 Good afternoon everyone and welcome to the next edition of the BioExcel webinar series. My name is Rossin Apostolouf and I will be today's host. BioExcel Centre of Excellence, as you might know is the leading centre of excellence for computational bio-molecular research in Europe, and through the webinar series we are aiming to feature notable scientists at their work, some of the novel tools in the domain of bio-molecular research, and anything exciting happening in the domain. And today is our pleasure to give you a webinar about how to understand better the parameters used by Gromax, one of the very popular engines for liquid dynamic simulations. Gromax is one of the core applications in BioExcel, and a lot of the development is running in the centre. And with that I'd like to present to everyone Christian Blau, who is a researcher at the KTHR Institute of Technology in Sweden. He received his PhD with Helmut Grumiller at Max Planck Institute for Physical Chemistry in Göttingen, and for four years he's been working as a researcher in Stockholm, currently in the group of Erik Lindow in his part of the core developer team of Gromax. His main work is on Ramos development, but he's also interested in driving simulations using experimental data, in particular cryo-electron microscopy studies. So welcome Christian, and pleasure to have you here. Yeah, thank you, Rossen, for the very kind introduction. Today's topic was picked due to lots of users expressing their concern about their simulation setups, questions whether they set up their simulations correctly. And this is an attempt to answer to these questions, and of course, if we are extremely thorough with everything, we would sit for another 20 hours, maybe, going through all the methods and all the details, going through all MD, but I still want to give you a basic understanding of what is important when you run your simulations. So what do we actually want from Gromax or our simulation setups? We want to take a structure, add a force field and some additional parameters. And from this via simulation, go to some biophysically, chemically relevant results via an analysis step. So it's quite a simple setup of very few conceptual, pragmatic ingredients here. And these things you see reflected in Gromax, different types of file formats. So you'll have the PDP and GROW files that give you the structure. You have the force field that is describing the interatomic interactions and so on in the top files and external parameters in the MDP file. So this makes for quite a simple split of things. And so far, this I think is a relatively easy to understand issue. The moment things get complicated, maybe, and maybe the first step where I want to clarify some things is when we look a bit deeper in what the parameter, the MDP option file actually does to your simulation and also to your analysis. So what we do in Gromax as a first thing is we gather all the things that make a simulation and pre-process them into a TPR file. The Gromax pre-processor does this, just gathering and lumping together all types of information. And from that on, we start the simulation. And we have a look at what the parameters in the MDP file actually influence. Then we see that, and what your simulation actually influences. We see that there's a bit of a spillage of what the parameters in your MDP file actually affect. So we observe that some things in the parameter file then actually do interplay with your force field. That's an extremely important thing to take note of. And I think one of the main reasons for concern for some users on what parameters to set and how to set them correctly. Then we see that some parameters don't really go so much into the analysis and results part, but rather how long you have to wait for them. That is a parameter that purely affects the performance of your run. But not much of the physical treatment of your system. And then there's parameters that just affect how much of your run you actually get to see in the analysis. And depending on what type of analysis you do, you might also see an effect only up to the analysis, but not in the results section than when you, for example, do an analysis race. Every other step of your simulation. And then as an extra thing, and I just added this for the completeness here, we of course have other options to the run that are not part of the TPR file, that are the command line options to the MD run command that also affect your simulation. That is something that will go into yet another webinar topic I will not cover today. So let's have a look at the parameters in more detail as promised. So the easiest thing we can do when setting up a grommet simulation, I think quite instructive thing to do is to just give no parameters at all. So it really does work to run grommet pp, but the completely empty parameter file that has nothing in it. It's just really a file parameters.mdp and the run the pre-processing tool. And what you observe there is that will give you an answer, you know, nevertheless, and also generate a TPR file. And that TPR files parameters are reflected in the mdout.mdp. Now that is the first thing I suggest you to look at, because this will also tell you how we interpreted your parameters and what kind of default parameters have been set in your run. And what I'll stick through throughout the talk is showing what happens when you just don't set anything. So what are the default parameters? What do you get as a kind of don't set anything? And then trying to discuss these things, saying which parameters you have to think about, which parameters you should set, what parameters are quite okay, always depending on your science, of course, to keep on the defaults. So let's look at what you get when you run such a default file. Header information is not to be neglected because this makes sure that you are looking at the right file, you get quite some information about how you generated the specific file. And you get to learn if you receive, for example, foreign mdp file, some information about how this whole thing was generated. Then we have another part that is called various pre-processing options here, on the right hand side, where you see the pre-processing information, for example. This goes into the top part of your simulation setup, that is the topology, the interactions can be altered to some extent by these pre-processing options. And something to look into if you have modified force fields or want to do very special things and useful if you want to use restraints, usually something for vanilla and the runs as we look in here right