 Hello everyone and welcome to our webinar. I'm Rostin Apostolof and I will be today's host. This is the fourth webinar from the BioXcel Educational series and today we will be having a look at some automation methods for free energy calculations using a tool called BMX. And it will be my great pleasure to have today Berthegroth and Vas Jager to present this material. Before we start I have to make a few announcements. First you should know that this webinar is being recorded. We will put the recording on the BioXcel YouTube channel and also on the website so that you can watch it later. At the end of the webinar we will have a questions and answers session where you can ask questions to Berthegroth and Vas. And you can ask your questions using the goto webinar control panel. There is a questions tab there and I will give you the microphone to talk directly. If you can't, if we have problems we call you, I will read the question on your behalf. Before we start I would like to give a very short overview of BioXcel which is the organization that is organizing this webinar series. BioXcel is a center of excellence for computational biomolecular research. It is established in funding from the European Commission and the center works in three main directions. One is towards improvement of performance efficiency and scalability of three key software packages that are widely used by the computational biomolecular research community. One of them is Gronk probably many of you are familiar with it. It's also the backend behind PMX that we presented today. The other application is Hadoq for integrative modeling of protein complexes and also small ligands and we also work with CPMD specifically on QMM approaches to modeling enzyme reactions. BioXcel is also working on devising efficient workflows including data integration. We work with several popular platforms and workflows such as Galaxy, Ignite, OpenFacts, Pashita, Verna, Comps you might be familiar with some of them and the center is developing an extensive training program. It's providing consultancy services to promote the best practices and make all users of those applications do their work better, faster and more efficiently. What might be of interest to you is that we are launching several interest groups on certain topics related to computational biology and some of these interest groups might be interesting for you to join. We are just starting with them that you can visit our website bioxcel.eu slash interest dash groups and you can subscribe to those mailing lists. BioXcel provides various forms of support. We have forums on ask.bioxcel.eu, we have several other ways of contacting us. The chat channel is also open and we have a video channel where we have copies of the webinars and we will be putting additional material in future as well. Now without waiting I would like to introduce today's presenters. We have Professor Bern de Groot from Max Planck Institute who's been working for many years in the field of computational well molecular dynamics simulations. He's working extensively on chromax and free energy calculations. We have also today Vance Jager who is a postdoc in Bern de Groot's group and they will tell us more about how we can automate the calculation of free energy and we're looking forward to their presentation. So welcome, Bert and Vance. I would ask you to change all the presenters to them. All right. Thank you very much, Rosen. My name is Bert Groot and together with Vance, as Rosen already said, we're going to spend the next 14 minutes or so talking about how one can do free energy calculations in Gromax utilizing the BMX framework and I should really say that this is to work primarily of phytotoscapsis who together with Vance works on the development of BMX and this all started with the work of Daniel Seliker. All right, so before we jump right into it, maybe a few words on why we would be interested in free energy calculations and particularly also of free energy calculations on protein mutations. Why proteins? Well, proteins are nature's nanomachines and they carry out most of the functions in our human bodies and also throughout nature and as such it's not surprising that actually many diseases are linked to mutations that occur in proteins so that's where a major interest into protein mutations comes from and the other key motivation to be interested in mutations is also because they are really at the heart of protein engineering and design so to be able to use proteins for for example industrial applications and so why would we want to do free energy calculations? Well, as you probably know, affinities are free energies, free energies of binding and that means that the affinity for a ligand binding to a protein or a drug molecule binding to a protein the affinity is given by the free energy of binding and the same holds for two proteins binding to each other so in the field of protein protein docking also the binding free energy determines their affinity. Another form of free energy is stability so in that case we talk about the free energy of folding and their folding free energy gives the stability of a protein for example against a denaturation or a thermal unfolding so again there also if we're interested in how a mutation affects the stability of a protein we might be interested in computing free energies and if we want to do that on a on a large scale then of course it would help very much to have an automated and user-friendly framework to carry out those calculations accurately and this is exactly where PMX comes in. Now a little bit to give you a sort of a background where where PMX fits in of course there are many different methods to calculate free energies ranging here on the right hand side from fully statistical approaches like docking for example to more hybrid approaches like PBSA calculations or also Rosetta falls into that category and then on the left hand side we have methods that are based on the first principles of statistical mechanics and molecular dynamic simulations would be among those methods and and among those we can again distinguish between two different flavors one would be free something so if we are lucky enough that the process of interest takes place on a time scale that we can