Bioexcel webinar #96: Multiscale simulation of biomembranes: shape, structure and cellular function

So welcome everybody to the bio excel webinar number 96. Today we will we will speak about multiscale simulation of biommembrane shape structure and cellular function. The presenter of today is Vera Percy Shan is coming from the Nilsborn Institute University of Kagen Denmark. I'm Alexandra Villa and together with Otto Anderson and Richard Norman I'm hosting this webinar in the name of by Excel the center of excellent for computational biomolecular research. The webinar is recorded. During the webinar, you can use the function Q&A that is at the bottom of the application of the Zoom application to ask question. Please type your question and uh and if you have a microphone then I will unmute you so you can ask the question directly. If you don't have a microphone please write no micro no microphone and then I will read the question. for you. Maybe you still have some question after the webinar. So you're most welcome to join us on the ask by Excel. Forum h in the category by Excel webinar. You can select the the topic connected to the webinar of Vera and then ask extra question. He will be available the next week. So something about this presenter of today Vera has a master degree in condensed matter physics and a PhD in molecular biohysic he got his PhD in university in Denmark. Then after the PhD he moved for a posttock in the Netherlands. Exactly was working in the group of severing in Grooning University. After that he went back to Denmark and he built his own group. currently is a group leader at the Nielsborn Institute and University of Copenhagen and his group is developing and apply multiscale computer simulation method to explore the cellular form. In 2024, he was awarded of the prize Frank Blenny Adward that is awarded by the molecular graphics and modeling society and usually is awarded to a young researcher for significant contribution of the field of molecular modeling. And now we are very curious to listen what he will tell us about multiscale. Uh hello everyone. First of all, I would like to thank Alexandra for the introduction and also for the committee for inviting me. Uh so my name is Uria Positionian. So it's a very difficult name. So please don't judge anybody for not not being able to pronounce it. And uh for today I will talk about aspect of my research in particular. I think it fits to the scheme of this uh web webinar series. And if you have any question later on, of course, I will be available. Okay, first of all, uh just to show you my group. Here is my group. Uh this is a cold weather in August University. We went for a Danish molecular dynamics day actually and it happened every year and this is all my PhD and postto but we also have master student and bachelor student as well and even abroad from Sweden it's often join us and also this is foundation that's supporting my research in particular uh the independent research fund Denmark and no this foundation for the s I would like to thank them for their generous support of my research search. Okay. So, u my group also de one of the focus of my group is to develop methods and tool to push the boundaries of molecular stimulation uh to the to the new level of uh complexity and also scale. And one of the focus area of us is to actually study the collective effect collective behaviors of biomolecles. For example, uh molecular dynamic simulation and many other computational modeling including alpha and others has really shaped our understanding of single mole and even few molecule together. But biology is also beyond that. For example, we have they have a high complexity and of course this one is a single molecule simulation. But what happen if you have many of them? For example, this is a model a simplified model of such a system that this protein are assumed to be non-interacting. So they do not interact with one another. This small red balls. But you see they class them. They are on the membrane. They do not interact with one another. they only locally rigidify the membrane and this leads to their classing. So it's something beyond the single molecule effect and biology works like this. So actually to understand biology we have to move on to this area and it's also important and because it's also very timely to actually move to this direction because specifically because of the advances in experimental methods. We have now we now have access to data from res super resolution microscopy that provide that with detail actually um level with the level of detail show us the dynamic. We have electron microscopy all these cryo em some of them in C2 and they provided with the high resolution structure of a very large system. We also have cross-linking mass spectroscopy for example that gave us actually the interaction between this protein in a scale of a whole cells or mitochondria. All this together actually calls for methodology for a stimulation community to go beyond the single molecule and looking at the collective effect and eventually reaching the whole cell modeling. Uh so here at in particular for this talk I want to talk about how we deal with this with this kind of collective effect in a context of biomemembrane. So biomemembrane we know they are very important. Of course we know that they define the boundary of the cell through the plasma membrane and divide the internal part into different organels. However, of course m membranes are involved in many cellular processes and they are not as simple as this. They actually form a highly complex both structure and composition. And uh to just put it in number we should know that up to 30% of our dry cell mass is just membranes and also up to 30% of our genome in con membrane proteins. So all together tells us that biommembranes are very