now, something we would not consider more depth exactly right now. Then comes one of the things I want to really hint you to, and this is just the easy device I can give you always is reading the manual, and you find this under the web address given here, manualgromix.org, current user guide mdp options. And if you replace current with the Gromix version you're using, you also get the manual for that version. But of course we always urge you to use the most current stable version of Gromix, and that's exactly where you get an overview over all the mdp options available with some explanatory text. Now let's have a look at the first really decisive mdp option, maybe the most decisive one on the whole run, that is what type of integrated you use, that is how do you want to propagate your atom coordinates in your simulation. And there's a overall six base types of ways you can do this, and most of the time we only see two of them in practice, that is energy minimization and nuisance equations of motion we want to solve. So then other features you might see are stochastic integrators that use loads of undynamics, so they say stochastic dynamics and Brownian dynamics integrated here. You can perform normal mode analysis with Gromix still baked into the mdrun machine because of this, like so look at all the force field interactions. You can test the chemical potentials of particles you insert doing a rerun of Gromix, and you can use the whole of Gromix as a slave for a quantum mechanics application called Mimic, where Mimic actually drives the molecular dynamics and Gromix is the part that does the interaction energy calculation. But these are rather special topics, so right here the main mindset I'm in when presenting these options is really the benedict and the run that really takes the protein and wants to sample its confirmations about many other things around it, but here and there I'll say something about the other options. So let's have a look at the options in detail when we just go through and as promised I'll really just walk through the options as we see them in the mdp files. Let's see what this default file gives us, initial time at zero and a time step of 0.001 femtoseconds, and here on the right hand side I show you what I think is reasonable with practice and maybe some clarification. For the most part the start times and so on should be taken care by checkpoint files, so that saves you lots of work manually changing parameters and you see a bit of the times remnant in parameter settings. You often don't have to care about from a time before checkpoint files when you had to manually rebuild a TPR file. Now you can really just use these and rely on these wherever possible and I suggest you do that if you want to continue simulation runs. Just right now I noticed that I also didn't change the end steps option, of course the zero steps is not very exciting. For lots of cases I think even with compute centers it's a good idea to use a minus one end steps and to really rather have your simulations limited by time and time applications. That is often something compute centers like to see in usage and also this allows you to post process and not decide in advance. That depends on how you want to do things. For the time interval you have for vanilla and the two femtoseconds is a good time step and if you use virtual sites which you get activated when using PDB to GMX minus V site then you can go up to time step of four femtoseconds. The next option is the center of mass movement removal option and there I really suggest for the most part use the defaults use of the linear center of mass motion removal to avoid a flying ice cube effect for almost all simulation settings that is just reasonable and if this would require some other setting usually you already know about it in detail. The Lousovan dynamics is something I will really not go into details here but if you need some more information about that you're interested in that there's a whole section in the reference manual that explains some sarcastic dynamics you can do with chromics. For the for the energy minimization next step of parameters the defaults usually are quite tight and really good parameters to use. In most cases you really don't need to change anything you might not even go down to the lowest parameter set here but you might converge to something else and stop the energy minimization slightly before. This is really just safe to use as a default for the largest part. Then there are some other energy minimization options for very special models again the shell molecular dynamics reliability of some certain models included but there again if you're using this you know what you're using here I guess you would otherwise go to the manual where you have a much more explanation of this rather special case and then as another option you can use is that you can use a con you get graded and LBFGS algorithms to achieve better energy minimized structures sometimes needed for example for quantum mechanical calculations also there the defaults usually break quite neatly. Then comes the test particular insertion option then for use with MDV1 I'll not touch on this again here you can see things described a bit more in detail in the manual. We jump to the next section so now we have determined what type of MD simulation we want to do and how we want to propagate our coordinates with type of a integrator. Now it's all about how much information you want to get out of your simulation and again here we see that historically we used to write full precision TRI files that took up lots of space and could be used for restarting new simulation that's where you have the coordinate xout velocity vout and force out output frequencies. Here you can say lots of disk space by just leaving the defaults to zero and take checkpoint files to continue your simulation run in case you need to. Unless you're specifically interested in this information usually that is not needed what you would rather use is the XCC files. So we see further down in the section the output frequency and precision for the XCC file defined here again you can use the default parameters for precision that had proven to be quite useful and the output frequency itself depends very much on your simulation system so it's all given in steps here that means if there's an nstx out compressed of a thousand that means you have output every two picoseconds if you use the two femtosecond time step or every four femtosecond time step. Then up there you see some other parameters for the energy output to the log file and log file frequency so you