easily simulate say for example we have the reversible folding of a protein or we have the reversible binding of a ligand then we can just count how frequently are we in a folded or in a bound state and compute the free energies from that but in most cases we're not that lucky and we need some sort of biased sampling and then we can go to enhanced sampling techniques like umbrella sampling that is based on an equilibrium approach or we can also use non-equilibrium methods like Kersinsky and the other branch of bias sampling is comes in by a so-called alchemical methods and that's exactly the topic for today now why would we want to work with an alchemical method but the example here shows how that can be useful if we are interested in the binding say of a ligand to a protein here on the top and experimentally we would determine the affinity by just computing the delta g of binding of this ligand to the protein now in a simulation this might work but this might also be problematic yeah if the affinity for example is high then it might bind but not spontaneously unbind or vice versa so we might have the difficulty sampling the pathway in a way that we can extract the free energy in an accurate way but if we are interested primarily not just in the affinity of this ligand but rather in the question well what if we change that ligand how would that change the affinity then we can apply a very neat trick and that comes from this thermodynamic cycle and so there the the question is about the difference in affinity between these two ligands on the left and instead of then binding to the protein in the simulation and computing the free energy change what we instead do is we change the ligand from its initial state to some modified state here we add a methyl group to the aromatic ring and we can compute free energy for such a process and we call that an alchemical transformation because really what we're doing here is we are generating atoms out of nothing and we can of course do that in the unbound state here on the left and we can do that also in the bound state on the right and we can compute the free energy for that and now the nice thing of thermodynamics is that it's the free energy is a state variable and that means it doesn't matter how we go from say the upper left to the bottom right if we go via here or via here and we always get the same free energy value independent of the path it also means that the difference between the two black delta g's and so the two horizontal transformations must be exactly the same as the difference between the two green transformations and that's exactly the trick that we apply so what experimentally one would obtain as the difference in affinity from the delta g3 and delta g4 is what we computationally get as a difference in free energy in morphing in the unbound state versus morphing in the bound state um as oops a similar thermodynamic cycle is what we can apply if we look at protein folding so again the folding process by itself now as a as a vertical arrow might be problematic because it's um it might take milliseconds or longer for a protein to fold so we might not be able to access the um the folding or unfolding pathway in any reasonable way and compute in associated free energy but again what we can do is if we are interested in for example how a mutant effects the folding free energy or instability we can make an alchemical transformation of one amino acid into another here from bloom to red and we can do that once in a model of the unfolded state and once in a model of the folded state and again the difference in the alchemical free energies is the same um by definition as a difference between the folding free energies of the native protein versus the um mutated protein now how how do we exactly compute the free energy across such an alchemical pathway we do that by introducing a parameter usually called lambda that we um that is a coordinate that we switch from zero to one where um it is zero in the starting state and it is one in the target state usually these two states are referred to as A and B and um we then get access to the free energy of this process of switching from A to B by integrating um lambda from zero to one over the um a force that is acting on lambda or the derivative of our um potential energy function with respect to lambda and we can do this either at intermediate steps or lambda that we each um simulate in a discrete manner that's then called discrete thermodynamic integration we can also let it grow um successively that is then called slow growth thermodynamic integration um the emphasis here is on slow because the assumption underlying all this is that during this whole transition process the system must be in equilibrium so that means we should be switching slow enough that the system is is always in in equilibrium um this gives us access to free energies of these alchemical changes and um maybe some of you are wondering well how can it be that this is now a free energy where we just said that we are just integrating um across the gradient of the potential energy function um and nevertheless I'm claiming that here we have also the anthropic contribution to the free energy I think this is a nice piece of homework and um I think many of you should be able to figure that out if not please feel free to contact us at any time okay I was um emphasizing this requirement of being in equilibrium actually um there are workarounds if we um do this switching between a and b so fast that we cannot assume to be in equilibrium anymore and this is based or one way of doing that is based on the crooks fluctuation theorem so now if we have a if we are switching between states a and b both of which for which we have an equilibrium ensemble with thermodynamic integration we would need to do slow transitions between the two and get access to our um free energy if however we switch fast then we get um the free energy difference plus an additional contribution due to dissipated heat so we get something different from the free energy and um in fact if we do a um a number of these fast transitions we get the distribution of such work values non-equilibrium work values and the same holds for the um backwards transitions and so we get the distribution