 Okay, welcome everyone to this workshop webinar for the BioExcel workshop in best practices in QMM simulation by Microsoft Systems. I won't take too much time, especially because people who have been to the previous webinars will already know what this is all about. Just to restate very briefly that what we're trying to do with this workshop is to really look at a variety of perspectives from very experienced practitioners who are using QMM simulation for bioelectric systems and look at what insights it gained about what could practice robust protocols for simulation, things to look out for, it holds, and also perhaps in some cases, some ideas about software to use. So to introduce today's speaker, we have Professor Maria Adriano-Ramos from the University of Porto. Maria heads the group of theoretical and computational biochemistry at the University of Porto. She has an extremely long list of publications that hasn't had a long list of high-profile propositions which I didn't want to put here, but I just want to highlight Maria's research interests and expertise, especially as relevant to the workshop. So Maria's research expertise is with regards to computational and symmetric catalysis, and there I thought what stood out to me in particular was a number of publications that she and her group have done on benchmarking and suitability evaluation of DFT functions, which I think is a really interesting topic and really valuable for people who are coming to QMM simulation, fresh, and I think she might have touched on this today in her presentation. Other areas that Maria has done research in and published extensively about dynamics, computational and geneticists, I think using a PDSA and my theatre blocking and director's covering. And Maria was also awarded in 2019 the Medina Vita, I learned so surprised by the Spanish role of science in the chemistry. I did my best. Okay, so I will, with that having said that, I will hand over the reins now to Maria. I just wanted to say hello to everybody and thank also Arno and Emiliano and Perit for organizing this series of seminars and obviously for inviting me to be here and to talk to you about studies on enzyme catalyzed reactions, which is quite a general name and today I am going to focus on mechanisms of enzymatic reactions and for that I would like to tell you first what I want to know from when we establish a mechanism of enzymatic reaction and to just give you an example of exactly what we want, my first slide features the phyldisophyte exchange mechanism, which is not terribly complicated, it consists on a nucleophilic attack by a methyl phthalate on a disulfide bond, it breaks it and another one is made. And one of my ex PhD students, Ruy Nevesh, who is now a, has a research fellowship in my group of research and has done a very nice work on the reduction of glutathione disulfide by PDI and using exactly this exchange mechanism, which is actually used in many systems in biology and what he did was what we knew at the end of it and all the other many enzymatic mechanisms that we have established is the, all the geometries and the energies of all the states that are important, such as reactants, transition states, intermediaries, products and by knowing all this we can actually visualise the mechanism, we know the steps, we know everything and Ruy has done a very nice video of which is done with all these geometries that he calculated. So this is PDI and here is the active centre and as you can see we've got for the first step of the mechanism we've got here the disulfide bond, the methyl, a phthalate that is attacking it, the bond is broken, the other one is formed and here it goes again, there's another methyl phthalate that is forming here, which is going to attack again the disulfide bond and form a new one. And as you can see, and I can't, I mean I was very happy to be able to visualise this, this gives us ideas and it's clear, very clear and that's what we want, we want at the end of the day to understand exactly what goes on with an enzyme when it reacts, yes the orders of magnitude of enzymatic rates of reaction based on experiment, so you can make an educated guess and that's what we use to do all the time and by educated guess I mean you know that if the rate of a reaction gives you a delta G of activation which goes say 30 kcal per mole or something is wrong, you've made a mistake somewhere, either the mechanism is wrong or the value calculated is wrong or the Hamiltonian is not good enough and provides that because with 30 kcal per mole the rate of reaction is too slow for any reaction to occur in your body and if it is a very important one you're more likely dead by the time it would sort of react with the delta G of 30 odd, so that's what I mean by an educated guess. So in 2015 and then it went on until 2020 we actually decided to do a computational study totally based on experiment and we went to the literature, we got out all the values of the rates of constant, experimental rates of constant, yeah and the rates of catalysis and basically what we did was we put it all together and here is and we just published that in 2020 and if you look at it and you've got here the percentage of occurrences in the individual classes as a function of the delta G of activation, so the stepwise graph is a sum of all classes and you have and the individual ones are with colors by color, so for example we started in 2015 with hydrolysis, it's the darkish blue and you can see that the dark blue itself goes from 12 kcal per mole to 22 kcal per mole but already 80% of that is in between these kcal per mole here, these numbers here and if you sum the whole lot, I mean you never have I'm saying from the values and we got about, so our study focused on a thousand odd experimental rates that were published in the literature and it very little time goes above 24-25 kcal per mole so that's what I mean if something is 30 