can say how often you want to write to the log file that is a good way to check that your simulation runs neatly also writes some energies to the log file here then the energy file is a quite useful thing to look into your simulation what happens however is that you're required to calculate the energy so if you output the energy every step you drastically include the of workload every thousand steps is usually very good indicative range of looking into the energies where you will learn what you would need about the energies and there's another slightly mysterious parameter the nst clock energy of parameter and really this one you can leave at the default or set to minus one because this nowadays is reset by chromics internally to match the energy output frequencies and other options you set in your simulations then further down you see the compressed x groups where you want to use the protein or the part of the system you're interested in maybe using system here that is a of course something to think about a bit more before starting a simulation that runs for a long time and gives you normal data you would like to use there's another thing you can use for reruns especially these are energy groups where you have a non bonded interaction energies output so we can't have this output on the GPU but what you can do is create a trajectory on a GPU use the md run minus rerun option and then look into the energy groups behavior here but use the output of care because the energy groups here don't take care of a long range non bonded interactions via pme so that is rather it can be used as an indicative measure oftentimes good now we dive deeper and deeper into the chromics engine and also go deeper and deeper into the md algorithms in this case here for the neighbors searching that is how do we find the closest atoms nowadays there's just one scheme we can use today so you don't have to think about it in the newest chromics release in the older chromics releases you could use the group scheme which already then was deprecated so really is really the thing to do and with this you can safely stick to the default nst list the neighbor list update frequency of 10 for the periodic boundary conditions you will know what you want but almost always that is the periodic boundaries in x y and z directions and almost always unless you simulate infinite periodic molecules you would also have periodic molecules set to know then for the belay buffer tolerance and this is the most determining step parameter for all the neighbors searching options this should stay at the default unless you know exactly what you're doing and gives you how much error you allow the system to do while doing efficient neighbor searching this error tolerance is quite low allows for the best balance of all the algorithm to be set and that this belay buffer tolerance itself determines the R list parameters so also they just keep the default it will be reset by chromics when the belay buffer tolerance is defined so all in all for the whole neighbor searching parameters what you need for the vanilla simulation case is the default parameters all over with no changes then we come to the other part that plays into the force field together so the options for the electrostatic and fundamental parameters and here I slightly resorted the output from the mdp file just to be consistent we talk about electrostatics first and then talk about lineage owns parameters and here the largest k has to be taken with respect to what you do in the force field most force fields are parametrized and broken with pme so this is the option you almost always want and also the column modifier that put the potential shift you want the default parameter here the R column switched is then ignored with a particle mesh eval to method and also for the R column you would use the default parameter and here we have with the particle mesh eval option this column radius parameter really just determines how much work you do in Fourier space versus how much work you do in real space and we see below there's a Fourier spacing parameter set and generally we suggest a ratio of Fourier spacing over R column to be fixed to 0.125 so this two parameters can be tuned also at the beginning of the simulation and are tuned at the beginning of the simulation and here this ratio of the parameters should be kept constant but this will lead if the ratio is kept constant this will lead to the same results here then these Fourier nx and ynz parameters are automatically in determinant and also there I strongly suggest keeping the default parameters unless you know exactly why and what you want to change here this goes all together before tuning the best performance for the Fourier space calculation part of the long-range electrostatics for the PME order that is how much effort do you do in approximating the particle mesh in the B splines so that's a B spline order usually 4 is a good choice but if you go to higher numbers here you can reduce overall communications if you simulate very large systems go down to 5 might be something to get a bit more performance of your system so your options are 4 or 5 generally in here then for the tolerance of the eval summation here I really suggest keeping the default parameters then we have some more parameters for the eval geometry and epsilon surface where we again suggest using the defaults so they are meant to enable some differentiation if you have a geometry of our systems that have a large z direction expansion then you could gain something with changing this option but for the most systems really the default option is the one you would like to use here as well as the epsilon surface value to zero that would incorporate some type of dipole correction in your system which then again is not recommended to use when you have moving charges in your system which you will almost always have here again all the other options following further down were options to be used for reaction of field for electrostatics and options that can stay in the defaults when you use the particle mesh eval to parameters then you will see an implicit solvent option and this really is not only deprecated but it's removed here so this parameter setting to yes even just to give you a warning that you cannot use implicit solvent that's a feature that was present in Gromix some time ago but had been removed over the time so this one is just not to be used now we come to one of the parts I think if you lost a bit of your focus I would really urge to take up focus again because this is now one of the parts where you really can influence