of forward and backward um work values and the crooks fluctuation theorem actually tells us that where the two um are identical so where the two probabilities are the same this is where the work values give us the um the free energy value so that's exactly at the intersection point of the two distributions um so applied back to our thermodynamic cycle of such um alchemical transformation this means um that we um um do that more thing once for example here in the non-bound state and once in the bound state and the difference um gives us the delta G that we're interested in um now um that was a bit of a yeah free energy calculations alchemical free energy calculations in a nutshell how does PMX help us with that um well um what you may have noticed is that um if you are familiar with MD simulations if we talk about the state A and the state B between which we um want to make a transition of course both will have a different topology so a different um molecular descriptor with different bonds and angles and atom types and what have you and this is something that um molecular dynamics engine has to know about of course if we want to make that transition um so in the gromax topology for example you can define an A state and the B state and this is actually um quite tedious to do if you want to do for example a um amino acid mutation yeah here we have um the generation of an aromatic ring that includes quite a few additional um atoms and and interaction parameters and now what PMX provides as a functionality is that for any mutation you might be interested in it gives you the framework to generate um such um parameters automatically so we have a mapping of each amino acid into each other amino acid except for proteins um and so we don't have to um generate these these mutation topologies by hand anymore um so within back of our thermodynamic cycle did now this now allows us to do a whole set of um mutation free energy calculations well as as one of the first sanity checks to see if the software works as it should um we did the number of um um at trivial cycles where we mutated um evading for example here to a phenylalanine and back in the same box as a two peptides one that was um simulated in one direction one into the other and um if we then um um make the calculation of course um a zero should be the results yeah because the beginning and end states are the same and um and and and quite a few errors in in creating the topologies and so um would occur in these free energy values to dba from zero so this is the first um check to see if things work as they should and we see here actually for quite a few mutations um even the changes where the charge is not conserved in quite a few different force fields we all get the scatter around zero so this is one sanity check um that appears to be fulfilled now of course the real um more interesting check is to compare against experiments and one thing that we were interested in is to see how well would modern day um molecular dynamics force fields um be able to predict changes in um thermosability of proteins uh due to mutations and for that we looked at um primarily at the protein barnace for which there is extensive um data in the literature and we looked at in total 119 mutations at 55 different positions and um we have data from calorimetry available so this allows us for a direct um assessment of the accuracy that we can reach with these kind of calculations um we um chose five different force fields that are popular so two amber flavors two charm flavors and opls and um we used the crooks set up that I introduced a second ago with 20 nanoseconds of equilibrium for each mutation and then 100 fast transitions in both directions which were really fast only 50 picoseconds each and we used PMX for all the topology generation as an overall accuracy um we get quite a reasonable correlation with experiments so here we have the experimental delta delta g for um these mutations versus the ones we calculated using the charm 36 force fields so we get an overall reasonable correlation with an um a mean unsigned error of about um slightly less than one k cal per mole and actually we find this for most of the force fields that we looked at so this charm varying does a little bit better than all of the others opioles seems to be um sort of on the maybe slightly uh doing slightly worse but overall one can say that on average we're sort of in the bulk part of an accuracy of about one k cal per mole um of course with that statistics we can also see where this error that we have where it comes from and I don't have time to go into that in detail now but we notice actually that we have a similar error in the calculations um one is a systematic error that comes from the force field one is a statistical error that we have from the something with the settings that we chose and the third source of error is experimental error and that's actually something you can see in this scatter plot over here this is experiment versus experiment and we also there see um for the same mutation in the same protein quite a good correlation but also quite some scatter so this also um leads to an uncertainty if we compare simulations to experiment what is also means means since we have a systematic force field error um we thought well maybe it helps then if we combine the results from multiple force fields and that's what we see over here so this is the individual force field results that we had before if we now take um a consensus approach taking results from multiple force fields we see that indeed we can improve quite a bit on the results of each individual force fields depends a little bit what consensus approach you take but even just um averaging two individual force fields already help so if you have the possibility it's always good to to compute things in multiple force fields and then take a consensus approach um of course we were interested to see if this not only works for barnage but also for other proteins so here we have an example on the Staphylococcal nuptias and a GPCR so membrane a receptor and we see that in both cases so for the nuptias we get a mean error again that captures around one kcal per mole for these different force fields which is similar to what we found for