or something is not quite right, so this is another way of you to know whether you've got the correct energy of activation and if you've got the correct energy of activation in principle you do have correct mechanism I'm not saying it's a hundred percent certain but you know you've got a pretty good chance of having it, so which are the the main problems that we were faced with during these years of work, first of all is the Hamiltonian then the long-range interactions then the reactional space and finally the conformational space, so this is a recurrent slide which is going to turn up quite often in the presentation because what I'm going to do is I'm going to focus on each one of these these major problems and tell you how we solve them, these are in fact problems that we have been dealing with forever not forever but anyway in 2008 for example we already talked about this and now we are in 2020 but we now have far more answers than we did then so now then they were real problems from the point of view that we didn't have any answers now at least we've got some answers, so with the Hamiltonian so which one should we use because there are many to choose from these are just a few CCSTD, LDNO, CCSTD, MP2, DFT, DFTB, semi empirical many more but one thing that you have to and there are protocols as well lots of protocols software to choose from as well but what we'll have to keep in mind is that whichever you choose it has to successfully deal with the nature of all energy contributions and of course which one to choose well that also very much depends on other things such as the group you start working with which will have their own expertise concerning a particular software usually a particular method as well or you know the computational power that you've got or or else even the software where you run your calculation sometimes you have access to some but not other so it very much depends on that but this is the main thing is that the Hamiltonian has to successfully deal with the nature of all energy contributions and if you're dealing with enzymes you have all sorts of energy contributions in your hands to deal with so I'll go a little bit more now with the Hamiltonian before passing on to other problems so the methods that we favor are when I say bioinformatics I put in here everything that sometimes we've got to use as well to help us setting up the system such as if the substrate is not docked into the active center we've got to use molecular docking if we don't know part of the enzyme or even the whole enzyme we've got to model it and do homology modelling etc and then we use molecular mechanics and quantum mechanics within the quantum mechanics we favor definitely DFT that's what we favor and but we also do mp2 ccsbt and even others but these are the main ones that we use we use QMM and have been doing so for many years and we now do QMM MD some of it anyway sometimes with QMM we have been running on Gaussian using onion and with QMM MD we have been using orca or cp2k mostly because they are software that are available where we can run our calculations and so we can do it without any problems and sometimes we have some problems into in setting up other software so for us DFT is the only practical QM theoretical level at the moment and and this obviously depends on the computational power that we've got the systems that we deal with however as you know functional performance is case dependent and as Arno was talking about just earlier on we started by benchmarking the DFT functionals before running our QMM calculations and we found out that in fact with some systems not all of them but with some the difference in the results was quite remarkable so we started dabbling with different functionals when running our or when establishing our mechanisms so so basically what we do is that we cut this is to benchmark the the functionals we calculate a reference potential energy surface at the ccsbt cds cds complete basic set level and then we benchmark the density functionals now this ccsbt cbs level has got to be has got to be done because when you benchmark a functional you have to benchmark or whatever you benchmark anyway you have to benchmark against something and so what we do is we calculate the something in ccsbt and we calculate the mechanism the whole mechanism with this small system that includes the zero content energy thermal effects the reactants and products of the term fluid I actually conclude I should do the same thing as you do the the enzymes as well and we then calculate single point energy values of all relevant structures that were calculated from this small system at the approximated level ccsbt cbs having geometry optimized with the three lip and you have then your reference to evaluate the accuracy of the density functionals so now you go to your whichever density functionals you want to look into and sorry I forgot to say this so but basically as a summary we fully optimize the geometry with bit three lip and then we improve the energy with the benchmark functional so when we do this we get accuracy which is close to the ccsbt cbs level in the chemistry which is the important part but it's not just the important part but obviously this is important and so you get a functional that is that mimics very well the or is very close so that's what I should say to the ccsbt cbs level and with a computational cost the dft triple the z level and a system size and cpu time scaling at the dft level which is excellent so this is our experience and so since then when we triggered that we have been benchmarking our density functions if you want to follow something like that you can use this first reference and in fact I forgot to put another one which is quite important which was a study which was done by Huy Nefs again sorry I've got it here and I forgot to pass it there which is from 2014 so it's earlier sorry the latest stage and this one was