100 nanoseconds, much more. And after that we pick snapshots from this trajectory and start crystallizing them on the basis of similarity of the backbone chain. And after that we pick up the cluster structures and the most populated one will be, of course, in most cases the most probable structure. And we picking up the structure which are as close as possible to the center of the cluster, particular cluster. And thus we can predict what the structure will look like. Okay. How it works, we are testing this on the set of mutants of the GIF, fluorescent protein which is called Teres Green, which was done collaboration with Kauleven, experimental group here. So, basically, we pick up a starting structure from the PDB, which is very well known for that protein. And we done 14 mutants in total, so here is just sticks, but we done over 14 mutants, for which we did not knew crystal structure at all. So we first done our simulation, and then we blindly checked that against the crystal structure which was measured afterwards. And here is what we get in the results. So first of all, of course, we observed that some of the mutants are basically RMSD plot, and this is the same, but in the frequency domain basically. And what we see that some of the structures they're showing double peaks, like this black one, which is the first one, like this magenta, which is this one. So there is up to five things actually showing the double peak, double peak in RMSD, meaning they have more than one cluster, typically. And we investigated all of these. And what we found that most of these clusters, the problem comes with not problem, but the feature which determines this comes in the amino acids on the surface of the beta barrel, because what's happening, for example, between this and this, and this mutant you see that this mutated histidine to serine mutation, it always in a different confirmation, you know this kind of mutants. And even in the crystal structure can flip in and out, it have a different thing. And what we're observing actually in the trajectories. And that's why there is such a double peak structure. The thing is that this serine can flip in and out of the beta barrel in the dynamics. So basically this gives us an idea that this protein, not only have crystals in crystal structure you'll probably find only one confirmation of this serine. In most cases in the solution, it will have two structures and you need to take into account this. When you do molecular dynamics simulation for the properties evaluation that potentially, you can have several structures in solution in comparison to the crystal structure. But sometimes even in crystal you can see this heterogeneous idea you know there could be two position PDB file support several positions for the same amino acids in one file. So, more funny situation is with these two mutants, which are, which are these two, which also have a double peak. Here it comes that histidine itself, and this amino acid which will also mutate to histidine, it's either here it means E protonate histidine, heaps means doubly protonate histidine. We test of course for histidine, you need to test several possibilities here it can be protonate by E, it can be protonate by D, it can be doubly protonate. How it could not be deprotonated in most cases. So, and we see that that histidine is also because both of them are on the surface. So here is the surface of the beta barrel. It's also can flip in and out of the of the of the chromophore pocket and out to the solution that you also need to take into account. But overall we can say that according to our simulations. So basically this pink structures is experimental structure measured afterwards. And we see that in most cases we have a very good agreement with exception of this. Sometimes of these several positions but for example in this structure you can see that even in the experimental one you have two positions of the same system. So we are not so wrong with this, and thus we can confirm that we can predict the structures quite well. Even we can go beyond the crystallography we can simulate heterogeneity in that structure. Okay, next I will go to the second application, which will be more interesting for most of the people so it's thermodynamic properties. And for thermodynamic properties using combination of PMX and Gromach's. So basically what's PMX, it's script or it's program that allows you to do so called free energy perturbation calculation so it basically generates for you topology that is suitable for that energy perturbation simulations. And it replace your some of your amino acid with so called hybrid residue, which can be in two states. It can be, for example, here while in and here is phenylalanine. So, and then after that you can do a thermodynamic cycle. So here we have for now we have a two properties which can calculate first is the folding free energy, and second is a dimerization free energy. So folding we calculate with respect to the for unfolded structure use a very generic model structure which is just a three peptide in a solution glycine some residue glycine. Yeah. And we are trying to calculate this difference energy between basically this and this. For that we studied in three mutants of the wild type GFP, or from you don't sorry, of the wild type GFP. The crystal structures known actually all of them. And we just trying to predict the free energies of folding and dimerization. Yes. Okay, first, we have a structures comparison. And here we have again the pink one, pink, white, and how to call blue one is a experimental one. And the other one is what we have simulated. And you can see that is quite well again agreement with the structures between the mutants and the, and the wild type GFP. And here is the results of our simulation free energy simulation. So, funny enough, we found one Newton which have a particular stabilization of the folding, which is this one. But most importantly, we found that this mutation, which is also actually was the same mutation in