barnage and for the neurotensin receptor we don't have experimental free energy values but we have thermostabilites in terms of melting temperatures and also there we get a positive correlation between our predicted free energy changes due to mutation and the experimentally observed melting temperatures. This is where I would like to switch to Vance Jäger who's a postdoc in the group and he does the PMX calculations on a regular basis so he's in a much better position to talk you through how PMX actually works in practice. So we'd like to present to you a practical walkthrough of how one might conduct these simulations. I'm going to show you the four major steps of setting up and running an alchemical mutation and how to analyze the results using scripts available within the PMX package. I'm going to give you a few bits of advice on limitations of PMX so that'll hopefully save you some frustration if you're a new user. First let me describe the protein that we're going to be mutating today and it's this tripe cage protein. Many of you are probably familiar with it. It's 20 amino acids long and it's the smallest polycoptide form a stable folded structure so for that reason computational studies often use it. What we're going to test today is whether or not we can substitute this central tryptophan residue for a phenylalanine residue and see if we can predict whether or not it will still form a stable fold. So the first step as Bear was mentioning of a now chemical mutation simulation is to generate this hybrid structure in topology and PMX is perfectly suited for this task. On the left you see the atomar structure that you can get from the protein data bank and on the right is this hybrid structure that we would like to construct and the hybrid structure is going to contain the side chain of both a tryptophan and a phenylalanine so if we set lambda equals to zero we have only a tryptophan we set lambda equals to one we have only the phenylalanine and then values in between are some portion of each. So I'm going to first show you how to generate these hybrid structures in topology using an automated server that we've recently set up in our group and then I'll later move on to show you a manual method if you want to script out some of these simulations for a wider range of mutations. So when you first arrive at this PMX web server that we've created you'll be greeted with this page and within a few simple steps you can transform your protein. So first what we do is we upload a PDB this PDB here is the the first model structure of the of the NMR structure for the strip cage protein we select the amber 99 SB force field we can use we can mutate the protein multiple times at once but in this case we're only going to mutate it once and we're going to change amino acid number six that central tryptophan to a phenylalanine and we suggest usually we use this PDB to GMX to assign hydrogens because if there are mismatches between your PDB structure and a naming convention within these force fields you will get some errors so we tend to use PDB to GMX to assign those hydrogens. So from that we can submit our submit our job and the next the next next screen you'll see is a confirmation that your job has been submitted to the server and it should only take about 30 seconds or so to complete that depends on the size of your system and how busy the server is and after that you can click on on the link on the link given here. So here is where we can download our results and the files these files the server will be cleaned occasionally so we suggest you download the files here on the machine and not rely on the link. So the two files that you're going to get out are a hybrid topology and a hybrid structure and what you'll notice here is at the bottom of the warning that says do not forget to set the GMX load variable to the mutation force field 45 directory and that'll be something something important later on otherwise Gromax Gromax will get confused if it doesn't have this GMX load variable set to this particular force field. So here are the two files that you get out from the web server they're on the left you see your hybrid structure on the right you see your hybrid topology and we know that the mutation is taking place and that it's correct because we have a column here that has a W2F and that means that tryptophan has been mutated to phenylalanine but if you are familiar with yeah so that those are the two files that you will get out of the automated server and now that I've sort of shown you how to automatically do this the easy way to do it I will show you a little bit more difficult way to do it the manual way to do it and this can be useful for people who want to run a large campaign of simulations for many different simulations for many different mutations so here's the manual or sort of method first what we'll do is pre-process the structure so we start with this one one l2 y pdb and we run it through pdb to gmx in this case we're still using an amber 99 sp force field and tip 3d water and this should be no problem for anybody who's familiar with Gromax and when you download pmx you'll find a folder of scripts that allow you to manipulate the topology and the structure files and eventually I'll show you later on how to analyze the results you get from the simulations so in this step we're going to use one of those one of those scripts called mutate.py to change our pre-process structure into a hybrid structure upon executing the python command shown here you'll be given the option to select your mutations here we select residue number six obviously and retain it to funnel islanding and we would not like to apply another mutation even though you know if you would like to you can apply multiple mutations you can also use this flag minus script if you would like to script out multiple mutations and you can you can look in the help menu of mutate.py if you if you'd like to do that there's more information in there so here is our resulting hybrid structure that we generated manually it's the same structure as we got out from the from the web server as you can see the w2f residue name here so what we'd like to do next is is create a topology file and this yeah so we're going to use the hybrid structure with