done and I forgot to put there the reference but anyway you can look it up and see this one which is basically the same sort of thing so here are the examples we did this so we've been doing them for many enzymes but these are four which then follow on to another example so that's why I chose these four and and these are the small systems that I'm talking to you about now if you focus on this one the time files and exchange mechanism it's the same one that I presented to begin with so that you would recognize all the selfish so what Huy did was that he set up this small system which mimics what happened in the active center of the enzyme if you remember the video that I showed with all the all the disulfide bonds forming and breaking and he set up the mechanism for this small system he also published here is I was looking it up but here is the reference it's JCTC 4842 in 2014 and then with this functional he then went on and did the enzymatic mechanism so that's what we do and nowadays it's actually recently we found out something which is quite exciting which is benchmarking at the DLPNO CCSTT level and if we look again at the same four examples that I just gave you a Pietro Piver which the student of ours has evaluated the performance of this method which has been which we owe to Frank Neese and his co-workers and at the same reference potential energy surfaces that we've been I mean we calculated these very accurate reference systems yeah so which are these four and others but Pietro only Pietro 5 only focused on these four because they are they they they they lengthy these things and so we've evaluated the performance we were not the only ones they have been established by Frank Neese as well and in fact but for these four this particular four which is our experience the DLPNO and so basically what was I what I was saying is that the results were very exciting from the point of view that we got a 0.5 kcal from all we were a difference from the reference ones the reference ones being the ones at CCSTT level and just two things that I was saying we didn't use any transitional mentals in this reaction just basically because they didn't have any transitional methods and they were the systems and the study and and we noticed that this is base dependent so you really have to use a good damning basis so this is more or less what I wanted to say about the benchmarking of density functionals and the Hamiltonian so what we obviously used the Hamilton the benchmarking to improve our Hamiltonian as best as we possibly can and software method delivered systematically better results than any density functional okay good let's go so this is what I basically had to tell you about the Hamiltonian or how we deal with the with that and how we improve the Hamiltonian that we choose and so we can carry on now and go to the long-range interactions now how important our long-range interactions we know that they are challenging to calculate technically speaking which model should we choose to use or I mean by we I'm talking about ourselves obviously I can only give my experience and or I I would only like to give my experience or should we just ignore them so what I can tell you is that in in an enzyme the long-range interactions are very many and they are they cover a very long range they go from 7 to 20 or can go from 7 to 20 angstrom so we have to be careful with them and I still just I don't have much time so I shall refer once study which focus on beta galactosidase which is an enzyme that catalyzes hydrolysis hydrolysis reaction of lactose into single sugars and as you see here so this is what really matters to me at this moment the yellow is the delta G of activation and this is a the active center of the enzyme this is a cut of the enzyme and this is another one so when we had a very small oops when we had a very small model of the enzyme with just obviously two small 35 atoms in the qm part we got extremely high delta G's of activation uh we then went on to 227 atoms in the qm part and sorry in the in the in the um yeah it was the qm part no this was a qm model which in total had 227 atoms so we cut the models so basically we used the cluster model and we also got a very high um delta G of activation but when we use a large part of the enzyme it wasn't the the complete enzyme but it was a large part of the enzyme then yes we got 15 kcal per mole always the same mechanism that was studied and which was was okay with the experimental result which is also 50 kcal per mole so in this particular case these long range interactions were determinant to get a good a good value for the delta G of activation which helped us setting up the correct mechanism um is this and there were no conformational changes here uh these these were really just long range interactions without any conformational changes so is it always like this no it's not so there are cases many cases in which you don't have such um such important long range interactions and they sort of um cut off much nearer the active center but you don't really know in advance when that is so so you have to be prepared uh for the long range interactions uh to be uh quite important for your calculation so we tend to use if we can the whole enzyme or at least the whole monomer if it is a case of uh many um of a dimer or trimer or whatever um but we we we are careful with the long range interactions we think they are important and we want to include them if we can um so for the reactional space well there can be great diversity of the reactional space with often quite unexpected chemistry when compared to solution chemistry so this means that you're in poor problems with the reactional space from the point of view that if it is complicated reaction your you have to uh sample this reactional space very well so you set up your hypothesis mechanistic hypothesis which can come from experiment or not uh and uh but whether it comes or