the physical accuracy of your simulation and how much you match the force field you want to use and how much you really apply the actual force field you have been choosing these are the funda waltz parameters you will see here so for the funda waltz type and overall for the funda waltz parameter options I would recommend using the parameters that came with the force field you're using and checking the force field publications by the way the force field publications show up automatically on the PV to GMX when you choose the force field you will see the citation of the direct force field in case you wonder where these parameters are published if you use other force field parameters again also our suggestion is to read up on where these parameters come from and are there a parameterized so in contrast to the electrostatics the funda waltz interaction type of for the most force fields should be cut off and kept to default we have implemented a PME treatment for force fields which is useful for example for special binary systems however there you would have to make sure that you find parameters that also match that so the for most applications that default is the best choice if you are really looking into membrane bilayer systems we suggest checking then again if this requires maybe some more careful treatment of the funda waltz type if your system is very inhomogenic otherwise keep the cut off and check with the force field for the funda waltz modifiers so at some point the funda waltz potential levels off and there you will have a choice between a potential shift parameter that simply shifts the whole function and a switching function that smoothly makes the funda waltz interaction go to zero after a certain cut off and there the amber and charm force fields use different philosophies and that's why you should match the force field you're using here usually amber potential shift and charm the force switch but again for different types of parameter sets in charm for example there's a slightly different race this had been handled so please again check the publication here for the cut off length themselves also here for amber you can use zero because we use a potential shift and not a force switch for the our funda waltz switch parameter charm usually uses one and for the funda waltz cut off amber usually uses one charm uses 1.2 extensive further yeah so these are the most general cases one thing we notice that people often miss is the dispersion correction that is applied to the funda waltz interactions and here the default option is uh no if you check and here we again urge you to check the force field most often energy and pressure is to be required here for dispersion correction and the further two options the eval-artal lennat-jones option and the lennat-jones pm combination rule really only apply when you use the particle mesh eval implementation of funda waltz interaction and there again the defaults are a good choice now we go further to setting the temperature in your system and if you want to run at a certain temperature for most cases now we find that velocity rescale gives you a very nice thermostat that is very robust and there you can use most of the default parameters set by increments now if you go through the coupling steps you can keep a default of minus one that means the thermostat will couple every neighbor searching step that is all steps when anyway global communication happens the thermostat will update the temperature which is usually verifying the frequency for doing that and then we don't get extra communication overhead from setting this to a different frequency that doesn't match the frequency very anyway to global communication in grommets between all different nodes and for other parameters the noses who were chain length and the printing of the variables they you can use the default variables and they don't have any impact on the velocity rescale thermostat for the temperature coupling groups here i see i was a bit too generous what i mean is protein and non-protein so usually that's the setup you would want to keep things equally warm because due to the heat conducting effects between protein and solvent that is usually recommended to have two separate temperature coupling groups between the two and for the coupling time how tightly you want to couple your system to a certain temperature we recommend that you use 0.1 even though the velocity rescaling algorithm here is quite robust in contrast to some other systems of course the reference temperature is the temperature at which your system should be at which you have to know usually 300 Kelvin now we come to pressure coupling if you are inclined to pressure couple for most of the biological systems you want to assimilate this is what you might want to do again it's almost always the parallel one pressure coupling you would like to use and also there you would like to keep analogously to the parameters set for the temperature coupling the default parameters for the coupling type and for the number of time steps during which you press a couple your system even though the coupling type you might want to change if you simulate membrane systems to a semi-isotropic the time constant is something that is possible to set around 40 picoseconds is one possible way to ensure that you don't have too tight but also not to lose the pressure coupling the common value used for the compressibility of your system is the value just quoted here 4.5 times 10 to the power of minus 3 parameter as well as a reference pressure of one by as a usual choice for atmospheric conditions there's one more parameter in here which is only relevant when pulling but doesn't show up when using pulling things directly in the pull section maybe something to pay attention to or something you should think about when applying any type of reference coordinate in your system that is absolute because suddenly with the pressure coupling what is happening is that you scale your coordinates system so you scale your simulation box that means also it's likely the position of coordinates does change now you have to think about whether you want or you don't want your reference coordinate to follow along with this movement and what type of reference coordinate so that all types of reference coordinates should just follow along with the scale box movement or just the center mass or not at all so one thing to think about when pulling otherwise not of relevance for the pressure coupling there's a whole bunch of quantum mechanics options which we can safely skip for now and they have also since the group scheme has