important but we should not look at them as a in a scale of a single lipids. We should go beyond that. So just to give you what here is more focus on shape does really shape matter or is just a result of something. Uh so here is a few example of mitochondria in different condition. For example you see the in um in fruitfly the young fruitfly has a completely distinct morphology of mitochondria compared to the old one. And also with the in the mice after paradoxical sleep deprivation you could see also a distinct morphology of the mito mitochondria in your membrane compared to the one the control one and even in neurons the one with high activity exhibit distinct morphology of mitochondria compared to the one with lower activity with spiking activity. So all this together tells us that membrane shape also not only is important for many cellular processes but it also gives us information about the whole cell and even the organism. So just to come to the story I want to show you one of the example that we dealt with was the case of sheigga toxin. So she toxin is the AB protein. The A subunit is a toxic part and the B subunit delivered or takes the A subunit into the cell. Okay, one can even ignore the A subunit because B the B part can also enter the cell and it can cause diarrhea is associated with diarrhea bloody diarrhea. But the question is for example how this can cross membrane or plasma membrane which is quite a big object. So the this question more or less was in an experimental setting was quite clarified or at least shown that it can enter the cell independent of any of um and other endocytos pathways. For example, cladine part a pathway uh which they showed in a this is in a model membrane they show it binds to the cell sheaxin it can cluster and then it form this tubular narrow tubular invagination which in the in the context of the real cell and active protein like dynamine come and cut the neck and then many toxin wrap around the membrane or a piece of plasma membrane and enter the cell. So in this mechanism we know how the binding happens because we know the cellular receptor for this protein for this toxin is actually GB3. Uh this is a glyolippids and we know the toxin can bind up to 15 of them which yeah make a strong binding to the cell surface. However, what what about the rest? How the clustering happen, how the invagination happen and to study this, this is a complex uh problem for computation uh in for simulation because there are many lengths and time scale involved. First of all, single protein is about 10 nanometer size while the tube diameter around 15 nanometer size and the tube length is goes up to 1 micrometer. So there is no distinct uh a single computer simulation that can actually capture all this scale. So I try to work on a like while well-connected multiscale computer simulation technique that we use all automar dynamic simulation to uh look at how single protein binds. then coarse graven simulation to actually look at how two of them interact with one another and make a cluster for example and then allow the result of these two modeling tech thinking into a misoscale to uh look how the tubular imagination forms. So as you can see this is very similar to what we see in experimental setting and then uh we uh so this gives yeah so this uh setup gave me an idea of actually building a more building a multiscale simulation a more versatile one to actually deal with the complex biological membranes in particular intracellular one and uh so Okay. Yeah. Okay. Sorry. Yeah. Okay. By the way, this is also another part that uh allows for back mapping of the structure into the molecular scale and then have the molecular resolution of the tubes. And of course, we could obtain a lot of details and how these processes happen all of them by studying every level of these simulation. But overall, this problem gave me an idea to actually build a multiscale computer simulation techniques. a bottom up that you start with the normal molecular dynamic simulation and all the way go up to the miso scale obtain membrane remodeling and then go back back map into coarse grain level and obtain bucular organization also in the remodel structure. So this setup allows us to actually deal with the time scale gap in molecular simulation which is a actually a hard problem. Lenga scale gap we are now experiencing that um there is cell modeling the whole cell modeling is happening but the time scale is also an issue and it won't be resolved anytime soon. So we need such a setup and in this setup we um uh so in my group and with in collaboration with other we actually further um further extend it now we can map back and also even bias it by experiment 3D EM imag imaging that actually perform bias simulation at the miso scale. So in this talk I will be focusing on um the methodology mostly in particular this miscale simulation model 3DTS that the both the code and model is develop and maintain in my group and also the TS to CG the back mapping procedure that allows you to go from misoscale to coarse grain and also allows you to build in principle any complex membrane simulation uh membrane uh initialized structure for molecular dynamics simulation and then at the end I will come back to this HMF algorithm that allows you to directly run simulation bias simulation by experimental data and also help to interpret uh the experimental data more correctly. So uh this will be my focus here. First let's go with the free DTS. So before doing that I want to also define misoscale as one can say in in cells there is really not a separation of scale happening. So therefore we define this methodology this misoscale in the term of methodology and so what we I mean by a misoscale simulation is that it's a model that uses that represent big biomolecolle explicitly as a particle while