not you really have to study all the reactional space which can be complicated and i'll just would like to uh run you through an example which sort of uh uh probably scarred my uh one of my very talented phd students build a fighter which very recently established the catalytic mechanism of human aldehyde oxidase and the whole thing's a bit of a nightmare because as you see the reactional space is very diverse it had two molybdenum atoms and they kept changing oxidation state uh quite a lot of steps of uh uh mechanistic steps and uh and it took a long time so um my advice if i can uh sort of give to uh each of the people that are not so used to doing this every day um the reactional space can be very complex and you really have to think that the more complex the or Hamiltonian the more difficulty becomes to solve the problem so you have to compromise uh you can't use a Hamiltonian that you know that probably we could give you a wrong uh answer but you have to compromise and still get one that will be able to give you the correct answer and also um establishing for me establishing an enzymatic mechanism is like solving a puzzle to a certain extent i every single bit of correct experimental or otherwise the information must fit and cannot be left out so you can't just say oh um this experiment or this experiment or this experience uh gives gives us i don't know the pka of histidine which means that it is um solvated or not or uh uh uh or not or whatever uh you can't just say oh i will disregard this no you can you have uh to to to fit it in and uh and nothing should be left out everything has got to to fit and so that you know that you really have got the the right uh mechanism and finally the uh conformational space uh which how determinant easy to sample enzyme conformations uh so a lot of people do q m m m m t kappa inelo metatynamics and all this and they do uh excellent work i mean really first class work uh other people don't work kappa inelo or anything else that includes dynamics this is not don't don't take me wrong uh is that i we have been doing it basically because we couldn't afford for a long time we couldn't afford doing other than q m m m and uh and we find that still do uh that we get a lot out of it a lot of detail out of it and i'll give you an example so that you understand my point of view and i hope you do um so let's start with single confirmation uh calculations so i'm talking about q m m m in which you start with a confirmation to start studying your uh your mechanism your your reaction mechanism and with it we can identify structural features of reactant conformations and understand very well the structure activity relationship uh so uh let me uh present to you alpha annuase uh this is what the enzyme looks like this is the active center uh uh we've got the sugar for substrate and two catalytic uh or two mechanics sorry uh two residues that are catalytic uh so we've got this glycosidic oxygen that is um sorry here that is protonated um by uh glue uh 233 and what happens is that this bond breaks and then there is a nucleophilic uh addition by uh 1 9 6 so it's not terribly complicated we established the mechanism for um the enzymatic mechanism we used onion so we did single uh confirmation uh calculations just one started with one confirmation that was basically the x-ray one um and uh we still had to optimize because of substrate to be put in place and all that we include we used um so p3 lip for uh the uh the geometry optimization and then m 06 2x for the energy um so we basically benchmark uh we included we did everything so included cpe so thermal effects etc etc and um um but we started with just one confirmation so you know the the doubt always uh is uh is there thinking uh what if i used another confirmation to begin with and to take away that uh problem we decided to run uh a dynamic simulation along one and selected snapshots from that md simulation uh these snapshots were not selected at random because you have to understand that when we run uh a long dynamic simulation the space the conformational space that you sample is incredibly large there is absolutely no way that you can sample or they can you can study with detail all these um all these uh different conformations so we wanted to sample those that were very near uh the the um uh the start of the catalytic mechanism because those are the ones that are going to be important for catalysis which is basically what we want uh start you know study catalysis but uh the catalytic reaction but with a different uh confirmation so um these these uh conformations that we looked for were those that were predicted to be adequate uh that uh that uh sort of uh uh were um followed adequate criteria for reactivity and how did we do that uh so we looked at our at our active center i included here well here are the catalytic uh residues i included this one as well last 300 because it's also it's not catalytic it doesn't go into catalysis but it's a structural um a structural uh residue because holds the sugar in place because of doing uh hydrogen bonding uh bonds with um the hydroxyls that are uh on these two carbons so what we did was we looked for the conformations that uh first of all the distance from arse 196 to carbon c1 was under uh 3.5 uh and strongly so i kept here the mechanism so that you are you remember it and then we also uh at the same time these conformations had to um um had the structural hydrogen bonds so these hydrogen bonds between arse 300 and the hydroxyl groups attached to c2 and c3 and the 2.5 angstrom uh so that's what we looked for those conformations and we selected 40 of them uh and we for each one of them we solved the mechanism yet again so we ended up with 40 free energy barriers of activation for each of them we repeated the exact protocol that we have followed for the first uh single conformation uh the x-ray basically optimized uh that we used to begin