p2p to gmx to create a topology file that contains the hybrid residue and let me remind you once again about this gmx lib variable if it's not set correctly you will get problems so after this if you're familiar with pb to gmx you will get a topology file out but that topology file that you get out will need to be edited so if you're familiar with alchemical mutations and you're familiar with chromax files this topology file is missing the the b column so it does not have the b state that we need and in order to create that b state we're going to use a package called generate or a script called generate hybrid topology and with the script we input that topology file and we we will receive back topology that includes that b state so i will show that here so here we highlight new topology file and you'll notice in and read the additional information added this b state so these steps that i presented here in the last few minutes this is how you manually set up the structure and topology and if you're not interested in doing many multiple different mutations likely the web server is enough for you to do what you want so after we have this topology and uh after we have this topology and structure we're going to minimize the pull rate and mutate the the uh the structure so for those of you that are familiar with bromax next few slides should be easy to follow we have simple mdp files available linked at our web server page uh so to go to go over the setup and minimization quickly first we added the box size and shape to make room for the solvent this is going to be a cubic box with 1.2 nanometers of buffer between the protein in the edge of the box and next we use gen box to fill the box with water we generate a tpr file with gromp that we then neutralize and ionize and then we can run our minimization and that minimization should be very fast for a small protein like this it should shouldn't take much time at all the next thing that we need to do is equilibrate the box in order to generate an ensemble of structures from which we'll we'll do our alchemical mutations where we drive lambda from zero and one and also drive lambda from one to zero to achieve the mutations so the important portions of the mdp file a few of them are labeled here in red you can set your initial lambda to either zero or one depending on whether you want to equilibrate for your a state or your b state and then you can set your delta lambda here that's how fast lambda changes to zero such that it stays in state zero or state one for a protein of this size we um yeah we're going to run the equilibration two separate folders and try to get about 10 nanoseconds of equilibration and we're going to use frames from the last athlete trajectory pull them out using trash can bear and once we have those frames and we're sure that they're that they're close to some equilibrium we're ready to mutate here are some of the important parameters that we might change for our simulation we need to calculate the energy every step that's dh or d lambda as dh d lambda every step as well as calculating the energy every step um and then we can also change how quickly we we mutate between state zero and state one so that's the number of steps in our simulation we can change the initial lambda either from either zero or one whether we're starting from the a state or the b state and the delta lambda is how much lambda changes with every time step so the more if you multiply delta lambda times and steps that should equal one or negative one some of the parameters that you might change within your simulation campaign or the number of transitions that you have and even the analysis methods at the end and and i'm going to show you in some future slides how changing those parameters might change the results that you get out so here's what you will get out after you do an alchemical mutation simulation this one we ran for 50 picoseconds and and what you'll get is this extra file called dhdl.xvg and as uh as barricade shown before we can get a work a work value out from integrating this integrating this dh or d lambda uh over the whole range of lambda and from that we can get a work distribution and we can begin to analyze we can begin to analyze what the delta g is for this mutation so here is here is another script that we have within the pmx package called analyze crook stop line and this calculates the delta g of the mutation using several methods and here's the output file it's called cgi.png this is one of the output files from this analyze crook stop line we can see the work distributions for the forward and backward transitions in red and blue and the intersection of the two Gaussian curves fits those distributions gives us an estimate for delta g using a method called the crooks Gaussian intersection method some other files that the script will put out are this results dot that so this gives you information about uh different statistics about how well the how well the fits the fits are as well as it analyzes using two other methods which I didn't actually print out the results the red dots there there there will be results in yours using the benign acceptance ratio and the jarzinski estimator so now that I've shown you these files I'm going to show you just one sort of check as to how you can determine whether or not your simulations are well converged this is one more file that comes out called w over t dot png and what you can see on the left is the work value the left part of this graph is the work value over time and if you see that the work value fluctuates around some central value rather than uh rather than than skewing one way or the other um or or having a large jump during your simulation you can assume that at this state um yeah you can you can assume that the the state that you're simulating here is actually decently well cooperated um so here what I'm showing you is are some of the differences uh what happens when you use different transition times during your simulation so this these are these are the same simulations except I changed the transition time from 20 ecosystems to 50 to 100 to 200 and as you see as you um as you increase the transition time uh your your gaussians uh the intersection of the gaussians stays near the same but