been removed and deprecated there's no current in QMM implementation in Gromix that uses apart from Mimic that uses Gromix as the driving implementation you're working on some things along these lines one of the next releases however simulated annealing is exactly what it says you heat up your system according to a certain time points I think this is quite explanatory but let me know if there's some questions about that then there's one other thing that gets easily overlooked and which generates a slightly funnier in fact and that is a generating velocity for your startup run if you don't choose to do that generating your velocities at the start of the run what happens is that your particles will have no velocities obviously and that means that your system is at a very low temperature and within the very first simulation steps then will heat up and the thermostat will take care of this so it's a nice if you start with an immediately a warm up system and start your production from that then one next section for the once treatment is something that goes also again into the force field mostly what we want to use for constraints is neither none or all but really the H bonds constraints this is also what lots of force fields have been parameterized for but again check your force field in this case if you constrain all bonds be aware that you might miss some conformational changes due to the constraints because then system changes are much harder for the system to do because the bonds can't stretch for such a rotational movement for the rest of the parameters it's usually fine to use the default parameters so we use shake or water molecules links for all other constraints usually and do quite fine with this the default parameters and then we are through the main talk here and I encourage you and I hope you have some questions coming up I'll try to answer you now in the next steps I think um Ross and Bill moderate thank you Christian thanks a lot I hope the presentation was useful for most of you and looking at the list of questions it seems like it is some of you have had problems without you I hope it's being solved by now um so the first question is from uh second uh from Horacio uh actually he has a couple of questions so let's see if we can uh have audio with him Horacio can you hear us yes hello excellent we hear you yeah you can speak directly with Christian oh perfect so thank you thank you for the seminar Christian I was just wondering um for example uh what the md v v a avek would be because I I haven't seen this option before I think it's a new one and the second question was how is the measurement of the verlet buffer tolerant uh so the last question I didn't get uh how this verlet buffer tolerant yes um yep uh okay I start with the first question um that is the um for the integrated the options there were many options and one of the options was uh molecular dynamics molecular dynamics velocity and then as a third option molecular dynamics velocity vb aavek and this is um also really explained in the manual and the idea here is that we use a velocity value algorithm where we um do the kinetic energy calculation at a different step of the evaluation so the idea with both velocity value and velocity value aavek is that you get um at a different point uh the you have the the velocities calculated and you get the full velocities at the full step uh and slightly but ever so slightly uh more accurate results for some special dynamical property in your system but it's usually really not recommended and to use for your plain molecular dynamic simulation um similar setup so the it is recommended if you're interested in very specific thermodynamic properties of your system at a given time um but uh yeah so this is a brief um uh just a rather pointer than a complete explanation of this uh but yeah again uh you'll find a bit more in the uh manual on the mdp options okay okay thank you aavek and then the second one um about the um melee buffer tolerance so the idea is that um if you use a uh list of neighboring atoms um to speed up your simulations so this is the whole neighbor list search approach the idea is you keep uh everything that is close during your simulation and during a time step conceivably might um run into your interaction range uh for example um for the um fundamentals interactions and um now if you um think of a like a case where suddenly um atom becomes or more water molecule becomes very very fast um then this might happen that you miss out on some interaction energy just because um you weren't even thinking that uh by keeping this neighbor list of all the the atoms that might come into your interaction range uh within a certain time um you didn't even think that this could make it um and this is exactly this uh energy tolerance uh so you're missing out on very tiny bits of the interaction energy uh when using this uh neighbor list approach and this is um a um an analytical estimate on how much you would miss out in terms of energy uh interaction energy between atoms when using this neighbor list approach and from this you can calculate how much of radius you need and you see this very very tiny to make sure that you are really um almost okay yeah the interaction energy so this is how it's um how it's um calculated so it's a the tolerance in terms of energy then yes exactly so the tolerance is in terms of um because we think this is a way you should think about your system is um this is the base idea that uh whatever you do you want to make sure that you don't um make up for energy okay but the do you have um do you have in this case or would would you would you keep track in this case um also the number of neighbors that are or would be enough as a function of density for example or uh no this estimate is made on the assumption of a homogeneous system um so this uh but it's a conservative estimate made here for the energy term um i yeah uh i think in the uh one of the grommix papers there's a bit more content on how this neighbor list of parameters are exactly translated into um distances but the really the approach is really uh to um see what's the velocity distribution of water molecules and um how far do you expect molecule to move within a certain range of time and then saying okay how much of an interaction are we missing if suddenly some for example water molecule moves in or out of the interaction range so this is so there the idea okay okay i got it okay thank you okay uh then we have a question from john cabrera uh let's see if we hear each other uh hi hi hi so um uh i would like to know uh how uh select the default