with and this is what we got so we got what we called a multiple pairs uh analysis so to speak in which you can plot this the activation barrier as a function of time and a function of each of the uh initial structures that you start with um to establish uh again and again again your mechanism and for the x-ray so the experimental value is 14 uh the x-ray we got 11 and look at this we have a very wide range of of barriers of activation and you have to remember that these were all chosen as being catalytic or near catalytic uh uh structures or structures good for catalysis so to speak with the distances good for catalysis and uh and these are the only uh in 40 of them uh the only four that we managed to catch that really uh could be uh important for catalysis so basically what happens is that you sample all this conformational space most of it is no good for catalysis and then eventually comes a uh a structure that sort of fits into those criteria for reactivity and and and you have a reaction so we've done this in the past for many many enzymes I just put here these this was the very first one in 2012 for HIV1 integrase but then we followed them went on and this is also the alpha amylase as I told you this is just to tell you to show you that it wasn't sort of something that is very strange it doesn't so this follows a pattern so we can say that the small number of barriers accounts for the very most of the observed rate this is experimental averaging so to speak and the average barrier is very similar so these average barriers the the the ones that are good for catalysis are very similar to the x-ray single conformation free energy and that's what we always found if there is not some big problem with the x-ray file obviously uh so going back to alpha amylase we looked into all the structures the 40 structures that we had initially selected um and we saw that these low activation barriers only occurred if a set of geometric conditions was verified and and it wasn't exactly the ones that we anticipated and that we looked for to begin with so what we found was that uh okay and a water molecule that goes and comes back so to speak and this was very surprised because I mean these distances are implicated in the reaction coordinate but this one is not um but so what eventually so we had to study this obviously and we came up with the conclusion that what happens is that this H bond lowers the activation barriers and basically it controls this the pka value of this general acid uh glue 233 and therefore the water molecule switches the enzyme on and off and I say on the time on the nanocyclin timescale because that's the one that we analyzed uh so this is a an example that I wanted to present to you uh to show you uh that these QMM calculations can be very interesting to find all sorts of details uh and just to see what we could find out by averaging over an ensemble of conformations since we only looked at one at a time we went on to do QMM MD calculations on the same amylase with the same um mechanism of reaction and so these are here the QMM results these are the QMM MD results I'll just quickly run you through this so that you understand so here for the QMM these are the results that you get out of that uh MD simulation from which we got the 40 initial structures so the black dots represent the conformations obtained during the MD simulation of the solvated enzyme substrate complex all these black dots and the colored dots represent the conformations that those 40 conformations that we got out of the MD simulation and which conform to those initial that initial initial criteria for reactivity that are pointed out to you and so the the green dots have got a lower activation barrier and the the red dots are those that go up to 30 kilocals per mole and what we found out in running the QMM MD simulations is that if we started with a red dot confirmation that was a bit more out of the the catalytic one still one of those 40 and we had some trouble or not some trouble but it took much more time to make it converge to the catalytic value to the to the the delta g of activation that the real real the one that the best one that we could get and it took quite a lot of the sampling to do that but if we went to a green one for example then we were already there it didn't take any time and to a yellow one it also didn't take much time so we started here and went on and and got the the the the three of the more accurate one so to speak and if we started with the x-ray we are also already there more or less so basically what I want to say is that these calculations are very robust very good and but these others gave gave us enough a lot of detail that we wouldn't have got from this because obviously you can't go on and study many of these confirmations you will get lost in this amazing conformational space that is sampled so obviously you cannot do the two of the time we were actually having a conversation this morning with Peter Freire over this but that really the best thing is to use several methods and see what comes out of this that's how you learn to know your design very very well so we have other the examples that also behave like this so João Coimbra is also doing calculations on protease hiv1 protease which take us to the same the same ideas so to speak and so well I've run used to through all these problems it's now time to to stop as conclusions well to say some of them activation free energy substrate binding free energy and the enzyme efficiency fall in a very narrow range of values for all enzymes as we saw the Hamiltonian has to successfully deal with the nature of all energy contributions if using the ft benchmarking functionals before embarking to mmn calculations is good practice the subset of catalytic competent conformations can be significantly small in comparison with the full conformational landscape of enzyme substrate complexes usually you find x