the width of the gaussians uh decreases and you get the values move closer to the actual delta g this this works decently well for the short transitions because we're going from a hydrophobic residue to a hydrophobic residue that are nearly the same size but for systems where we have very dissimilar residues oftentimes you do need long longer um transition times 100 or 200 pico seconds from long uh next we we select a different number of frames to analyze rather than using 100 frames uh as as displayed in the bottom here uh say we use 20 or 50 frames 20 or 50 transitions um you can see that using fewer transitions over uh fewer transitions increases the uncertainty but the estimated free energy remains similar overall so in this way you can you can tune the number of transitions in order to uh in order to reduce the error so now that we have this estimate we're going to use this estimate of negative 6.27 kilojoules per mole now that we have that estimate we can complete the thermodynamic cycle here is the thermodynamic cycle that Barrett showed earlier and as as he explained we can use some of these unfolded uh these unfolded uh mimics these gxg peptides where g is glycine and x is the amino acid of interest so we use that to simulate the unfolded state and we do that exactly the same as we did with this folded protein so when we add up all of our values uh from our simulation we would estimate that the phenylalanine cage the delta delta g for that mutation on the on the folding free energy is 14.3 kilojoules per mole and we actually do have an experimental value for that of 12.5 kilojoules per mole so we were happy to see that uh that there was good agreement with experiment in this case uh next i'm going to go over just a few considerations that you you might want to take into account if you want to use pmx first pmx uh does not use or does not do proline mutations uh because of the changes in the backbone topology second if you need charge mutations running a typical simulation would cause problems because there's a change in the net charge of the system over time and that will create some artifacts so we suggest a similar setup as Barrett uh had presented earlier where you have both the the folded and unfolded protein in the same box and you can read more about that in the paper that i have that i have cited below here and finally these terminally modified residues are not supported you can't um you can't mutate them with pmx and with that that is all that i have to present today and we would like to thank and acknowledge our great members especially uh Danielle and and Vitas um these two are the the main contributors to pmx and we would like to thank the audience for their attention and we welcome many questions thank you once thank you bet now we have already several questions and first we have a couple of questions questions from wow damas i hope i pronounced them correctly and i'm going to unmute you how can how can you hear us no okay i'm going to read out the question so the first question is is the unfolded wow type and mutated proteins in related the entire protein or just a portion around the mutation um right so maybe we went a bit fast there what we usually do for the unfolded state because it's it's we think it's it's a rather um broad ensemble that is difficult to to simulate in a converged way we simplify that by um simulating a short peptide and so we usually actually use a tripe peptide to model the unfolded state not even in the context of the sequence of interest but usually like a gxg tripe peptide so the the peptide of interested a flank by two glycines that also means of course that um there are also only there's a limited number of mutations that you can do in such a tripe peptide so we have them all tabulated and you can just um look them up in our tables you don't even need to simulate the unfolded state yourself if you don't want to thank you i will uh let actually how has two more questions but let's uh let's sweat uh vanguetti uh sweat can you hear us hello hi hi set yes we can hear you can ask your question oh so uh i was wondering is it possible to create topology files for disappearing residues instead of mutating them um yes um that is uh well not out of the box but it's actually um a rather um uh and similar exercise so you mean extending a protein chain on the end or the c-terminates right yeah either extending it or like disappearing at like removing the residue from yeah i mean it's it's um it's not something that you can do with the current software but it should be easy to add as a feature because it doesn't need any new functionality so if there is any interest in that then uh please drop us an email and and we'll put it on the to-do list sure thank you and i had another question so um what if you have to modify a residue instead of mutating it to one of the default residues is can pmx do that a phosphoserine or something like that yeah yeah um well then you would have to um dig a little bit deeper into the um um library files and if you download pmx you will see that um there are mutation libraries for all of the force field that we um support and um that would mean to go to your force field of interest and then for example if you want to mutate the serine into a phosphoserine um you just copy those building blocks and um use um yeah the analogy for example of what we have done for mutating a serine to a three-umeen um you use the same type of of um approach then to create a morph from a serine to a phosphoserine okay so that is well it's it's um it's not as easy as running a supported mutation for sure um but also there the pmx functionality will um should help you a long way so as long as i have all the atom types and correct structures i can just add and actually okay if you have no parameters for your modified amino acid then we cannot help you of course oh yeah that's right yeah i get that thank you okay thank you and uh question why you are on kills hello hi hi so i wonder uh you have the you intended to use gromax package but you didn't test any of the gromax force fields so there is a particular problem with this force fields for this type of application no it should be um um sort of um wait forward to also support gromax i mean we didn't