the, but other way around it was rising to our need back mutation towards the nearest green. But anyway, so basically mutating this residue on the surface of the data barrel with lysine, which is charged of course and which is very hydrophilic. And it renders it to preform an American form, which is actually very well understandable, because you're doing something which is something hydrophilic. And a replacement hydrophobic is hydrophilic. And then of course it's starting to be less dimeric. And the dimerization of course actually drops a lot. So it's several, it's 10 kilocalories. So, then we started of course to investigate what's happening why it's also big number and actually what we found, even more that if you just do normal dynamics of the dimer with that mutation, what you see that in the record dynamics your dimer starts falling apart. And first, first thing of this is that your angle between the two data barrels become very unstable. So, and you see some rotation about this around this timer, like almost 40 degrees, but then it becomes very unstable. So you can immediately see even this without the free energy complex structure is that you have an unstable system. Okay. And this is actually what happens. So basically this lies in, it starts to point it when it was all in, it was normal, it stays here is here. But now lies in just points outwards to the solvent and just prevent this, sorry, prevent this dimerization, it just wraps this dimerization interface, which is should be made out of the hydrophobic. Okay. And finally, the third application which I want to talk about is for the chemical properties. So basically we picked up absorption and fluorescent spectra, of course, which is one of the key, key property of the fluorescent proteins. Yeah, and for that we're using the following approach so basically we again do large molecular dynamics. Again, equidistantly pick up a snapshot. Yeah. And in that snapshot we do a short simulation to relax it. And then after that we calculate absorption spectrum. So absorption spectrum we calculated for now is the EGT but potentially many methods can be used here. And after that we convolve total absorption spectra with Gaussian functions like this. There's transition cycle moment, there's difference in energy. Okay, and how to calculate fluorescent spectra, this is a more advanced thing, because potentially to access the excited state, you need to do KMM and we actually doing also that, but then we realize that it's very long simulation. And instead of these, we just trying now to provide trying to generate the excited state for two parameters. So basically we're trying to make forces for excited state. Trying to make we just use forces for excited state. And then we do one on the second simulation excited state and the complete the emission spectra from that basic fluorescence spectra. But here we can also also do a KMM in reality it will be just much, much longer. Okay, so here is a spectra which I calculated. So basically this absorption is magenta and green one is emission. So and we see very distinct picture that some of the, so all of them are fluoresced more or less at the same position around 2.25. So absorption differ could differ a lot. And why it is so well, because actually, many people know, people who are working with the fluorescence proteins know that the chromophore in that proteins could be in several protonation states. So it depends on the presence of absence hydrogen at this OH group of the tyrosine ring of the phenyl ring basically. But depending on that you will have either a very small stock shift, or you can have a very huge stock shift, because this chromophore is actually for acid so we try to throw out the proton upon excitation. And it is very well known that typically the fluorescence come from the anion. So basically when you don't have proton here but absorption can come from both from anion, then it's like this or from neutral one. And after which you see excited state proton transfer and see a life stock shift because of this. And finally, how we can model this with excited state dynamics is just an outlook, because it's now not yet in the workflow but potentially it can be done. So here's the example of the DDFT simulation molecular dynamics of that exactly ultrafast proton transfer so it is very well known, not very well known but there is a mutant in which if you mutate this history into the aspartate. And then you will have ultrafast proton transfer which is scale of this is under under the hundred seconds so it's really ultrafast. And that's how we can simulate this so basically we start this configuration when the protein is exactly on the chromophore. It's a simulation and it almost immediately goes to the to the aspartate. Basically it happens within like 50 seconds or so. Okay, so with that, I think we are. Yep. There is a short summary of my talk so basically, we, with our full product workflow we can do simulate we can simulate, we can predict structures of the mutants. We can predict chromodynamic properties of the mutants without doing actual experiments. We can evaluate some photochemical properties. We have some databases for some parameters for several chromophores and can win parameters. And, yeah, we have a full basically mutagenase protocol in silica for simulating fluorescent proteins. So some acknowledgments, yeah, collaborators and by Excel and European horizon program. So thanks for your attention. Yeah. So as Alexander already put in a chat, everybody please feel free to enter any questions for me three in your in the q amp a panel q amp a box. Then we will go to medical. Thank you very much. What I want to talk about. It's also part of the use case on the QMM simulations I and I want to talk about proton dynamics and mass spectrometry. So, proton transfer between particular of DNA molecules and ammonium ions and the mass but from the conditions so in the gas phase, and I want to first introduce the hydropathromatic techniques which are used commonly