the smaller one they will be as a constraint or as a collective. So a small molecule like lipid is not represented explicitly while something like protein are represented explicitly. Uh so the first one is 3DTS is this software that is available in our GitHub and you can use and also do it. And the way it's set up the membrane part is model as a triangulated mesh that is contain uh consists of some uh vertices connected by edges u and creates this triangulation and they are is a dynamically triangulated surfaces which means these edges can break and flip. So therefore you have you give fluidity to the surface. So it's not just a elastic surface but is a fluid surface that is required to model membrane and through a set of discretized geometric operation we and on each vertex we can obtain uh principal curvatures and a normal and each vertex represents a patch of membrane consist of hundreds of lipids. So eventually we evolve these vertexes by our simulation method. But the vertex has a characteristic of a surface element. It does is not just a particle. So it has multiple degrees of freedom uh not just the position but also the surface area and also the curvature. And each vertex energy is uh coupled to such equation which is eventually harmonic uh potential of um curvatures. And this is known as a halfish Hamiltonian in the context of membrane uh theoretical membrane modeling. And uh so a membrane is only have three model parameter a bending rigidity a gaussian modulus and a spontaneous curvature that characterize that um a tendency of the membrane to bend toward to a specific direction and it could be it can be set to zero in most of the cases. Then we have some global constraint like osmotic pressure or membrane tension if you want to study something in periodic boundary condition and more. So there is a manual that one can look at that. Then the main part comes the prin. So uh each vertex can be also uh can we can assign a vector an inplane vector to each vertex. So it plays as a protein role and in this uh this vector can jump from one vertex to another and also modifies the local uh mechanical local mechanical properties of each vertex locally. So in this context we can have two type of proteins one that break in plane symmetry and one that is in the plane of the vertex is symmetric. So for modeling this we bias the simulation with this potential when a vertex uh attach or when a protein goes on a vertex on a surface element and it requires four model parameters and when it's symmetric it requires only three model parameters and uh this model parameter in principle can be obtained from molecular dynamic simulation. So you run a single molec called a single protein simulation with membrane and then you have you analyze the data to obtain these parameters and then you can load it into this kind of methodology. Then we also have a protein protein interaction that consists of three part. The first part is just at um attraction low very short range attraction. If they are close by they interact. If they are not the interaction is zero. Then the another part which is the most complex one is actually this term which uh which you have to translate this proteins through the geoysis of the surface. So you consider the curviness of the surface and this is a complex part but this allow you to correctly model the create this kind of misoscale modeling. This is a reduced dimension but it make the mathematic behind it more complex. And then also we have another term which eventually considers that if two pro the two two protein come in close proximity then they want to bend the membrane. This is seen for example in the um uh in the ATP centase when you have a dimer of them suddenly they bend the membrane but individual of them do not bend membrane as much. Uh so to just show you a few examples. So this one uh first of all the simulation procedure correctly capture that this shape transition of a very simple vesicles the fl it can capture the fluctuation of simple membrane and also membrane containing inclusions. It can uh also validate the fluctuation tension and me input tension are the same which is these are more theoretically trackable result that we have about membrane. But our simulation shows that it can get all this result and also that nanot tube formation by pooling that also is very common by a physical experiment is done that it also captured that one well but also we can also study much more diverse range of processes that are not as easy to study in the theoretical modeling. So we can also include fluctuation. Of course this is for example if you study something like er something like we have a transitional next also we have a back map example you can study face separation lipid face separation in both course grain level and also um in the misoscale level and make a link between them. We have viruses simulation and then you can go back to the course level and all that. Yeah. So the code is over here and we anytime that we make more uh place but we do not change the older version. So that be consistent with uh every there is a set of manuals and also many tutorials that you available to actually answer your question and if you have a something new in mind you can just either contribute directly or also ask us to develop. We also have um a policy that whenever we use this software to publish a paper after the publication we write a tutorial for how exactly do every part of uh the result in the manuscript. Okay. So to show you one of the showcases for example that now you actually have go to the GitHub and you can obtain the result in this work but it's about nuclear envelope. So nuclear envelope can be considered as a one single membrane sheet in a high topological genus which is look like which is we call it stamatite. Uh although they are often