x-rays okay as a starting confirmation if there are no major errors obviously of the x-ray confirmation single confirmation to mmn calculations enable us to understand better the structure activity relationship however many other techniques that explore more efficiently the conformational space of the enzyme during catalysis are used providing a more dynamic picturing to the analysis of the potential energy surfaces associated to catalysis and I just would like to thank first of all professor bill of finance we've got we have my colleague we have been collaborating for years my three this is not all my group but they are the people that contributed the most to some of the things I said here so Rui, Joanne and Nathesia are have got research fellowships they were former phd students of mine and then my two phd students Peter Freire and Peter Piver so Peter did Peter Freire did the AOX mechanism Peter Piver did the MPNO CCST team work and these are not my two students but they are my ex phd students uh so the oracle center martins is now at Scripps and Rita Kalishtu is at the university of Uppsala and well here are some of the people that have been helping us both with money and computational power and well and I would like to thank you for hearing and now you can ask questions if you're interested thank you thank you very much Gloria that was I think really really valuable for people who especially are coming new to this to see this outline very clear outline of you know the different things to take into account the protocol um I'm sorry if it's too long do I have my biochocolic Scherz-Ameriano would you like to ask any questions before I do yeah so while we wait for questions to come in I can already ask a few questions that came up while listening to your to your very nice talk um so when you do the MD simulations to scan for or to look for conformations from where to start the QMM and what you didn't find is you find different barriers however the time to to to make transitions between these conformations is apparently really short in the order of nanoseconds so guessing that the barriers between them are really small so what I then do not so well understand and we also encounter this so it's not that I'm doubting the result it's just that I don't understand why during optimization you then don't cross that small little barrier and then take the lower barrier confirmation that was uh that belongs to a to a different type of confirmation perhaps one is closer to the x-ray structure can you comment I'm sorry you're why can't because I only just heard the very very last part of your question this thing just went off so if you I'll repeat yeah please yeah yeah so I'll repeat the question so in the in the plot of in the in the last part of your talk when you're discussing the the the contribution of sampling conformations there you see that uh if you simulate the the system for a while a couple of nanoseconds and if you're taking different snapshots at let's say nanoseconds apart if I recall correctly you find that the barriers they vary widely my question is what I don't understand and this this is something that that we also observe but what I do not understand is if you do an optimization why does the optimizer not bring you first back to the original confirmation because the barrier must be super super low because it's sampled with a nanosecond so the barrier cannot be very high between these different conformations why does the optimizer not take you first back to the lowest say energy confirmation from where the energy barrier to the transition state is the lowest well because it can be very far apart in the potential energy surface and the uh if you think well they are but it's only 200 picoseconds for example so between 11 and 11.2 there's only 200 picoseconds difference and between well 13.2 and 15.6 only two nanoseconds different you cannot cover so much conformational space in in such a short time I would assume yeah but it with the potential energy surface the algorithm of optimization is not good enough to to do that so when you try to to use a qmnn method that doesn't happen so you you you do the best you can but it's not it just takes a little bit of of change in your activation in your site in your active site to give you a very high barrier uh and uh the the problem is the uh and you can show this mathematically in fact the uh and that's a preconceived sort of a preconceived idea that people do have is that the the the um the catalytic confirmation is the lowest uh that you can find but it it doesn't have to be it can be it doesn't have to be uh in the um in the lowest point so we we it does make a difference and also the the the algorithm that you use and we use several whatever we can uh are not good enough uh to immediately go to the one that you anticipate to be the catalytic confirmation um so uh this is what we get and uh it happens with all the cases that we have looked into uh and uh well that's basically what I can tell you okay the rest now being quest there has been questions I have assigned it to you so I switch off my my microphone now uh Torio says very nice presentation uh regarding multiple PES calculations clustering analysis is going to help in order to reduce the number of structures from a large MD simulation that's a question can PES calculation clustering analysis help in order to reduce the number of structures from a large MD simulation yeah yeah yeah we we we did that as well uh so clusterizing you know and then uh and then um uh truck to get to the uh catalytic structures from there that's something you can do as well but the answer is always the same uh so if you get something that is a little bit different from um from the um um the the catalytic uh distances then the activation barrier is very high um so um that's basically that part of the other part of the question was do you have any study comparing clustering