in biomolecular applications. And, yeah, the ionization process is typically done with electric space ionization so the analytes are in an actual solution and spray directly through a capital out when the high electric field is applied into the gas phase. And one of the advantages of mass spectrometry and the setup is that the benefits require just the very, very low concentrations of analytes. And after that droplets are formed first several analytes inside these droplets, and due to an evaporation and coolant repulsion smaller and smaller droplets are formed so that finally the analytes are in a single droplet and then are fully disolvated with a complete evaporation process of these analytes and then can be analyzed to study biomolecules one has to pay a bit attention, and there's a specific technique which is called native azim MS. So, a native just means that it has to be compatible between the ionization process so the solution conditions have to be compatible with ionization, but also the biomolecular system of course should be in the same state as would be observed on the physiological conditions and what typically gives good results is that they are, yeah, potential solution of ammonium acetate solution. And to get additional insights on the analytes, often it's covered to the eye mobility mass spectrometry, what happens here. So after these analytes are, yeah, disolvated and the entire circle drift tube is accelerated to the main electric field. And from the other side and in a drift grass and that, and due to the collision of the drift cars with the, with the analytes, which can have a different shape or different confirmation so a different form and size. So these can be separated within this drift tube and then separately analyzed. So finally, which are the experimental observables and information we get is the mass and charge ratio of of the analytes and the total charge distribution, and then also the collision across section here from these separations within this drift tube, which can then be related to the conformations of molecules, the topology of complexes and also to identify conformational changes upon the binding. But of course, at a low resolution and here comes molecular dynamics and term and I'm simulations into play to get an atomistic insight into these biomolecular systems. And in my talk I will focus on DNA in the gas phase and typical computational protocol which is done is outlined here to get to simulate these DNA molecules and the gas phase. So this one typically runs in classical and these simulations and accurate solution. Then one selects randomly snapshots or one plus there's different conformations and select snapshots and removes the molecules for DNA. The DNA backbone has to be, yeah, has to be protonated in the form that the experimental observed total charge status is obtained but that I will discuss later, and then we can run extended gas phase simulations and to write the properties and compare with the experimental results and from our cell from the simulations get into a domestic details between of the biomolecules of the conformations or of the type of the complexes. An interesting application was published by the group of Modesto Roscoe who studied the DNA of the complex here. And they figured out that just selecting the one of these snapshots and removing the solvent, they could not reproduce the experimental collision with crush actions. And then they did some of the tools, even if they sample up to the microstate types here. And what they did is the then tested or carried out in the evaporation process and so they started from such a water droplet and then simulated these in the gas phase water is released and what they observed is that during these evaporation process already the DNA duplex is compact or compact and the size is decreased and starting from these configurations then they finally could get reasonable or good results in comparison to the experimental data. So I want to talk about a bit more on the proton dynamics and the gas phase also aren't DNA. And of course, and by molecular simulations we typically apply a classical force fields to study which then we cannot study a proton transfer reactions. So it's very suitable for the M.M. mythology and and also the previous study and the group of Professor Moroscoe and Paolo Caloni from Ulich, they studied the proton dynamics of single oligotides and the gas phase. So this is, which will also be focus of my results. And the structure as shown here it's a heteronucleotide which has such a hairpin form and accurate solution and consisting of a short BDNA fragment and then here in DNA loop. And then the gas phase one has to localize as I mentioned before charges at the phosphate backbones. And from QMM and up initially and dissimulation the study different hydronation or hydronation patterns and what they observe is that when gas phase protons can be transferred between adjacent phosphate groups, they also can induce structural changes, for example, if these phosphate groups moves up here it can form a different hydron bonding pattern. And yeah, which can all the the shape of the molecule and the confirmation of the molecule which is then related to experimental observations. So what was not studied so far so they studied such isolated molecules. What was not studied so far is the protonation process of the DNA from the counter ions, which are the experimental setup. So DNA is known to be poly ionic with negatively charged phosphate backbones. But the energetics then that will be different the in the gas phase. So, the backbone is significant portion is protonated. And since in the experimental setup there's ammonia acetate the proton transfers likely to come from ammonium ions to the DNA backbone followed by dissociation of NH3. And as a brief illustration I've shown here the model system ammonium ions and dimethyl phosphate, which is known to be fully surveyed from the PGA well as one can see it under solution conditions that if we move to the gas