considered to be double membrane system that um this nuclear uh pore complex connect the both membrane together but in principle in mathematic way or in physical way you can also consider them as a single membrane sheet that these protein are uh assembled around uh this necks and uh yeah so uh so if you can see it it will be like this and these are the next that the complex assemble around it. But we ask different question because it's not easy to ask question about single molec interaction in misoscale because that's the model parameter. But we could ask one more another thing. What happen if suddenly the in internal uh osmotic pressure increases or the tension of the membrane increases? the same thing. What happened to the size of this neck? And what we found was actually that there is a critical uh size or that is also size related to the membrane tension and the rigidity in which uh above that suddenly the neck constrict uh below that if your tension is below that it will constrict and if it's above that it suddenly dilates and interestingly even if you have a protein complex around it This is just because of visualization that is outside it should be also inside but um if you have a protein complex around it that doesn't change the scenario below certain threshold but increasing osmotic pressure or tension it all lead to the constriction and then above a threshold suddenly start to dilate. is actually has also uh also if you look at system with more of these poor you still see the similar phenomena of course with a much more diverse uh range of uh possibilities but it's the same principle and interestingly something like this both the dilation and um constriction have been observed in experimental setting which is quite interesting here and then another one was I showed you this one in the beginning that um you have some protein that do not interact with one another but then suddenly they form cluster because they locally rigidify the membrane and this is called fluctuation induced protein clustering which is quite uh non phenomena but uh yeah it also applies here and one interesting aspect here something that even you can't uh really look at it theoretically is if you have different type of this protein and what we see is of course if you have a two type of protein one very rigid and one a little bit softer. The rigid one attract each other but then suddenly the rigid one also attract the more softer one compared to the soft stuff interaction which means that rigid protein can form a cluster center and lead to the nucleation of a cluster formation in a biological system and all this emerges from the fluctuation of the membranes. Okay. So the second one is the TS2CG software which is actually to backmap structures from misoscale to uh coarse grain level. So this was initially the main idea was initially developed as this that you have a such a triangulated surface you can create both monollayer of a billayer and then you have a certain mathematical rule that you keep so you extend this surface so that that the geometrical properties of the original surface remain constant. So you keep the curvature at each vertex the same and then you add more uh points and then extend the surface. But this is important because the mechanic of the membrane come from the bending rigidity and the curvature and therefore you have to maintain this while developing the surface farther and uh it consists of multiple step and all the way that you add the protein you find the a lot of points then you add the protein and then you add the lipid and the protein of on the uh points but of course not all the point will will be assigned a lipid it only with the certain uh projected area or with the area assigned to each lip. So the concentration will be as you defining the input. So uh these points which they have also orientation uh only find the place for lipid placement but does not necessarily a lipid will not be assigned to it necessarily. So it this was the original version and it consists of two different uh script or software uh pointalism which eventually takes a triangulated surface and gives you points with orientation, normals, curvature and everything. And then the second one which is a membrane builder which is a script pcg uh which plays the routing for you based on the defined input file and also the lipid based on the composition defined. We also have another script that called salt to solve this kind of system in a fast way. Uh in particular if you are dealing with the something in the level of cell size. So one of the example is was this case for example we had the misoscale simulation of this tubular invagination and then we use TS2CG we backmap it we run molecular dynamic simulation and we obtain organization of biomolecol in such a curved system which is probably this takes forever if you wanted to start um from the scratch and see to observe the whole imagination. So this allows you to somehow uh accelerate the simulation time step in a different uh in this scheme or another case was actually to build uh SARS COV 2 envelope structure for molecular dynamic simulation. This is very challenging as you can see building this system often you can lead to the collusions and then uh explosion of your simulation because of uh u because of very close contact of the bit. But what you could do is start a misoscale simulation of the system define everything. You run the simulation you obtain the right organization of the protein which only takes half an hour on your laptop. then you can backmap it to multiplar scale and run a multiplier dynamic simulation which is quite stable and nicely done. Yeah. Uh then in the second version we actually thought about it okay this is very good to building a structure why not just make it also something structure builder or initial coordinate