 So welcome everybody. And now we start the student webinar. This is the by Excel student webinar. So our students that have participated to the summer school. 2020 second edition. And those students have won a poster price and the poster price was a webinar. So the presenter of today are Costanza pattern monster from the University of Trento, record Perry from the University of Newcastle and Serena are key to from the front comes Institute for advanced study from the international Max Planck research school for cellular biophysics. I'm hosting this webinar together with Martha from the European bioinformatics Institute. The webinar will be recorded. And during the webinar, you can use the Q&A function that is located on the bottom of the Zoom panel, and depends on the operating system that you have you can see this symbol, or you can see the symbol. And since we have three speaker I will ask you when you type your question to put at one for Costanza at two for Russian and at three for Serena. And at the end of the three presentation, we will read Martha will read the question and share the Q&A section. So everybody can hear the question and the speaker in charge will answer. Okay. We have three different topic today. So so Costanza will speak about poor forming toxic, in particular investigating the action mechanism with molecular dynamics simulation. Rafael will speak, will speak about the Riemannian geometry for molecular suffrage approximation. And Serena will speak about mechanistic side into the early events of the activation of the C metal receptor during the stereo invasion. So now I give my the work to Costanza please. Good afternoon everyone. Thank you for the introduction now I will share my presentation. I'm going to talk about a part of my research project which concerns, which is focused on the so called poor forming toxins, and how I investigating them with the aid of molecular dynamics simulations. So the poor forming toxins are particular kind of proteins that are released by bacteria during the host infection. And there is a huge variety of these toxins that differ from for sizes and structure but they all share a similar mechanism. They are released as soluble monomers by bacteria, and they attack and able to bind the membrane of the cells of the host. There they assemble in an oligomer and form a transmembrane channel so that like the one that you can see in the picture that eventually led to the death of the cell. In particular, I am studying the gamma hemolysin, which is a poor forming toxin used by the field of cocos to evade the new system of the host and gamma hemolysin is released at two separate separate soluble components of the so called you calf in blue. HLG2 component that are able to target and kill the leucocytes so the immune cells of the host and to form the transmembrane channel depicted here. So as you can see they have a high structural similarity, but also a low sequence similarity which in the end accounts for the different roles that they have in the binding and in the assembly process. So here it is shown are shown some key domains of the two components. The cap is the more rigid one, the ring domain, it is the one which is involved in the interaction with the membrane, the stem in green is the one that undergoes a conformational change and then constitutes the transmembrane channel, and also the amino latch is reported because it's unfolding, it is important to allow the interaction between the monomers in the poor. So, in brief, this is the mechanism of the poor formation, there is an initial stage, as I said before, where the toxins are released as soluble water soluble units by the bacteria. So then they have to bind the membrane. Then the two components interact on the membrane in order to form a heterodimer, and then the heterodimers have to interact in order to form an octamer, so an oligomer. That led to so called pre poor, where the stem in green is partially refold, as you can see, and then this brings to the final transmembrane channels where all the stems are completely full. Unfortunately, there are just several stages of this process that have been captured by experiments, and so molecular, you can say dynamics simulation can complement this experiment by reproducing the dynamic process and from these dynamics we can extract some relevant information, for example, we can ask ourselves something about the ensuring mechanism or also the interaction and then the conformational changes of the toxin. In particular, my goal, my aim is to study the formation of a membrane bound dimer from the interaction of the two components on the membrane. And to do this, I'm employing a full atomistic molecular dynamics simulations. The problem is that we lack a proper experiment experimental structures of the bound dimer for comparison, and it's also so difficult to think where to start for this dimer and which features we should inspect. So, I started from the results of Thomas Taranti, who is postdoc in the group where I work, he performed some cross grained molecular dynamics simulations of the process of binding of the two single components, which are HLD2. So he tested their ability to bind the belayer membrane using using purely lipid belayer membrane which makes composition of partially unsaturated lipids, which are the phosphatidicoline and cholesterol. What he found was that there are some differences in the binding ability of these two components, as we can see from the distance of the components between the components and the membrane and the main brain over time that UKF