phase. We have just scan the, the transfer of approach on to to the DNA to the DNA, I've made it a redact surface scan and monitor the electronic energy, and we observe in the gas phase of course here, the neutral bone is much more stable and also for small illustration we simulated these with up initially and the molecular dynamics, where the ammonium ion and the diameter of phosphate is in the gas phase, and we simulated this process and due to cool up attraction they meet each other and immediately transfer these this proton. So, we wanted to study the protonation of the DNA and the countering complex along the evaporation process of such DNA molecules. So, first of all, a bit what one has to do it, of course, is that first the proper total work so molecular dynamics and the accurate solution, we re-examined the solution, the actual solution structure of the septal nucleotide and also in our simulations. So with ammonium ion as counter ions and then the experimental conditions and carried out 500 nanosecond classical and these simulations also in collaboration with Modesto Roscoe's group from Barcelona. In fact, we see what we have expected that these hairpin structure here, what's shown here is preserved during the whole simulation. So no strong fluctuations of the RMST values. And then we analyzed also the formation of transient ammonium phosphate pairs. And, as mentioned before, in the solution. There are only just a few transient pairs so that there are no stable complexes formed. And these are just 100 from 50,000 from 5000 structures so the low person region and not long. So I simulated the evaporation process. And I want also to go a bit to the technical setups in my talk. So from our accuracy solution simulations we catch droplets with around the center of mass of the DNA molecule. 500 people second chance of gas phase and be simulations for these gas phase simulation. These can also be done with grow much just this information one has to use a large simulation box and then direct column summation to evaluate electrostatics and that the periodic images are not interacting with each other. So we use this during these computational protocol water molecules are released, which are shown here so at the first stage, it's more or less released linearly with simulation time. And at some stage, when only a few waters are still available. They get much more stickier because of stronger electrostatic interactions and no sudden spinning anymore, and to remove the last one, typically also increase the temperature during this process. The ammonia mines instead, they do not evaporate so they at the first stage they prefer to be in, let's say in the more solvated state but after a specific time and more and more water release, they interact much stronger with the DNA molecule and form afterwards also here so specific phosphate groups and form stable complexes afterwards. And from these snapshots are from these simulations we are then going to analyze proton transfer reactions. The proton transfer between the ammonium and the phosphate groups, and some general considerations, which I've done when I carried out this project was, of course, we have somehow to identify suitable configurations and the descriptors are typically used. Of course, the total number of water molecules, which is also of course related to the time span of the evaporation process or more water is released. Then of course, we selected the one which has, which showed contact ion pairs so the distance between the ammonium ions and the DNA phosphate groups. And as a third one also the local coordination number of ammonium ions to characterizes are much if there are a lot remaining water molecules coordinating with them. We applied them to an MMM. And there are two interfaces, which are developed and by Excel we applied. The first one is mimic which couples a CT and D and grow marks that uses a DFT with a plane with pseudo potential approach. And the second one I will focus on my talk is the grown up CP2K interface, which uses DFT with a mixed Gaussian and plane wave approach and the multi grid implementation. So after we identified suited with that shows one has to think about the QMM mythology. And in the first part of course, which atoms have to be in the QM region. And I've shown you I've highlighted them with with stairs, you can see here of course the ammonium ion is shown to be in the QM region. The proton acceptor, the phosphate group, and then parts of the DNA backbone including this oxy ribosa sugar and some carbons. And red, we have them to cut covalent bonds. And we choose these C4 C5 prime bonds and also the we cut it here along the C1 prime and here the nitrogen atom from the nuclear base. In fact, this one would comprise the QM region. And yeah, we then used, as I mentioned, the grown up CP2K interface with the GPW approach with DFT and PBE functional. And in the first stage, which is one has to do is to get benchmark the QM settings and we bench not be a plane wave cut off the relative cutter for the multi grid approach and also the number of grid of the multi of the multi grids. So during the performance of QMM simulations, which uses hybrid MPI OpenMP, open MP publicization, we benchmark them before starting the simulations. And we find that up to 96 course so two nodes with 12 tasks per nodes and few CPUs per task. So that's the reason of the performance or good performance. And with that set up we were able to simulate 3.5 to 6.0 picoseconds a day, but dependent on the QM size. We have then carried out these QMM simulations and I want to go through this approach. We are minimizing we have already come from MD snapshots with the MP potential and put to first minimize and to adapt to the new QMM potential. So to minimize in a few thousand steps or 2000 steps we use. We then reheated the system from zero Caribbean to 3000 Caribbean and 5000 steps with a step size of 0.5 10 seconds which is typically used in QMM simulations. It really rated afterwards the the system at 300 