builder for molecular dynamic simulation. So we um in between the um the pointalism script and PCG membrane builder script we created a a point class. This is a Python script that allows you to define different different thing in the local in the on the surface. It gives you for example you can define inclusion, exclusion, holes uh different uh domains to and so on so forth. then eventually allows you to build membrane with any shape and any composition given enough information and uh to just show you how it works. So the pointalism script uh divide this surface when it's extended triangulated surface divide it into domains which each domain consist of collection of points which each point is for example it's a position area and all this but the point is uh you can define uh define domains and reallocate the point to different domains and each domain inside your input file get a specific specific composition. So with this you can uh create any domain you can create many domains with any shape and get get them to have a specific composition that you want. And uh yeah and this uh this new point class that we have that you can use it actually already has several uh predefined script in it. You can also define your own one in it. Uh first of all it allows you uh to create a domain uh at specific domain around a specific inclusion. Inclusion means a protein. So you have a protein you want to make a specific domain around it with a specific lipid composition. You can define it. You can um uh for example say oh I don't really know how to distribute my lipids but I know this lipid prefer curvature the negative curvature that one prefer positive. You can give this information and also it's placed based on the curvature for you. It can do do the same based on protein for you. But you can also define your own way of lipid lipid and protein placement and it does it for you. And based on this we actually been able to do many very complex shape. This is one obtained from uh realistic experimental uh data. is a whole mitochondrial membrane that we use cryet data to actually build this structure. So the size and the compos uh composition is uh quite realistic. Of course this is only two nancond without all the protein. Of course it will disassemble if or but it is simulatable. It's stable but not for long but we can have all this complex composition. This is a a segment of our representative of a crist membrane that you can see the lipids and the protein are placed based on their curvature preference. We to show this we even uh uh try to have a globe. So Martini globe to actually make domain based on the lipid composition. Exactly. So have a global uh the world map and have each continent have a different lipid domain. And we even started to some non-physical membrane and still can build it. And all of these are simulatable of course. And also we were able to make a martini exhibition that you can actually create artifacts just for fun. And then they are also very they are numerically stable. Of course if you run for long they will deform because um membrane mechanical membrane do not allow for such a shapes generally. Okay. So TS2CG also could be found in this um in on our GitHub and there is also a list of large actually a large number of tutorials available that tells you how to do everyone from a simple to a very complex system and uh our team also is available to answer your question and also we are very open to get your uh your suggestion if you think something can be done or something new feature will be helpful. We will be happy to consider them and uh possibly add them in the next uh round of in the next uh version. Okay. And uh then the last one is HMF algorithm. So HMF uh so it's not a new software al although it is also implemented in a new software which is in my collaborator Yan Konski lab they have done it but it is also implemented within 3DTS and allows you to actually get system like this data like this and bias your simulation based on such a data first obtain the missing part but also just to see how stable this kind of a structure is And uh most of the work of course is done by Valentine and uh so yeah so just to give you what we do is that a 3D image data we take it and we create a um from their cloud points we create a potential based on the density of the points and we bias our simulation with these potential. So this will be a coupling potent uh coupling parameter then this will be an EM or ET data and this will be the potential of our simulation. So the larger the uh this coupling parameter is uh the simulation tend to go and follow the experimental data however the smaller it is the uh simulation simulation potential um dominates and perform this. So this allows you to bias the simulation in particular if you don't have all the information if we are dealing with cellular system we don't have all the information about where the protein or a force applied to it. So that this help you that use the experimental data to define a new boundary for it and then also run the simulation which is a physic base to obtain this and just uh yeah this also extend our multiscaling from a top down. So we go from a 3D imaging go to misoscale and from misoscale you go to coarse graining level and just to give you an example here is for the influenza a virus. So we use the cryo um uh data and then our simulation based on a little bit helping with the simulation of course to to adjust the parameter can find the membrane surface correctly which many segmentation tool may not be able to and then run the simulation go back all the way to coarse grain level and run simulation at the molecular scale. So some seem completely top down from uh experiment 3D EM imag imaging all the way to molecular scale. And this also yeah as I said this uh HMF could also be run or with many