uh time step uh you said that it was two femtoseconds the valid buffer mm-hmm it's a very good question um so this is one of the fundamental question in molecular dynamics so what you want to do is to calculate the forces on all your particles and then move the particles forward according to the forces as much as you can to get the good something of your system now if you move your particles only a very tiny bit um according to the forces you have so this would mean a very very small time step then you spend lots of simulation time in very uninteresting system behavior so you would see your system very very slowly moving um only and you wouldn't even reach the time scales that are interesting for your system now if you um start as you calculated your forces and then you choose a large time step that would mean that um you move your particles forward a long bit then what happens is that suddenly you end up your particles in a position that is um far away from the position before completely different forces and energies and interactions uh up to the stage where you keep shooting particles around randomly so if you imagine you calculate the forces on your protein and you take an enormous time step um that uh then atoms on the very top of your protein might end up at the very bottom and everything will get enormously mingled so the question is how to decide on the time step that is exactly um the right um order of magnitude for this and for this you can look at an harmonic oscillator approximation um so uh there's a really distinct formula that says the fastest vibration uh that means the fastest movement in your system should be time limiting and for this we use the force constant to really estimate the time step from this interaction so um the usually that's why we use also the hydrogen bonding constraints are usually the strongest um interactions that we still want to simulate to determine the smallest time step um in the vibrations so that we can get a stable trajectory with the largest possible time step um yeah this is um long story short um for all the force fields we have uh two femtoseconds without virtual sites or four femtoseconds but virtual sites is uh just the recommended option that has been um found out to be exactly in the right ballpark to give you a stable simulation protocol that gives you still efficient sampling and uh lots of a sort of thing enough to protein trajectory okay thanks christian uh we go with the next questions from uh oscar who has a question about PME order oscar you can speak um hi hi oscar yeah so thank you very much i just wanted to ask about the PME order because i think that you i understood that you said something about uh like it was better for larger systems in the in the terms of communication issues but i just wanted to clarify if this also changed somehow the values of the electrostatics no it shouldn't uh so um we see that we receive the same accuracy um if you go to a higher order so between order four and five so you should get the same results um you really just uh all right so it's a matter of a data compression in the end only um so how you um between hearing whether you have a fewer grid points and compress them better or you have more grid points and compute press them less and this is a way of shifting work between um so say individual separated um evaluation units okay okay uh the next question from navine i believe doesn't have audio uh so does the protocol that you were describing works for both protein ligand complexes or only for proteins no sir this should work for protein ligand complexes as well so um there all the options um apply the only thing to uh that is really important is the temperature coupling should be um protein and non-protein and not protein and solvent as a was put in here uh to make sure that you get the right um the right um that everything in your system is a couple to temperature both uh correctly but that uh yeah okay and the next question from partiba can you say something partiba hello hi hi hello christian hi hello christian thank you for your wonderful presentation uh i have a question about uh the parameter that i should use uh if i am planning a system based on ph so could you explain something about it ah yeah that's an interesting question so what you would like to think is that uh because ph is some kind of external thermodynamic property that you could just set ph equals some number which at the moment you can't do in gromics so if you want to have a certain ph set in your system what you would do instead is um take your protein and protonate or deprotonate accordingly uh to your ph value and for this you would use this um uh thing um estimating the pka of each and every amino acid um at a given ph so um for this there's a different tools around um i was small blackout too but something okay sorry i have to refer you to google in this case okay but the the concept is um you um manually protonate all the amino acids according to the ph you want to assimilate for the time being uh it has been a long-standing um idea um of the whole the community and the and on charm there's a way of doing um simulation where you just set the ph at the very beginning at the outset and something we are working on but there's something uh yeah um not for now okay thank you very much all right and uh the next question is from Zavett you can speak hi hi hello sir hi thank you so much uh yes uh thank you so much for explaining uh the basic steps i want to know something from this uh uh in the prior versions of 2015 if the law firm used to get close evaluation time of g2 slash cp ratio and uh that is not there in the 2018 uh series right now i'm using it was an 18.7 so uh i want to know that that optimal performance of this gp and cp is it have any relation with the computer is spacing value that we just explained in the your lecture that is point one two and in the file which i have downloaded from the website it's set to point one six so does it have a relation to the optimal performance of the uh simulation so um okay um i'm not exactly sure what ratio you're saying so there is the for the Coulomb interactions there is the R Coulomb versus free spacing ratio that i suggested to keep to constant value that is then tuned was this the ratio you're referring to um um uh so it's in the last of the log in the log point towards the end we used to get off close evaluation time it seems that optimal performance of this ratio should be close to one so this reports something value greater than one or less than one it means i have some performance loss during the simulation