Kelvin. And then we were interested in scanning the protonation or the proton transfer, or the prefer or we were interested in the preferred position of the of the proton and these DNA ammonium complex. And we would then scan these proton coordinate. For that we use Gromax and CT2K, which was patched with plumes so that one can also apply the functionalities of plume for for bias simulations. And this is a simple proton transfer coordinate, which is the difference distance between the an H bond and the H O bond. And the first step we just scant the these protonation all these coordinate. The first year to a year we moved first to the ionic configurations and then scant the coordinate and part of 2.5 seconds QMM simulations. Finally, we carried out umbrella sampling to get these local protonation or proton transfer profiles with different starting configurations upon these evaporation process. We used 10 equidistant stairs windows and the range of minus 1.0 to 0.8 so minus means it's the other configuration and plus means it's the neutral configurations. And for these, for these 10 different windows we carried out 25 picoseconds QMM simulations to recover the energetic profile along these proton transfer coordinate and what we observed is what we see is that we get ionic configurations, which is what that we have configurations where we have these ionic profiles so that it's not. Yeah, prefer to transfer the proton but we have also observed proton transfer configurations where the proton from the ammonium complex can be easily transferred to the DNA backbone and finally to and then to dissociate. Yeah, these simulations on the final stage of analysis and also prepared now for publication and yet I hope I could show you some interest in details on the proton dynamics between ammonium ion and DNA and the relation to the experimental. Yeah, important. And yeah, with that I want to close my talk. And give over to Arnold, I think. Thank you very much Mirko and thank you very much to me three as well. These nice talks. I think it was nice that we also included some shared some methodological details as well which are not always presented but I think it's very useful for people to see. So we have some questions, which I'll go ahead with now anybody who hasn't asked her question yet please feel free to still do so. So the question is for three comment is from a moment very less talk is a question if the flora for is conveniently bonded to protein, how to deal with it. Okay, the question. Yeah, of course. So basically these two options. Most of the first some proteins actually have the chromophore bone into the embedded into the amino acid chain, because it's formed out of the amino acids. It's autocatalytic uniform and how to deal with it in molecular dynamics you typically define a new residue. And that's how we are doing basically we have amino acid residues database where we have a special residues for the chromophores. Some of the fluorescent proteins, however, there is several like, for example, SFP code, they in them, the chromophores cleaved. So basically it's cleaved I mean it's separated from the protein itself so there is two reaction going to stop the information and then cleaves. Then you have a separate chain basically so basically your protein is forms to change. So that's the question so basically you need a specific special residue for the for the force field, and you need to parameterize and this actually consumes a lot of time this is the main time consuming part, because you need to do this. And I guess I can do the second question. I can read it. Thank you very much. The second question is how much do we have to simulate a system after mutation to see if there is a conformational. Okay, there is a very tricky question because actually it depends on the size of the system. As I showed you, in one of the slides, I typically you need to see at least that your backbone stop fluctuating. And if it starts fluctuating so when you see that your MSD changing, you need to be sure that you captured the transition from one state to another state several times here. So just look into the dynamics look at the dynamical parameters like currently like try to clusterize try to build to see how many populations you see and try to see what is different. What is the transfer time between this population, only after that you can say, well you can say that is it enough or is not enough because molecular dynamics is never enough. So now work be done from 100 nanoseconds up to half of the microseconds basically trajectories. Okay, thanks. I think next question is for me to go. The question is what is the shape of the box of water molecules during salvation is the octahedral. If so, then what's the advantage of using octahedral box rather than cubic. And also does the shape of the water box have any effect on the MD simulation or cure and simulation. So, the box I use the cubic box and my so in the actual solution simulations I think it was asked. And I used one can also use an octahedral box and the advantages of course that you have less water inside less water molecules and that was speed up the simulations, but I was not focusing on the actual solution so these 500 nanoseconds on the small DNA fragment was small enough that I can set up with briefly such a Yeah, such simulations and carry out the simulations with them. I don't know remember it was not much simulation time required. So computation time required. Okay, that makes sense. Thanks. The next question also from you guys. This is about the QMM QMM and the QM treatment parameterization, could you comment on your cutting of the C and covalent bonds, did you do some tests on it. And also simulations where these are inside and where the nuclear base was inside the QM regions. And I did not find any. Yeah, any issue, then cutting this bond. So, I cannot hear that you shouldn't cut by system if it's not a pie. It's not a pie. So, and that reduces of course the computation of course. So, and was already done also. But