analysis structures with the protocol that you're working with you may have already addressed that um well yes but uh uh yeah uh uh I can if you send me your uh your email uh I can send you some references um I can't yeah that's fine okay we can do that who do you want to put your email into the question then we'll get you in touch yeah yeah that's not a question I think um um the question is can we correlate the calculated free energy barrier with experimental k-cat or k-m values if the rate limiting step is not known if so what is the best way to correlate them and if we know it is a good correlation can we use that as evidence to confirm the identity of the rate limiting step in enzymatic reaction oh yes and yeah well it's it's really the the um rates of reaction they vary so little they they they really are uh so if you don't know them you can always make as I said at the beginning that educated guess because they really most of them are in within the range of 14 kcal per mole and 20 also um that's that's a very very narrow range of values they're all there 80% are following that uh uh in that region so you can correlate anything like you know as soon as you get those values basically hang on I can see Amani's question now so um yeah yeah yeah yeah yeah so uh yes with the iranian law you can correlate sorry actually I understood the another question you can calculate the free energy barrier uh from k-cat so basically what I was saying is that you can calculate the um the delta geovactivation from the k-cat there is an equation that lets us do that iranian's equation so you can also go the other way around if that is your question so if you have the delta geovactivation you can go back to the k-cat as well so another question is in your multiple pers calculations do you impose any kind of restraints in the coordination to many groups of the ts structures uh so uh as I so um I imagine that in your multiple pers calculations you mean by to you mean that in the we are using your we're calculating structure by structure yeah one confirmation confirmation at a time so when you do one confirmation at a time no we did not impose any restrictions in the coordination to the main groups of the ts structures or at least we try not to unless we see that um uh there is or we know that there is a particular interaction that has got to be there uh and then we can restrict it say uh two groups or whatever but on in principle no we don't I can't see anything else okay so I can I can read out uh the question um so uh thanks for the very inspiring talk to somebody else I have a question in case the x-ray structure is not available would you still recommend to perform such calculations using a homology model well um if you have to use a homology model because you really have to um do some sort of project that in which you have to um then you know um it's not such a good idea or it can be difficult because if you make a mistake even if it is a small one with the homology model then you might not get to the correct mechanism or at least if you are in the correct mechanism you might not get to the correct um uh barrier um this doesn't mean that we don't do uh the mechanisms with homology modeling but they always pose more problems than the others that's our experience in a matter so I'm not saying that you shouldn't do it or that we don't do it I'm just saying that it is um we're in for more problems usually um so he responded actually saying thanks it makes sense so thank you for that there's a follow-up question to head of the question uh from Dmitri thanks for the very nice talk right and I have the following question so basically it's kind of following up to the get it question uh so basically when you do first this kind of PS calculation yeah and in the next slides you also show showed the uh what happens if you do QMMMD free energy calculation basically I suppose it's a raw assembly for something like that for metodynamics but for me seems like you can take from at least from the QMMD you can take even a very bad starting structure and after some not so long like vibration time with QMMMD and something you still can get a correct energy barrier no yes of course so yeah if you are you want to see me if I could do the multiple pairs out of the QMMMD simulation my question is what is more expensive and what is should be done in general so you one one thing you can do a lot of PS population well second you can do QMMD long QMMD you you can do QMMMD and still look into single conformations obviously I mean nothing stops you from doing that but the amount that you sample is such that I'm not quite sure how easy that will be so what we like to do is or personally what I like to do is QMMM because I think it's the much more fun QMMMD is more robust I'm not going to dispute that but it's very automatic as well I don't like very automatic things I always find that if you have to think about something you get more out of it but QMMMD or multi meta dynamics or whatever are more robust methods so I think that it is a matter of it all boils down to a matter of personal taste and what you can get out of the methods that you have at your disposition the amount of computational time that you can afford the software that you have available etc etc and that's why you've got so many so many groups with all doing different things to basically get a mechanism of an enzymatic reaction somehow okay yeah yeah thanks yeah so I don't see any other question from oh Gerrit still has a question that you would like to ask Gerrit go ahead you can go ahead well always yeah the question I have is going back to the observation that you find that the barrier is always around but was it 40 kilocalpum all I wonder is this because enzymes have been optimized to work at the diffusion limits so that at a certain point the time at which it takes for the for the reactants for the substrate to actually find the