is able to form a stable binding whereas the other component is not. So if we also add the experimental evidence that the gamma hemolysin has a permealizing activity, also on this model membrane or we can say on main brains, which are purely lipidic. We can say that this suggests that the, the binding of the UKF component is necessary is preliminary for subsequent binding of HLD2. I used the information to start my research to find a natural dimer so I started from a membrane bound UKF component on the same kind of membrane. And then I run some atomistic molecular dynamic simulation of the single monomer. So then the resulting structure was the starting, we can say structure for the interaction with the other component. So the question was where to put the HLD2 monomer, which has to be in solution. And also Thomas one showed that the UKF bound to the membrane had a tilted orientation, which exposed a solvent, which expose a surface and the surface was found also as one of the interproteomer interfaces in the final poor, because we have the structure, the crystal structure of the final poor. So this surface can also be thought as a dimerization interface, moreover, so this is the angle that you said. So moreover, this surface also contains the amino latch that I mentioned before, which is disordered in the final poor, and so it is supposed that it's unfolding is needed to totally expose the interface. So I replaced the second monomer close to this exposed interface surface and moreover, because, because of the fact that you kept is not tilted in the final poor. Probably the interaction with HLD2 should induce a reorientation. So I was able to run one microsecond of simulation. So I simulated the interacting monomers, and I wasn't able to find an heterodimer in the sense of membrane bound heterodimer, but in a replica I could find the two monomers interacting with the structure mobility of the two monomers, starting from UKF, especially I found looking at the root mean square fluctuations, the average displacement of the residues during the simulations that the amino latch, so the N term and the C term were particular, we can say fluctuating also with respect of the same RMSF that I found in the simulation of the single UKF. So we, I saw more mobile, more mobile terminal terms. So also the RMSD confirmed this behavior, and also here are reported to configurations of the amino latch during simulations. So two representative configurations, and I also also investigated the structural mobility of the other monomer. It was that interesting was to find a jump in the RMSD of one particular part of the HLD2 monomer, which is the RIM domain, the one that should interact with the membrane. So this we can say reveals kind of local rearrangement in the RIM domain. So I wanted also to inspect the interface area to tell if the two monomers were really interacting during the simulations, so I estimated that this interface area using the SASA, the solvent accessible surface area. So to compute this interface, I simply calculated the SASA for a single UKF, the single HLD2, I summed up these two and I subtracted the one from the dimer. And moreover, I wanted to compute the contributions of the different domains to this interface. And what was the result was that the majority of the contribution comes from the RIM domain of both the components. So I also investigated or computed the angles between the axis of the UKF, which was tilted at the beginning of the simulation, and with respect of the membrane normal. And here I reported the distribution of this angle. In red, the distribution of the UKF axis angle during the simulation where I just simulated the single UKF, so you can see that it is really tilted the monomer. And then in grey, I reported the distribution of the same angle, but when UKF is interacting with the other component and during the simulation time, we can see that this distribution sheets toward lower angles. And I also reported the value of the orientation angle that UKF has in the final pour. So, I run some molecular dynamic simulations in order to find to investigate the interaction of the UKF and HLD2 components of the gamma hemolysin in order to obtain an heterodimer. We can see which yet proper stable membrane bound diamond formation but some preliminary results. There are some preliminary results we can say that there is an high mobility in the UKF terminal beta sheets, one of which is the amino latch is supposed to have a pivotal role in the HLD2 component because it has to unwind to better expose the demeritization interface, and also found that the ring domain of UKF but especially the HLD2 component as a dominant role in the interface so that has to be elucidated, and also that I saw some changes in the distribution of the UKF due to the presence of HLD2. So, thank you very much. I finished. And I have to thank my supervisor who is Professor Gianluca Lattanzi and the Thomas Tarenzi who worked with me and at this research. Okay, thank you. Thank you very much. Okay, and now we go on with Thresho. Good afternoon everybody, and thank you for the introduction. And today I'm going to be talking about the work that we've been doing as part of my PhD, developing a new method for approximating the 3D shape of molecules by applying mathematical theory of Ramanian geometry. This is in the case at the beginning of a drug discovery project that with the structure of the protein that we're trying to target is unavailable, which prevents the use of structure based methods like docking or FEP and the identification of potential