have a uh a real urge ourselves to do the porting um um because in in the number of tests that we ran previously um we sort of came to the conclusion that we prefer a modern amber or charm um force fields for most purposes and so it was just a matter of priority that we didn't have around to supporting gromax um yet okay and another question is um so if i use the web server that you to generate a mutation for instance but if i want instead of uh calculating by uh affinity uh energy free energy for a folding protein but i want to calculate free energy for a protein ligand complex then i would upload the complex of the protein ligand uh and and generate a mutated let's say that i want to generate a mutation in the ligand yeah i i think that there there are two possible things that you might mean here so let me answer one by one the first one is what if i am for example i'm interested in the ligand affinity for a wild type protein versus a mutated protein then you can just use pmx as we presented it um and and if you want to use the web server then i would suggest uploading the protein without the ligand um let pmx do its thing and download the um the topology and the coordinates for the um uh with the mutation inside and then add the ligand afterwards and then you can go ahead and and you will get um the the difference in um ligand affinity due to the uh mutation if um if you if you want to mutate something on the ligand side um then uh this is something that is um um also in progress of being supported by pmx but we are still developing that at the moment so that's not something that you can run um out of the box as of yet okay yeah that was the second thing that was a what i have in mind similar to what you explained the first example you gave before go go into the 40 example you show this alchemical modification in the ligand right to calculate the free energy of magnesium yes because it's the frequent or a frequent example of how these thermodynamic cycles are are used in practice and that's certainly also something um uh we are um doing it's just that with the protein mutations in principle there's only 20 times 20 possible mutations so that's a whole lot easier than doing the 10 to the 60 square mutations that you can do in chemical space so that's why um uh we are taking a little bit longer there to reach the same kind of accuracy but just one more thing like if my ligand is a peptide and i want to mutate just for a regular amino acid could i upload the ligand alone in the web server or then it's a protein just like any other and you're ready to go okay thank you okay uh we have just a few minutes i'm going to read a few of the next questions sivastan is asking does pmx plan to support mutation of nucleic acids in the future um yes already in the in the very near future in fact so the the previous version of pmx actually supports nucleic acids it's um it's not supported now because we we went to newer force fields and also the web server but um so either if you um use a the previous version or the next version will also support nucleic acids again okay uh we have one question by musumi khazra he's asking whether the last slide the blue and red color bar represent two different states of protein system here probably um and no it's the same well both represent a transition between two states red here is the um is the are the so-called backward transitions where lambda is switched from zero to one uh from uh yeah from from one to zero sorry and blue goes from zero to one so um um both are coming from transition trajectories and they just are different in the direction of the transition thanks uh there is one question by salman serini uh he's asking whether it's it's possible to calculate the free and the combining of a minos to a metal surface in aqueous solution to a metal surface yes in in water solution um well we cannot help with the metal surface but if you have that setup then yes um PMX will generate the dual topology for your protein mutation and then you should be able to calculate the free energy the different differential binding to the to the metal surface one would only need to work out what is the proper thermodynamic cycle there so it's once associated to the surface and once dissociated maybe something like that but that all depends on the question you want to answer one more question from uh wow why gxg mimics a x a mimics would be the more would be more standard really um he doesn't have a microphone so i can't put him on okay yeah and i mean gxg there is not a real good reason to use that in um in uh when we were were um generating the software and and playing around with it and we we um simplified the unfolded state more and more and we came to the conclusion that actually the gxg works surprisingly well um there are probably also other choices that work um just as well or maybe even better it's just that sort of by accident we found that this works um surprisingly well and and and helped us to get these accuracies within 1k cal from from all from the experiment thank you berth uh this is a very interesting discussion there are a few other questions that could be followed up but unfortunately we are running out of time uh i would suggest to all the attendees to use the forums at ask.bioxcel.eu we have a category there for the free energy interest group and feel free to start up topics there we can follow up the discussion you can continue with the questions that you have today and also i would like to let you know that our next webinar will be on the 30th of june from uh the same time for p.m central european time and next webinar will be on qm and m approaches using cpmd specifically with application to biomolecular simulations this is another of the codes that bioxcel is uh working on will be presented by emiliano impoliti from new league and that's all for today i hope you will you have enjoyed today's webinar there will be a recording on the website so you can watch it later you can share with your colleagues and again keep in touch and if you have any questions please feel free to use our support forums on ask.bioxcel.eu or get in touch with us through any of the other channels thank you all thank you very thank you guys and we'll get in touch again bye okay bye bye