 Pričo, da se počutimo na antibodijske desine. V tem spremu smo se vseznačno vseznačno, in dokončno vseznačno, na antibodijske desine. Imam Alessandra Vila, imam v Stokom, v rojstvu tehnologiji, in to, če sem izgleda, je vseznačno vseznačno s Susana Zaldova in Aleksandar Bruban na universitetu Vutrektu, na Naderlanske. Why antibody design? In in the last period there is a growing interest for the pharmaceutical company in antibody as a drug. Indeed, already antibody has used in therapeutics to cure some particular diseases, like remotoning, arthritis, multi-sclerosis, multiple other type of cancer, like prostate cancer. They are used also as a detector in diagnostic, and, for example, in the pregnancy test, they are used already in the pregnancy test. Antibodies are proteins that bind to a specific protein, called antigen, that is an oxide of a pathogenic, of a virus, of a bacteria. And this will treat an immune response, so L will tell the immune system to destroy the site. So it's very important that antibody has very high specificity. So usually antibody is an epsilon shape, or we can think about like a protein form of three arms, and at the end of the arm we found this variable region and where we have the hypervariable loop. And those loops are the ones that contribute to the high specificity to the antigen. The other part of the region, the fragment antigen binding, that region is male in constant and is not specific of different antibody. Another peculiarity of antibody is the low immunogenesis. So which is the challenge in design antibody? One we would like to predict the antibody three dimensional structure, only knowing the antibody structure we can know also, we are able to model the antibody binding mode to the antigen. Of course once we have a structure and the binding mode we would also like to improve this binding mode, to improve the binding affinity. The specific aim of the work that I will present is just to start just to willing to improve the docking protocol for antibody, a standard protocol if that you take the unbound antibody, the antibody structure, the antigen structure, usually from experimental source and you plug in in a docking protocol in standard one of adog to get a model of the complex. We would like to improve, so that means to have a better model, a model with high quality, that is what means a better model. But of course we want to achieve this, still thinking about a balance between accuracy to have something very accurate, but in the same times that it will not take too long, otherwise it's always a problem if we have a very accurate model approach, and it takes too long to get the results, we always get the results too late. And also we would like something that is easy to implement, so that we can implement in a workflow that can be distributed to people to use. The approach that we thought about, the following is to combine a docking technique and this is where we perform with adog and molecular dynamics technique, perform with chromosomes. We need a test, a pool of protein system, complex, body antigen complex, and this pool of complex, we took 11 complex from docking batch mark, affinity batch mark, and one requirement of those structure is the availability of experimental structure, both of the apofoam and the olofoam. We will use the apoform of the antibody and antigen as input, while we will use the olofoam as a reference to test our model, the model that we will build. In this year we have looked different type, we have tried different type of workflow, first creating a model, then simulate it, but it was not working so good. Then we also tried to simulate both antibody antigen and then combine it to get out of structure, but also in this way we were not so happy with the results. If you are curious on one detail, on those protocols you can just ask in the Q&A section. The fanical workflow that provides us a better model than the original one is the one that I will present now. So we start from the unbound antibody and unbound antigen, then we will perform a simulation, we will perform a molecular dynamic simulation of a standard nanoseconds in a NVT ensemble, in a standard condition, only of the antibody. Of course here we can meet and see some issue, for example in the original structure might be missing, there might be some loop missing. We know that we are running a simulation with a fixed protonation state, so the protonation, so all what we simulate is according to the protonation states that we have, so if we have a change in the interface, auto-unbounding that we cannot mimic, we might be also the present, if binding is promoted by the present ion and interface, that also will be not included in this approach. Then when after the modeling we get a pool of structure and the question is, how do we analyze all of them with Hadoq? We have to have a clever way to select a pool, a strict pool of structure that we can plug in in the docking program. And so we decide to look to two different cluster approach. We use the Arctic Patrick approach and the Gromov approach. Here I briefly describe in this slide I will not go in details the difference between the two approach and the Gromov approach provides us better results both in terms of each rate and success rate. Again if you are curious of the comparison between these two, please ask in the Q&A section. So what we have done this fitting, first we fit with this clustering, first we fit the backbone, then we select the hyper-variable loop, then we apply the cluster analysis only on the hyper-variable loops and we extract the centroid of the cluster of the most 20 most populated cluster. So now we have this pool of 20 structure for each of our antibody. We add to this pool the original crystallographic structure and we plug in those antibody structure together with the original crystallographic structure from the antigen in ad hoc. So we perform a standard docking starting with the rigid body energy minimization. We will take as a result 100 times the number of structure of antibody structure that we have plugged in. We proceed with simulating annealing and here we start 400 structure. We go on with refinement solvent where we get 400 final structure that we will analyze. To ases the quality of the model that we got, we have two different approaches. The one approach is one parameter is docking and the other parameter is ad hoc score. So the ideal scenario will be when decreasing the value of ad hoc score, so decreasing that is based only on energy, it's based