running in a gpu workstation with uh of course two gpu parts okay um so if i understand this correctly this um this is a matter of how much work of the simulation you um you put on the gpu versus how much work you do on the cpu and for this there's an indirect influence of the mdp parameter options uh in the how you set the electrostatics and there also the pme order can influence this type of work but then most importantly it's a matter of a command line options you use for starting the simulation run and um how you um how many gpu um um how many gpu you have versus how many cpu you have and this also drastically changes with different chromics versions of this case so um it's a bit hard for me to just write from a distance um immediately understand uh how to uh to give you directly the advice on what to do to achieve the optimal performance here um on the setting so usually um yeah um i would defer this a little bit and suggest uh you send me an email personally just um stating the issue again because this makes it a bit easier um to answer your question yes sure sir yeah okay thank you thank you so much yeah thank you okay uh next we have a question from amit let's see amit do we hear each other you can speak okay i'll read the question on your behalf if i have a system that contains protein water and a single or multiple copies of a ligand should i perform a separate coupling for the lica particular temperature coupling so is this recommended to couple the as one group of protein and separately water and legal together or you have protein water and ligand is three separate distinctive temperature coupling groups i think in this case uh the temperature coupling um would work also with a protein and just non-protein um as uh was given as a temperature coupling not to three systems but to two systems um should be fine in this case uh the matter is the question is the matter of um heat transfer between the solvent and the ligand and um in this case um if the ligand multiple or single copies of a ligand of 40 atoms um the ligand is quite small and the heat transfer should be uh fine so in this case uh the coupling so say the water itself and the ligand environment acts as a um fair temperature both so the differences shouldn't be large and uh choosing protein and non-protein for the temperature coupling should also just give a very good result that um yes the correct temperature in this case okay thank you uh next we have a couple of questions from bakari let's see if we have algil bakari can you say something we hear some noises hello hi yes we hear you if you speak out okay uh uh i had a question about um running um mdb hydrogen mass repartitioning and so it's about the time step when you use hydrogen mass repartitioning do you use v side yeah when you use pdb 2 gmx or you use minus abh yeah uh yeah that's a fair point i didn't touch on this um yeah as i briefly went to um at the beginning um the time step the question is uh the fastest oscillations in your system and of course if you use heavy hydrogen atoms then you also um dampen the dynamics um that means um the hydrogens become heavier they move all less quickly and uh thus um you can have you have um oscillations and you can uh increase your time step also in this case um however um i think um if you want to get better dynamics of your system um then v sides are better um because they will give you so if you have the choice between the two um i would suggest you use virtual sides um because then the simulations will also reflect much better the actual dynamics of the system uh various the heavy hydrogen atoms um option is is also a nice option to give you more something um and it will give you the right free energy distribution of the system but will give you a different dynamical behavior of the system so um yeah yeah okay uh thank you and then we have a question from der gehli kohut but uh he doesn't have audio um so his question is what would be an optimal temperature coupling tau t for no say hoover thermostat um is it better to use no say hoover for production runs instead of v rescale um so usually i would say um velocity rescale algorithm proves to be just more stable and produces the right and correct thermodynamic ensemble um and um i would i would personally prefer v rescale over no say hoover thermostats for this case um the optimal um coupling um temperature coupling constant for no say hoover um i just don't know um right on the top of matter i don't want to say anything wrong actually but the days um in this case i can really refer to the um manual section um in the uh grommix manual if you go into the md algorithms today's section that talks about thermostats um and looking at the grommix menu under thermostats there's a directly um very nice detailed description of the no say hoover thermostat right the tau t the coupling optimal temperature coupling constant is also discussed exactly for the no say hoover thermostat um which will give you a much better overview than i can just give you in a few sentences here um so um yeah there you see also even implying value printed for the tau t okay um in interest of time i'm going to read the questions of the rest of the participants because we are already of the hour uh question from silayman what are the most critical options which should be changed if i use another force field um uh yeah that's a trend or asking for summarizing all the slides where there's some like blinking red thing on it um so um the things to look out for is um the fundewalds interaction type and the um lena jones um radio the cut off a radii for the fundewalds interactions to check the dispersion term is all important here and um then the um constraints using um these are the things to check when um using different force fields i think these would be the most critical things to have a look at okay next we have a question from mayank uh who is asking about more details on the ref court dash scaling option during pulling simulations seems to have uh problems with the system blowing up in the direction of uh pulling through a membrane um is there any specific to this option that uh users should know about uh yes the question is uh and this bit hard for me to just understand from briefly reading this here is um whether this um the position restraints are absolute in space or relative to all some kind of atom positions so if you have a system usually what i would suggest um to do is to try to define your position restraints in a way relative to your molecule um coordinates um you want to pull towards want to pull away from where uh you would not have to