enzyme is becoming the rate limiting step and further optimization doesn't make any any sense from an evolutionary point of view sorry I didn't hear your question okay so this goes back to the observation that all the barriers fall in the same range that 80% or so in the interval of what was it around 20 kilocalorie per mode and I was wondering is that could that could you speculate whether that is maybe due to the fact that the diffusion limit would probably then become the rate limiting step if you optimize the enzyme further so there would be no evolutionary pressure so to speak to actually make the enzyme even faster because then the rate limiting step is the diffusion process of the substrate into the active side so this is what you're talking about right yes okay so this is the activation energy that we get out of experiments so to speak so I think that what it is I mean it's related to the fact that we need the reactions in our body that are fast enough to happen in a very short time I mean if my trypsin that takes hours to calcium amino acids I'm going to have indigestion that's for sure so this is connected the the the the rate of reaction and the the delta G of activation is connected with that you're talking about evolution if I heard correctly yeah so what I meant is that the enzymes have been optimized throughout evolution to reach the point that the rate limiting step is now the diffusion process so that the time it takes for a ligand in this case a peptide or whatever the ligand is for the enzyme to move into the active side so the if the enzyme would be more efficient than that then it don't gain much because the rate limiting step would then be the diffusion would be the diffusion limit yeah yeah I understand that yeah but in fact the it can be also other other other roles can be other physical process can be in play mass flow diffusion mass transfer area all this can come into play now if if you ask me that evolution has optimized the enzymes for the diffusion process I'm not sure if that is the case because I think that with this apparent apparent activation energy that comes to us from hello yeah yes yes we lost you at this activation this apparent activation energy is coming from something so okay so these are the apparent activation energies that come from the experimental values and experience experiments coming from labs that usually do these things now they don't know exactly if reaction rates come from the the diffusion processes or mass transfer area for example that will be into play as well mass flow it can be from all sorts of as well as the chemical processes as well also so it can be several factors and I am not sure and I don't think anybody is if mass diffusion is the major player and it comes from evolution I personally am not very convinced about that I think there are the factors into play that's my idea do you have any idea about that no no but the observation yeah no but the observation I don't think so yeah sorry no no I was just gonna confirm that no I don't have a clear idea either but the observation that barriers are all in the same range that has been observed yeah in several studies so I was thinking there must be something underlying it yeah yeah well it's got to do with the with the fact that with very high barriers you know I mean our our body wouldn't survive but if exactly you know enzymes were optimized for that I think there are many factors into play I don't think it can just be that but you know this is just a speculation good I asked for the speculation so this is fine thank you yeah okay thanks I I have one quick question one last quick question and it's about software so you said that in your group you use both cp2k and orca I was just curious whether there's a preference for one or the other in the context of the availability of particular basis sets particular the functionals or whether it's about what size of number of qm atoms you're dealing with or whether it is to do with your computational resources basically when do you use one which one and when you use the other and why um whenever we use we want to use dp the lpno the ccsbt we have to use orca because it's implemented there and for so that is one reason why we use orca but orca is quite nice to use and we were we started with the cp2k because we started running in oakridge national laboratory and it was difficult to set the other software there and so we were a little bit pushed into cp2k whether we like it or not we had to start using it and then we sort of got used to it basically because you start with one software and you start getting used to it but we run both and probably at the moment we're running more cp2k than anything else but as I said if you want to go into the lpno you have a ccbt you have to go into orca to do that we like them both really provide the answers yeah okay great thank you very much so with that I think we've come to the end of our question and answer session as well so I just wanted to take the opportunity on behalf of you know all of us organizers from bioxcel and all the attendees as well to to thank you very much for the talk I think it's very valuable for people who are coming to this maybe fresh to draw on your experience and you know highlights in different different aspects that you take into account and especially sort of this is benchmarking protocol I think I think it's quite nice for functionals so okay so next Monday we have another webinar coming up that's relevant for this workshop part of this workshop which is agent mahaland will be presenting on the topic of tubers chemical accuracy in qm modeling also enzyme catalysis and also protein vegan binding so it's all lots of good stuff still to come as well next Monday but for now thank you again Maria for for your presentation and for dealing with the go-to-webinar technical issues and yeah thank you very much okay thank you very much okay bye bye bye