actually if you do protein simulations never cut over the peptide bond peptide bond is a pie system. In case of proteins you should do that. Okay. Thanks. So the question is, from the same person is, is, is it, the question is, is it grow is it use a plume and patched to get the, you know, the scriptures of the. Yeah. Is it, is it grown access plummet patched, which is possible, or is the CP2k that's plumets patched, which is also possible, is it, or is it both. In other words, where is the blue and patching that you're calling in when you're running the simulations. So, so gromax is an empty driver for these QMM simulations so it's patched with gromax for mimic it's red, red user CTMD red CTMD is molecular dynamics driver that would be patched with CTMD but maybe probably Dimitri would have more information about that. Yes, so I can say that for what Dan Mirka of course it was gromax patched with with with plant but fortunately enough. In 2022 we now have the so called transformation pulling variables, and they can suit the same exactly like as collected variables is by Mirka so you even don't need to patch gromax is planned now to make the same kind of thing, because it's already in gromax, the similar functionality is already inside the gromax from 2022. Okay, then a question from Mirka, why are we simulating at 300 Kelvin shouldn't we be simulating a physiological temperature 310 Kelvin. Yeah, it's, yeah that's a good question. Also related to mass spectrometry it's not really clear what temperature it has to be so typically I think it's the experiments often set up at 300 Kelvin so I decided to go to 300 Kelvin. Yeah, okay. Then the question for both do we need to install gromax and double precision to use along with CP2k for the question for Dimitri actually. Yes, because I'm actually doing a certain interface but yeah. So, basically, typically I would say that yes, you need to install the gromax in double precision if you want to. So, typically QM is very sensitive to the precision the problem is not the gromax the problem is QM. The problem is very sensitive and the single precision is typically not enough to single precision of coordinates, which will provide by Gromax. So yes, you need to do a gromax and double precision. You don't need, you can do it in single precision but I really strongly suggest to compile these double precision. So one question from that I think is double precision is slow with GPU but I think that's a point about perhaps about gromax where it's really the performance here depends on the CP2k. Okay, for CP2k so. So maybe Arna you can say more about GPU and CP2k. Yeah, in the next webinar which I will mention it. Yes, so in two weeks will be a webinar where it will be discussed specifically the scaling because I think it's for QMM your performance is not almost not dependent on the gromax it's fully dependent on the QM code. And it depends on how effectively your QM code can use GPUs and it will be in the next webinar I guess yeah. Okay, then Alexis, thanks for the nice talk. And this is the question is from Mirko is you mentioned two scenarios from your umbrella sampling simulations, namely, double well potential, almost equal potential wells, and Ionic with one minimum and no proton transfer. Can you please explain what are the prerequisites of each of them to take place. Do you obtain them under the different initial conditions. Which I've shown with the descriptor so different amount of water molecules across the round, and also in the actions of the ammonium iron with not with a single phosphate group, maybe also with a second phosphate group which was close by. Okay. A question for I think this one is also for the Mirko which is what is the time step you used for the QMM simulations. Okay, thank you very much. Those are all the questions we have from the audience. I have one more question for both of you which is that how did you choose your, why did you choose the functional you chose. You said you choose, how did you choose the QM system size, you that you medical you said you did some benchmarking. So you change, you grew and shrunk this. But if you have any comments that you think are useful tips for people. Yeah. Okay, of course, benchmarking everything as I've shown that's always a good idea. And of course living in the literature of similar studies have been done and getting some inspirations and some information on. Which type of function is to use for the specific problem or which QM sizes we try it there. Also a lot of studies them. Yeah depends on the problem of course. Yeah, I cannot hear that, especially if you are PhD student or whatever to try to develop the bicycle, most probably a system already been studied and just look into the literature. And only in the case if you see that no one ever studied similar systems, then do a full scale benchmark of your functional of your basic set and size of your QM system. And for that we have a very good last year we have a very good webinar series on the QMM best practices where the professional high level scientists and that field that talked how to benchmark which of this thing. Yeah, I've just put a link in chat everyone with the link to that workshop. Okay, thank you both again very much for your talks. And thanks everyone for attending I just want to share some details about the next webinar that is happening. So I stopped sharing right. Over over taken over. So, the next webinar that's happening about so webinar to its time is related to this. It's by my my colleague here the PCC Holy Judge and myself, talking about efficient usage of chromax with C2K, C2K to do QMM simulation of biomechanical systems so we will look at this in a little bit more detail about the performance that you can get, and how you use both different kind of processors and GPUs. So hopefully that will be of interest to people. So with that, I wanted to thank everybody again for attending. I hope you enjoyed this. See you at the next bio sub webinar. Goodbye everyone. Bye everyone.