worry about the reference coordinate scaling uh in this case um not sure this is directly applicable otherwise um mostly scaling along the box it might be what you might want um if the system begins to blow up in the exact direction they read it here yeah okay uh then um a question from sekie about normal mode analysis and uh it's just being curious what are the applications of normal mode analysis um particularly in draw marks so yeah i think um normal mode analysis spelt or played a role in the uh estimating spectra molecular spectra um it there is some analysis tools to um look at the data you get from normal mode analysis however this for protein doing normal mode analysis for proteins is a quite expensive procedure that looks at the the um changes in energy or interaction energy um just at the very specific point of confirmations so i would say uh for most of the things um we do with chromics um the type of analysis that tells you okay where would the protein move what are its degrees of freedom of movement uh often pcaa proves to be um a method of choice rather than normal mode analysis thanks then a question from moritz after the uh so the question is that after the grew cut of skin was removed is it possible to still use flexible williams force field unfortunately i don't know because i don't know enough about flexible williams force field um so for the most part yeah might be possible might not i really can't say uh sorry for this yeah uh then we have a question from yogesh what settings should be used for doing nv ensemble simulation so yeah if you want to keep the volume and just the energy constant and not the temperature then um all you have to do is really just not um make sure that your system gets up to a certain temperature then the um energy should be conserved and make sure the system doesn't change the box size that means that uh the pressure is not kept constant so in this case setting all these things all the recoupling algorithms to know will result in an nv e simulation um now if you set up such a simulation and don't generate velocities this gen velocities in the beginning then you will have a extremely cold system because there's no extra energy coming into the system and your molecules if they're not already in a local minimum from the energy minimization they will move towards the whole system will move towards the energy minimum and fluctuate around that which will result in very low velocities for the particles which will be reflective of a very low temperature if you then calculate the temperature if you want to do an nv e simulation and uh then that is somewhat reflective of some kind of higher temperature you can generate velocities but the question is what what it really tells you in terms of physics uh so yeah um so some things to something to do is to generate velocities if you really want to do an nv e simulation that um has particles moving quick question from the banjo how to configure ndp files so that it generates xdc files as outputs yes um so here um the ideas that you use a whole pipeline of which i i really just focus today on one aspect that is you need a structure force field and simulation parameters once you have this then you're going through the pre-processing and starting a simulation will generate the xdc outputs um but for this i really suggest looking into the tutorials um again for example mdtutorials.org will give you one of these pipelines thanks and uh then we have a question from the hit is it possible to predict thermal stability of proteins at various temperatures using md uh yes uh to some extent um what really works neatly and uh there's one example of a barnace study by um people also by itself um veto the skapsis and the bear to hold for example it um as a reference um we we can look at the changes in thermal stability upon mutation of a residue um the thermal stability of a whole protein in itself um is um you can also access but this is a harder task but most often you because then you really have to look at the changes in the complete protein structure um but most often biochemically you're interested in um looking at what happens at the mutation and this we can definitely do very well the other thing you can also do conceptually the question is one of compute time and whether the results are reliable or not thanks and the last question we have is what should one consider um when setting the constants for pressure coupling the time constant yes um what properties of the system are important to be considered um this is a matter of a the change of system behavior usually so if you use a very tight pressure coupling then with a very low time constant so very frequent pressure coupling then often if your system is not already very stable um in the simulation behavior then you might see some changes if you use the parallel um pressure coupling um of a fluctuating box sizes so the main consideration and this is a bit more of a rule of thumb is that you want the coupling tight but not so tight that you see large fluctuations in your system that might lead your simulation even to a crush at some point um the system property is really um of course one of her um yeah uh fluid liberation and how well your system box like the volume of the system really represents the um the volume of the system at this very pressure you want to simulate at so yeah something to try um and uh unfortunately no extremely easy answer of this is exactly the value you should set the coupling constant to but usually again i can refer to looking at the parallel element of pressure coupling um written up in the manual using the default value uh or the values no sorry the values i gave in the presentation should give you usually a stable result for bite range of systems i love it thanks uh i'm sure there will be a lot more questions coming especially as chromium valves further but uh we are already away over time um so if you would like to follow up we have more questions please visit as though by excel.tu we have forums there or the grammar spelling list and pressure questions there will be happy to follow up and i'd like to thank again christian for the great talk i believe he's been very useful for the whole community and thank you all for attending today and follow up with all the events and training that we organize on bio excel we regularly do various webinars workshops we are starting with virtual training so you might find a lot of useful information there thank you all again for attending and until the next time let's finish the webinar thanks christian yeah thank you rossom for the next host thing