hits. Instead, what we often have are a small handful of molecules that are known to bind to our target. And from initial experiments, they could be natural products that we want to mimic, or often in this case, they might be another company's drug that we want to compete with. And we can use these molecules as templates to screen large databases for other potential hits on the basis that molecules that are similar in some way are likely to share biological activity. However, the problem with similarity is that much like Putey, it lies in the eye of the beholder, and there are many equally sensible ways to justify how similar to molecules are. One approach that's gained popularity recently is comparison based on molecular shape. The shape of a molecule is a good predictor of whether it's likely to display activity for the protein target, as it needs to be the right size and shape in order to fit into the binding pocket. There's currently no absolute definition of 3D shape, and instead we depend on mathematical approximations to condense the shape down to a vector of numbers that we can then compare. There are currently three main categories of shape similarity methods. There are those that depend on the overlap of volume between two molecules. Those that use the distribution of atomic distances from a set of fixed reference points, and those that make use of the molecule's surface to compare shape. Compared to other approaches, the use of the surface is still in its infancy and drug discovery applications. As molecules don't have a true surface in the same way that an apple has a skin, but it's still a useful feature to consider in order to describe its size and shape. The surface offers a good compromise between the other two approaches as it captures a lot of the same shape features as the volume does, but it's less expensive to calculate. So the primary focus of my PhD has been the development of a new surface based shape similarity method using the mathematical theory of Riemannian geometry. This allows us to produce a vector descriptor of molecular shape that's quick to calculate, easy to compare, invariant to rotation and translation of the molecule, and it doesn't require the optimal alignment of two molecules to be found before their comparison. We've implemented this in Python and the code for the project is all available via github and I will share the link to that at the end. I won't have time today to go too much into the maths, but the basis of our approach is that there's an object associated with the surface known as the Riemannian metric that captures in detail the geometry of the surface. It's possible to approximate this, but by treating our molecule as a series of intersecting spheres the van der Waals radius to begin with, we can obtain the metric in its explicit form, which captures a lot more detail. These metrics themselves can't be compared, but they can be used to compute the Laplace-Baltrami spectrum associated with the surface, which gives us a nine element vector representation of eigenvalues that describe the shape of the molecular surface. These descriptors are easy to compare for two molecules using the inverse Bray-Curtis distance, which gives us a score of zero if they have no similarity and therefore are likely not to show biological activity for the same target, or one when they're identical and very likely to have the same activity. So in the development of our method there were a few factors that we needed to consider. The mathematics itself only consider surfaces that are free of poles, so we treat any rings such as benzene or pyridine as a single sphere with a radius of 2.25 angstroms, which we can see as the bigger spheres in the pictures that I have here. This does give us the limitation that anything that is truly shaped like a donut is off-limits, so we can't consider microcyclic molecules with our method. While the descriptors are rotation and translation invariant, they do depend on the choice of the initial atom that we use to construct our descriptor from. For consistency, we take this as the atom closest to the geometric centre, which I've shown in grey in the image in the middle here. The description of the shape is best about this point and the quality tails off the further away that you get them. And finally, in order to simplify the maths, all of the surfaces that were considered are treated as having a surface area of 4 pi, so we include their true weighted surface area in the final description to account for differences in size. As an initial test of how well our method might perform, we made use of a series of PDE5 inhibitors of known shape. So, we took Viagra as our starting example and two of the follow-up drugs. Vardenifil, which is a classic Me2 follow-up to soldenifil that only has a few small modifications in its structure. And Tidalifil, which has a completely different scaffold that is known to occupy a similar volume in the binding pocket and has a much better performance as a drug. A soldenifil to Tidalifil jump is a good example of the main advantage of using shape similarity. Because we're considering broad shape rather than any sort of specific chemical detail, these kind of changes in structure can be identified. This is particularly useful for either trying to compete with an existing drug to establish an IP position and is also helpful to address unwanted issues such as problems with toxicity or insolubility. In theory, both Vardenifil and Tidalifil should be classed as having similar shape to soldenifil. So, compared to other approaches of doing