on energy contribution, we will have an increase of docking. Indeed docking is vary from zero to one where one is when the complex is identical to the experimental structure where the model is identical to the experimental structure. As you can see how is defined docking is all defined with some structural criteria based on native contas and Rosmy square deviation of the interface area and the ligand position. So will allow us also to divide the structure between the model obtained between incorrect up to higher level. So here we look our results. On the left side we have the results with the standard docking protocol starting from the two x-ray structure. We can see that we have most of the model that we obtain are accurate. There is not really clear trend between ad hoc score and doku and then here are the results with our new protocol. So we can see that indeed we have some of the expected trend between the ad hoc score decreasing the ad hoc score we have an increase of the value of doku. We have a higher population of medium type of model and also the peak of the distribution of the model is moving through a medium model. So we have an improvement in the quality, clear improvement according to these two criteria in the model scoring. There is another way to evaluate the quality of what we got and is the success rate of the each ray. The success rate will give us the percentage of complex that has a structure of high, medium, acceptable quality in the first structure in the top one structure top five, top ten. While the success rate, the hit rate will provide us the percentage of the complex that has a high, medium, acceptable value from the old structure. We can see that while performing all the doki we have nothing while we perform the doki in plus MD we see that we have an improvement. It clear we have more high quality structure appearing that we were not appearing before and also on the top five we have an increase of the number of medium quality structure. Of course that is an accounting for all the eleven antibody antigen complex that we got. We would like to see more in details what happens for... So that doesn't mean that all the system are the best. If you are curious we have an overview that I can show you later. And so we have, for example, one case that has three v sets that is as not performing so good. Indeed, as you can see there is not much difference applying the two type of approach the doki and MD doki. And this is also can be expected since if we compare in GRACE we have the original X-ray structure of the complex in MAGENTA and in RANGEN you can see the loop, the position of the loop in the X-ray structure of the antibody. You can see that there is quite a conformation of a range between the bound states and bound states. So that means that if we meet such conformation states our protocol is not able to provide us better approach. So then we change our protocol we decide to do a new protocol where we use an accelerated weight histogram to try to announce the sampling of the antibody structure. What we do we combine the unbound antibody with the unbound antigen X-ray structure in an ad hoc so we do a pre docking we extract the best model and this best model we will simulate using accelerated weight histogram also here we perform 100 nanosecond computational we are on the same level and how we perform such as now we have to define a reaction coordinate and the reaction coordinate we will be the distance between the two center of mass of the two proteins and the method is somehow having a bias potential to make the effective potential flat and that will allow us a better sampling of the hyper variable loop in present of the antigen. Then we proceed we take only the antibody structure that has been sampled in present of the antigen and we remove the antigen and we apply again our clustering and we extract 20 structure we add the X-ray structure and we perform the protocol like before and we perform we run ad hoc in the same way we run before and we look to the results so the first column is the docking the second column is docking combined with standard molecular dynamic simulation and the last one it's when we combine with accelerated weight histogram approach used to generate a conformation for the antibody and we can see that we have a clear improvement not only in the graph where we show ad hoc score against docking where we have also an increase we have a shift of the peak of the structure between a settable toward medium quality structure and if we go to look on the it rate we also see that we have higher medium structure 100% medium structure in the top one where we before we never had a settable, a medium quality structure so that as now we have to refine this protocol to stand of course on more case but it's very promising so the conclusion is that we we trade this work is that it's important to focus on the sampling of the apple form of the antibody that is the main message so probably because the antigen is more rigid molecule so and we need to have to improve the model this is a key to have a better sampling of the impervariable loop now we are working to put all this workflow now we have this workflow we are trying to import in the Jupiter notebook some way that is ready to use from everybody we will go on to generalize the approach where we use an accelerated weight histogram and we will aim to apply to ex-novo design of antibody where this there are some aspects that are still missing and as one has to think about and that are important is the force field effect we have run only with one force field all the results were running only with Charm 36 as a force field we might want to consider the option to perform binding affinity calculation so to extract free energy difference and to try to improve the affinity and we would like also to consider we know that all this approach since all the sampling in one case somehow the sampling in the first protocol the protonation states of the of the antibody is the protonation states at the interface in the second case is still if we have protonation of Korean binding level or ion player role on the binding interface or how to address incomplete experimental structure so these are still open question that we might want to investigate further so I thank you for your attention and I hope I would like to see you in a Q&A section with a lot of questions on this and if in the future you have any question about Gromax orado please go to our forum and ask and use the forum thank you, bye