refinement tools inside the mosaic uh that is developed by Valentine and you may also try it and even I think uh you can load free DTS immediately from it also to run but of course HMF also is available on free DTS alone. So just to see how so in the past when we were modeling biommembranes we were for simple system it was all our modeling was based on simple system a larger scale we will look at halfish Hamiltonian which was capturing actually the shape of single component system very nicely and at the molecular scale we will dealing with the concept like shape factor but nowadays this are all dominated by molecular dynamics. simulation and also miser scale simulation that allows for realistic complexity and much better solution and one can expect that in in even now and in a close future we will see active me simulation of active membrane and also in C2 condition like what we see in a wholesale modeling system that you have the environment also around the membrane to be realistic not only the composition of the membrane itself Uh yeah. So also something about if we are dealing with this complex system one could say okay how could we calibrate this misual scale modeling because we can't maybe it's too expensive to run all this molecular dynamic simulation to calibrate them. But one option also is of course uh experimental data. For example, in this case, this is a somehow proof of concept. Of course, it also has some finding which we use cryo trust linking mass spectroscopy data to build a simple model of a mitochondrial system which we capture a small uh part of it with the different layer of the protein which is represent as part of the membrane. And then we use this cross linking data to find that the that tune the interaction between different this uh different protein and also based on uh uh alpha fold information on the protein size and we also define the size for them and run a simple simulation for these cases and found for example certain cluster formation or um formation. And what we found actually was that uh these two protein uh they were often found uh it was ambiguous that are they in the cristine membrane or in me um in membrane part of the uh and in a boundary membrane. And what we found was they are actually more in the Christ one. And uh this was also confirmed by our experimental collaborator of course but the conclusion here is that yeah to go forward we also don't have to rely on the completely bottomup approach. We can also use a top-down approach and calibrate this system from experimental setting. Okay. At the end I would like to first of all thanks all my collaborator um uh collaborator in particular in this projects that I show you C Yan Maring and uh for many different project and also Yan Steven for his contribution to TS2CG version two. uh John Ipsson for uh being involved in the uh 3DTS development and Yan Kunski for his contribution to uh HMF and in particular Valentine that he did the major work there and also also all the other collaborator to being involved in different aspect of this projects and this is my groups uh and also the foundation that has supported my research. church and thank you all for your attention. >> Thank you very much. That was a great Sorry, do you hear me better now? Thank you very much. It was a great presentation. I really enjoy I I while I'm I see that we have some question probably I have one things that I was asking you. Sorry. Uh could I share your you mention a lot of time your GitHub repo? >> Yeah. Can I share in the chat the main the main repo? So just get up with your name is correct that one. It's shall I share in the chat? >> Yeah. Yeah. Yeah. I can share share it myself also if you want. >> But you cannot share with everybody. You can just send to me if you want. You can send to me and then I share it with everybody. >> That's true. >> Yeah. Okay. Here is >> Yes. So in the meantime uh >> and >> yes thank you. So I will I will share to everybody. So here you have the GitHub that uh was presented by may be more than one time so that uh yeah >> yeah that yes >> you you should see and then is the second one. So we have two. Yeah. Yes. Thank you. Yes, thank you very much. So, we have a question for Fat, but I will try to unmute. I just uh I just Sorry, I just lost the control. So, let me see if uh as uh uh yeah, I try I allowed you to speak. Could you speak please? Otherwise, let me know if you don't have the microphone. Oh, sorry. Okay, thank you. So, I will read the question. Hello, thank you for your presentation. It was very interesting. I was wondering what are the approximate box size in is in the messoscale simulation. How long it typically takes to obtain results on a supercomput. Uh so the misos scale simulation for something around 2 to 300 nanometer system you need a single CPU within two days and you obtain good results but if you wanted to do a bigger system of courseh you can also use openmpp paralization scheme which you need to allocate this and it's more or less linear but for most of our result actually that we have published we are using the single core uh and but of course with many replicas and often two days you can get up to 10 million uh steps which is enough for yeah for enough sampling of these systems. Yeah. >> Okay. Thank you. Now I I fat is thank you to say thank you. Interesting. So then I will unmute Vish that maybe he can speak. He has a microphone. Please could you could you >> Hello. >> ask your question if you would like. >> Hello. >> Sure. Thank you. Uh thank you for the lecture. Uh I would like to know if you can model the membrane being affected by Ross. >> By Ross you mean the ROS as a protein? Yeah. >> Uh reactive oxen species. >> So can you tell me what do they do to membranes? I I'm not familiar with this. Yeah, >> I might be wrong, but I think uh