shape similarity and a chemical fingerprint-based method, at first pass our method seems to do quite well and classifies these three molecules as having quite similar shape. In all cases here, a score of zero means there's no similarity and one means they're identical. So, while USR-CAT and ShapeIt, two of the other shape similarity methods, give the same relative ranking as we do, the scores that we get are much lower than the typical threshold of 0.7 that's used in cheminformatics to identify similarity. So, while this study isn't enough in its own to prove the methods capabilities, it certainly does indicate that there is potential. One of the other main benefits of using shape-based similarity is the ability to account for different comfomers, and so we also wanted to investigate how our method would cope with this. To do this, we took 10 random and 10 low-energy comfomers for each of the three PDE5 inhibitors produced using RDQIP. And from the overlays here, we can see that we'd maybe expect slightly higher similarity amongst the low-energy comfomers than we would in the random ones. So, the plots that I've shown here show the pairwise shape similarity comparison for each pair of comfomers in the set, compared to the Ritman square deviation of their atomic positions. And what we see is that for the most part, shape similarity between comfomers with the same molecule is generally quite high, despite true differences in their atomic positions. This is likely a result of the insensitivity of our method to small deformations in the surface, which in theory could be quite advantageous as it would allow us to ignore comfomers where only minor changes have occurred, and we could capture the same shape information with a much lower number of comfomers, which would offer a big efficiency advantage when we're screening through big databases. So, we also wanted to investigate how well our method might do in a real-life virtual screen. As a small test case, we took a set of 43 further PDE5 inhibitors with known nanomolar activity, and used soldenifil, bardonifil and tidalifil as queries to rank them. As these are known actives, we would hope that the method would assign them as such with a score greater than 0.7. So the top five scoring molecules compared to each query, we do observe a very high similarity of over 0.9 in each case. And inspecting the space filling models by eye suggests that these do genuinely have similar shapes. It's also worth noting here that we do get a slightly different top five for tidalifil compared to our two other closer structural analogues. So diverse eight amongst our queries will also be quite important in a real screen to obtain a diverse set of potential fits. And we also use this study to take an initial look at how many comfomers we might need to use in order to get good results from a real virtual screen. So we compared the scores obtained using just a single random comfomer of each of the test and query molecules. The binding pose of the query taken from the crystal structure, which is the ideal case if it is available, compared to multiple comfomers of the test molecules. And then finally, we've generated a series of multiple comfomers for both the test and the query using our decade. And the number of comfomers that we took in each case was determined by the flexibility of molecules with an average of around 200 used per molecule. And so compared to the single random pair, the consideration of multiple comfomers led to consistently higher similarity scores, which shows that the consideration of multiple comfomers will be important to capture the true shape and find all of the possible hits. And we can see that the pattern that we obtained for the multiple comfomers comparison also closely matches that of the binding pose queries, which suggests that we've completed sufficient sampling of the conformational space in order to capture the binding pose that the molecules take up. So hopefully that I've managed to convince you that the similar property principle is a useful tool in drug discovery and is efficient way of screening large databases. And we've introduced the use of Romani geometry as a method for molecular shape approximation, which gives us a nine element vector descriptor of molecular shape that's alignment free quick to calculate and easy to compare. And our initial case study shows promise of the method compared to others in the field and in considering comfomers and as a tool for virtual screening. So the next steps for the project. And we do have a second method of describing shape in the pipeline and that doesn't depend as heavily on the choice of starting position in which we think might address some of the issues that we've had with that. And I then have to complete a full validation study using the directory of useful decoys and to compare our approach to shape existing shape similarity methods and validate its performance. And hopefully if this is successful will then apply the project in real drug discovery projects working with the Newcastle Center for cancer. And there's also quite a bit of scope to expand the method to include consideration of form for properties as well as shape in order to further improve the quality of the suggestions that we get out of a virtual screen. And so to finish, I would just like to thank my supervisors Danny Stewart and Matt for all of their help and support with the project. And we'd also like to thank Dr. Thomas Murphy CS Fullerton for his input on the mathematics side of the project. And I would like to thank the engineering and physical sciences Research Council for payment and the Alan Turing Institute and the molecular science for medicine CDT for all of their help and support throughout my PhD. And then the QR code here at the end contains the link to the GitHub repository and the paper covering the method and my contact information as well. And thank you for listening. Thank you very much. Now we go we move on to our last speaker. Serena please. Good afternoon everybody. So let me directly introduce the main character of my research, which is this human cement receptor. Sorry. Yeah. Oops. So, this is my receptor is a terrorist in his receptor that is placed on the plasma membrane of tissue specific capital cells and in these cells it regulates cell growth, migration, migration sorry replication, and as well as cell survival. When they regulate and instead this receptor leads to metastasis metastasis development and as in the case of my project, it can mediate the internalization of bacteria such as the list area monocytogenist. And this bacterium is an intracellular bacterium that exploits the internal in be so an invasion protein that is known to bind to the cement receptor and ultimately it leads to the internalization of the bacteria. So what we don't know is how this happens so my big game or better the beginning of my project is that one is 10 the activation mechanism by these invasion product in the internal in be our hypothesis is that once bound to the cement receptor. This is forced to assume active confirmation that makes it more favorable for dimers to form and once the diamonds have formed, which are almost dimers. Then, at the level of the cytosolic domains where the time design kinesis are placed the signaling cascade can be triggered and the internalization can happen. In order to go through my results I need to give you some more details on the topology of the receptor. And in particular, what is important here is that for my research for the results that I'm going to show you and explain you the, I have focused on the upper portion of the domain of the receptor, as in these actually divided in three regions, which are the actual domain that I don't get outside the cell, the plasma and so the transmembrane region that anchors the receptor to the plasma membrane and finally the cytosolic domain that contains the juxtamembrane domain and the time design kinesis. And for my research, the main point, the important part is composed by these three domains, the sema domain that you see in atomistic representation on the side in yellow, then the PSI domain and finally the first sema domain. Why this? Because these three domains are the only ones involved in the binding with the internal. In particular, this happens via two interfaces, one that is placed between the inter-repeat region and the sema domain. And one that is placed between the leucine-rich-repeat region, so the concave side of the internal and the first immunoglobulin-like domain. But something here is missing, let's say in the scheme, which is a quite important detail, that is the fact that this receptor is not just a protein, but is a glycoprotein, meaning that on the ectodomane and on the juxtamembrane domain in particular, these receptors have 13 anglycosylation sites spread around these two regions. And in the portion that is important for irrelevant for this talk, we have eight anglycosylation sites, so that if we were to give a closer look to the receptor, it would look like this. So we have a green, let's say, chains that are the glycans, so that in our goals, or among our goals, we have to understand what is the role of the glycans. Additionally, we want also to understand what is the active conformations assumed by the receptor upon binding with the internal in B. And finally, what is, and if it's possible to discriminate between active conformations. So in order to do so, we decided to use atomistic molecular dynamic simulations. And our idea was to compare the trajectories explored, let's say, assumed by the receptor in four different situations. One is the naked receptor in isolation, then naked receptor but in complex with internally, and finally, the same two situations but in presence of full glycosylation. So, our approach was that of considering all the configurations assumed by the receptor so the red portion in each of these models, and then put all of these configurations together in one data set and applying on the data set a non linear dimensionality action in order to obtain a representation and lower dimensional representation of the explore configuration of space. Sorry again. Okay, this is what we have obtained so two dimensional representation of the configuration of space, and on this I then applied a clustering technique with which I identify the three high density clusters which represent the conformations explored by the receptor. And the good part of the advantage of this approach is that for each of these points in the new representation we are able to map back to the original configuration which allows us to, or enables us to identify for each of the configurations, which are the models that explore that confirmation, and these are important in this, in this table here. Additionally, we're also able to identify the representative configurations for each of the conformations so to gather them between and open conformations. And finally, or better lastly, we are able also from this representation to understand what are the global motions that lead one, let's say the receptor to pass from one confirmation to another by building the pseudo