the oxygen atom has like a sort of like an unstable electron that binds to the carbon in the lipid. Uh so what I should tell you that to actually see if the at least the miso scale that I am working with uh can help you or not is that you should run a coarse grain like martini or all atom simulation and see if this uh molecules or any of those has a major effect on the membrane. When it does in particular membrane shape or composition when it does then you can allow this data into the misoscale to see if it has if it be useful even to model with because if it is uh um uh what if it is insensitive to any membrane is insensitive to this stuff or any change then uh they are not really useful. The point is you should know something what what do you want to know and why do you want to use this method to use it not just um for example it's not like molecular dynamic simulation that you say I wanted to know how this molecule interact with each other here you should know what you are looking for as well >> okay thank you >> thank you very much so now there are no other question but I I have I have also a question. How do you rank uh how is uh complex to build a system with your model? >> You mean how complex is in term of size or >> size? How much time you have to dedicate and so I saw that you make a tutorial for any type of system. So assume there is also a challenge in building the system not only in sim in >> in modeling the system what I mean built is really to model the system and then you have the more part the pre-processing part should we say and then there is the processing part the data generation then the post-processing yeah >> yeah so to start simulation uh so we have a lot of scripts to build setup for you and also from input file you can decide what you want and to build it for you as well. this is uh so most of it are available but if you suddenly come with something that I haven't thought about it uh then you require to code a little bit to actually define the especially building a triangulated surface in a specific topology or geometry beyond what we have could be challenging because it has to satisfy certain requirement but apparently if I I haven't tried it yet but uh Valentine has told me that if you go with the blender give a shape go back to HM uh to mosaic it can gives you something that free DTS can easily handle so in principle is not very difficult to deal with no yeah >> okay so if you want to how much you expect user to spend time in the prep in preparing it in running and in post analyzing it how you will share percentage >> if you have the old project I will spend 10% or 5% in the first part and what is the >> I think if I could say that you one should only spend uh 5% of their time for preparation uh not much time in running of course running depend on how much simulation you want to run but most of on analyzing because this simulation are good for sampling so you can run many simulation and then combine the result and look for what you want. Do you get this? So it's about interpretation of your result is the main challenging part than just running simulation. >> Okay. Okay. So it seems comparable the the prepurching and the running of the simulation as time. >> Yeah. But it's not as difficult in molecular dynamic simulation. If you don't have a good structure then you are really in trouble and you have to spend a lot of time here is much easier to start. >> Okay. I was asked uh we have maybe other no this is uh yes so we don't have other question so I don't know if other people has question now otherwise I will move on to just make the announcement. Okay, that I Oh, we might have one still. Oh, yeah. So, we have a question from Eager. Eager, I will try to unmute you. Um, oh, you can ask your question. Oh, no. Sorry. You say that you don't have a microphone. I read laughter. Okay, I will ask I will read the question. I would like to know if I can use the model to study hydrophobic compound that bind to the membrane like chemically modified steroid molecules. The objective is to see membrane alteration and possible diffusion to the cytoplasm. >> Um yeah this is a bit um it requires a little bit more knowledge in the molecular scale. For example, if uh you could study how rate can change the membrane shape and composition during this kind of these processes, but it's not as straightforward as you one think that we can do this. So you should have more knowledge if you wanted to uh this which kind of changes make to the membrane. If it's uh just crossing the membrane like permeation process then I don't think it's a good idea to actually use this model at all. uh but if you want if it they go through a process like intoytosis then this model is very robust and strong for such a processes. >> Okay. Yeah. Thank you very much. So now I share my screen to to get the final. So I want just to announce the following webinar. We had a webinar that will be the 28th of April. This webinar was supposed to be at the end of March but he has been postponed. And so Etsy Karaka and Bersin Barlas will speak about Binadin and it will be the 28th of April. uh and then after that 12 of Maine we will move on with David Sancho that will speak about the driving force in molecular condensate from atomistic simulation of model peptide. Yes. And here you can see with a barcode all the other webinar that we have this spring. And finally and not least, I want to remind that we have a bio excel conference from the 27th to the 3rd of September in Bernno. You are most welcome to register. Please come. We have several speaker and we are have open the call for abstract is still open. Please enroll. And finally I want to ask thanks a lot Vera to be with us. Thank you very much and thank you for everybody for attending this webinar and we see in two weeks. Okay, bye-bye.