trajectories along the coordinates that span these two dimensional representation of the configuration of space. And out of these analysis, I have two comments. The first one concerns the glycans role. In particular, if you look at the table, you will notice that the naked receptor in isolation so the one of which the name is in red explored all of the conformations. Why the glycosylated receptor in isolation only explored the purple configuration, the conformation, sorry, which means that the glycans are actually reducing strongly the configuration of space explored by the isolated receptor. And as we will see later, this is definitely not trivial. Then, we have the second consideration instead concerns the representatives of the conformations in particular. The fact that the conformation assumed by the complexes, which is the one in CN in the representation, is actually the only one open, open in what sense that based on the angles spent by the axis. Let's say trespassing the semi domain from top to bottom and the immuno globally like domain from one side to the other as it was a cylinder. So looking at this angle, we are able actually to discriminate between active and inactive conformations and we call this angle data angle. And remarkably, looking at comparing our findings for the two complexes distributions of this angle. We have across three microseconds trajectory for each of the two constructions, we noticed that the angle assumed by the two complexes so the distribution of these angles for the two complexes as is actually overlapping with the angle observed or assumed by the native auto dimer so the dimer formed by the receptor in presence of the endogenous legion which is the epatocyte growth factor. This is remarkable in the sense that this might mean suggests that the active conformation assumed by the receptor is actually in the independent. So on the other side, always looking at the triangle, we can see that if we consider now the distributions of the triangle for the isolated structures. We notice that while the naked receptor in isolation so the orange distribution overlaps with the active conformation. So in the case of the glycosylated structure of the receptor in isolation this never happens, which means that the receptor and the sorry that the glycans are actually selecting between active and I will suggest that the glycans might select between active and inactive conformations, and these is related to the fact that during these trajectories, the glycans form these bridges among them with a fuzzy fashion so these bridges are not steady but they instead are alternatively formed in this form, and this is particularly non trivial because they do not represent an aesthetic impediment to the motion of the receptor but they rather via these fuzzy interactions seem to keep the receptor more compact and also impede today more globally like domain to freely orientating space, for instance, towards this direction when the receptors isolated. To come to, let's say summarize, we have found that the glycans seems to select between active inactive conformation or allow this activity, then, and that the angle theta is actually an optimal discriminant for active inactive conformations for the upper region of the after domain. And finally that the active conformation, it can be identified with a theta of around 110 degrees. And now for the outlook, so for the research that I'm now carrying out. I want to start like I want to describe it to you starting from this conformation configuration sorry that I observed for the, for the isolated receptor with fully glycosylated. So the glycans seem to stretch on the top or actually are stretching from on the top of the semi domain, but then if you imagine that these receptors actually connected to his portion is connected to the rest of the chain. And this chain is then connected to the membrane. And it seems that these glycans actually stretching toward the surface of the membrane so that they might interact in similarly, as they do between themselves, for instance with ganglion sites, there are glycans, like a single lipids. So they have this glycan head that has a composition that is similar to one of the glycans termini. And they might interact with the same fashion so and also in this way influence the dynamics of the wall economy of the receptor. At the same time, looking at also other players of the plasma membrane as for receptors we notice that these conformation of the order is a structure of the, yeah, this is this conformation of the receptor chain. It seems to have a strong similarity with the, with the conformation assumed the inactive conformation assumed by the integrins that are in particularly alpha integrins that has that have also a strong structure affinity with the receptor. So my research is going towards this direction and I want to investigate. So the role of the glycans, the participation of the glycans in the latter organization of the receptor, their effect on the dynamics of the wall to domain. And finally also to understand to identify the dimeric conformation of the homodimer formed by the CMET internal and B complexes. And I'm going to do this exploiting optimistic and the simulations, although I'm also exploiting the course grain simulations for this task in particular so that the identification of the dimer. And we're going also to integrate our data and we are actually already did partly this with Fred data coming from our experimental collaborators from my column and screw. And that is all. Thank you all. And I want to thank all the people of my group and friends also and all the institutions there on one side. Fund me and provide me the computational sources, the resources. Thank you all. Elena. Thank you all the speaker. And now we might want you participant to ask question. So please use. So if you can use the bottom on the bottom you find a Q&A function. It can be this or this according to your operating system. And you can start to type your question and Martha that is sharing the section we read for the speaker. So I invite people to do that maybe in a while we are speaking and waiting that people start to type. I can I have a couple of, I have a couple of questions. I will start with Rachel. I was wondering, do you know, I have two question one, how computationally, which have computational advantage we have to use your approach against other shape based approach. The advantages. How much do you have on the computational cost. It will go down. Yeah, it will be expensive or it will be go down how much do you have an estimation. And they're relatively quick. So I think compared to the other two that I had said, the atomic distance based ones are quicker, but they're not quite as good at producing good hits. They're comparable times and maybe just a second or so more and they're quite a bit faster than the ones that are more accurate and sits somewhere in the middle of the two. So roughly for one compound, what is the time that you expect. Depending on the size of the molecule two or three seconds. Okay, this is pretty standard desktop PC as well. And, okay, accessible to what lab researchers that might not have access to high performance computing. Okay, and you spoke about a lot of the advantage of your approach, do you see any limitation. Yes, so the main one is that we can't look at motorcycles and anything with a hole in the middle of it is off limits. And this dependency on which atom do we start from and might lead to some limitations as well. So because the description is better around the point that we start from. If you get that point in the wrong place you might discount things as not being similar when they actually are. Okay, okay, thank you very much I see popping up question please not up. Yeah, Shravan wants to ask my voice, so go ahead. Yeah. So it was nice presentations and I have question for Sharon, that in the last talk Serena mentioned that she's going to use the Fred, Productions that experiment to validate her study. So how civil going to validate that study using the Fred and comparison with the molecular dynamics. So we are going to we are actually using the threat to identify distances between some residues of in particular they internally be proteins, and use these ones to guide our research of the monomer form. So this is a sort of dimer alignment, because we are the say triangulating the, the position of the two internally be with respect to each other. And these known how the internal be binds to the receptor will lead us to understand the, which dimeric confirmations we get from the course grain simulations that I mentioned will be accepted actually fitting this data. Okay, so that's good. So now, I mean, which techniques I mean intrinsically fluorescence or extrinsically are you going to use any dice to label that your residues and then you are going to measure the fluorescence of Fred. So is it intrinsically or extrinsically. And the threat experiments are done with dice there. So they have mutated the residues on the internal in the protein and are doing the, the experiments. I would say, extremely, if I understood what you meant, we are not doing this by simulations they are actually doing experiments and then we use their data for the say checking our the dimeric confirmations that we observe from the simulations. Okay, thank you. Thank you. So we have another question for Rachel. They say thank you for all the nice talks. And the question is, do you have some examples of the other properties you would like to analyze in the future besides shape similarity. And so it's actually something that a lot of the other shape similarity methods have done as well. And the standard thing to look at is things like kind of hydrogen bond donors hydrogen bond acceptors, number of aromatic rings, typical things that would potentially influence binding. And or you can look at putting the electrostatic potential on top of the shape similarity as well. Thank you, Rachel. So people can ask more questions I also allow you to raise hands if someone wants to talk to Stravon, you can raise your hand and we'll allow you to talk. I have a question for Costanza Costanza you explained very nicely you know like the different steps of how one subunit goes and then the other one and then they get better, but you don't have to say the full four point yet. So, can I just ask you how do you envision it what do you think will be like the next steps without you, you know, working hypothesis. It's a nice question because when you lack experimental data and you cannot say if what you're seeing on your simulation is reliable or not but we have the actually we have the poor crystal structure of the entire complex and so the prepor so we have some informations about how the heterodimer can can form from this but not the preliminary stages so not about the formation of the so we suppose that that exposed interface could be maybe a good interface also for demarization, but of course we we also have to try different possible conformation so this is the probably the most difficult point to face. Okay. Thank you. I have a couple more questions for Serena. How do you think this could be interesting when talking to experts in clinical settings. Can you envision a way, this could be relevant for the treatment prevention of listeriosis. I think that the, that our research will, let's say, the aim is to understand the activation mechanism, especially for what concerns the actor domain and normally grow factor receptors are all targeted with any leaders that there is any kinase that is inside the cell instead. So I think it would be particularly interesting to develop therapies so from the point of view of the clinical setting that he did instead the activation of the actor domain. So there is the say any beating the Tarzan kinase is not always optimal in the sense of Tarzan kinase is like not only the grow factor receptor has the epithelial factor receptors so the cement has a Tarzan kinase so it's hard to be selective is that the active domain specific for receptors. So it should be more selective therapy in the sense. You raise your hand. Go ahead. Yeah, yeah, I have a question for Rachel that she has used the new set similarity approaches to find a based molecule for the protein. So the question is, did she use any experimental techniques to validate her results. And we would hope eventually when we go on to do sort of real drug projects we would then pass the results that we get back to the wet lab tests. And at the moment it's just a kind of initial case study of with this in theory work. And so we've not done any experimental validation yet. Okay, thank you. So we have another question for Costanza. Are there experimental data on the interactions between the proteins on the membrane. Yes, there are interact experimental data about that. For example, when I talked about the fact that the Thomas the postdoc made around this simulation for the single monomer that bind the membrane. And he could compare his result because he used different kind of membranes with the experiments that were also held here in Trento and with some using some basic calls. So model membranes, basically also with some flourishing calcane inside. And so the, we can say the action of the memo is in was. So the citrolytic activity was measured with this kind of flourishing experiments because the gamma molyzine makes pores on the membrane and so its activity was measured by measuring the flourishings, we can say. So for the specific interaction between the proteins and the membrane so they were found have been found some, we can say pivotal residues. On the ring domain of the UKF. And also with the simulations Thomas found other two binding sites so one it's we can say in line with the experimental one and the other one wasn't. It's still detected. Okay, but this is not an experiment, we can say that he, he was able to see a bit more. Thank you Costanza. Another question for you, have you compared the homologizing structure with other homologizing, maybe to see any subtle differences that can have an impact. No, I haven't personally. The fact is now I would be interested in a heterodimer confirmations. I know that there is a kind of gamma molyzine, which actually forms a heterodimers in solutions and not single dimers that and then this heterodimers can find. It could be useful to compare to use this kind of structure, even though, of course, maybe also to to see what. Yeah, probably it would be useful and I will do this, even though this kind of heterodimers, if it can be forming solution. It has some kind of differences of course. But maybe to see which are those differences could be, could be useful. Thank you. You have your hands raised again. Yeah, so I have question for Costanza that sees investigating the four formations and using this heterodimer. So I don't know much about it but after finding that how the pool is formed and the civil investigate, what is the target how I mean we can use to target the toxins, what is the applications of that by doing this all structural investigations and the fact is that these toxins are released mainly by bacteria during bacteria pathogenesis so the main, the main we can say that aim would be to prevent these toxins to the binding for example if we know the, which are the the binding sites of the protein, or also watch what can causes the conformational changes so that the major aim would be to to prevent the formation of toxins, the poor. So, maybe eventually research like these could lead to the development of some, some drugs. So, if we can find a proper target. Okay, thank you. Thank you. I think I'm gonna leave it for one more question. Rachel, I have a question for you so you talked about you will benchmark your tool can you tell us a little bit more about that so how will you do it against some other tools or just against the best tool or yeah. So, it'll be against the same methods that I presented for the little table and of just using three PD five inhibitors with the exception of the chemical fingerprint method. And so the gold standard benchmarking set is the same one that's used with a lot of docking the directory useful decoys that's 102 different protein targets with a set of active and presumed inactives for each target. And we'll run through those and see how well the method does identifying true actives compared to the decoys. You can't use that for the 2D fingerprints because there's an inherent 2D bias in the data. So it's too easy or put them. Thank you very much. I think we'll leave it here today. Thank you very much to all the speakers and thank you very much to all that and this for having been with us today. See you another time. Bye.