Webinar Grupos científicos de la SEBBM: “Grupo biología sintética y biotecnología molecular"

[00:00:00]Hello.

[00:00:02]Good afternoon. Welcome to one of our webinars.

[00:00:06]Today we have Ilenia Jabalera. She's the coordinator of the group Biología Sintética y Biotecnología Molecular. She's a researcher in CIC bioGUNE and she has a an important uh researcher as an invited.

[00:00:27]So uh Ilenia, if you want to introduce our Reyes.

[00:00:31]So >> Uh it's my pleasure to to welcome our speaker, uh Reyes Monje-Franco.

[00:00:38]Reyes is a postdoctoral researcher in the Computational Chemistry Group, also in CIC bioGUNE. This group is led by Dr. Gonzalo Jiménez Rosés.

[00:00:48]Uh her research is focused on the application of artificial intelligence and molecular modeling to protein design with special emphasis on stability optimization and function improvement.

[00:00:59]Uh she completed her PhD in 2024 at the University of the Basque Country where she explored protein-ligand interaction, particularly the molecular recognition of ligand using a wide range of computational tools.

[00:01:13]Her current work includes the engineering of thermostable protecting bio with potential therapeutic application as well as the use of deep learning deep learning tools to enhance the solubility and catalytic efficiency of the protease. I would like to highlight that this work was a collaboration with uh Professor David Baker who won the Nobel Prize in 2024.

[00:01:38]Currently, her work focus on the novel design of protein for a variety of application ranging from biotechnological oriented enzyme to mini protein binders with therapeutic potential.

[00:01:49]So Reyes, the floor is yours.

[00:01:52]>> Thank you very much, Ilenia and Patricia and the Society of Biochemistry and Molecular Biology for this invitation.

[00:02:00]And yeah, thank you for having me here today. I'm very happy. So, I let me share my screen 1 second.

[00:02:16]I hope you can see properly the slides.

[00:02:19]Right. So, yes, I Ilenia just said um I was a postdoctoral researcher at the Computational Chemistry Lab led by Gonzalo Jiménez Osés.

[00:02:30]And today I would like to share with you some of the uh research projects we've been working on in in line with the topic of AI-driven protein design and optimization.

[00:02:42]So, uh when Ilenia invited me to this webinar, I was thinking a lot how to orient it because I know the audience is very uh wide and diverse in this group. So, I decided let's hope it's okay to do uh first a brief introduction uh regarding the protein design area, then share with you which is, let's say, the uh standard protocol or the general and common steps we follow to apply these tools, and then share with you three real case examples in which we have applied this technology for very different purposes in each of them.

[00:03:22]So, let's jump into the protein design.

[00:03:26]So, this field is normally divided in these two main areas. So, we have protein engineering, and what we mean uh with that is that uh we have an already existing protein uh that already exists in nature, and we want to modify it for improve uh some of their properties. So, this can be in order uh to improve stability, solubility, expression yields, or enzymatic activity, for instance, when we are working with enzyme.

[00:03:56]And then, on the other hand of the picture, we have de novo design, and what we mean by that is creating entirely new proteins from a scratch.

[00:04:07]And this, as you can imagine, can have a lot of many different applications, ranging from uh designing new protein binders, nanobodies, biosensors, enzymes, or even new materials.

[00:04:20]So, even we have these two main lines, the both of them they share general uh steps that we might follow for doing the design protocol. So, of course, the first step would be to identify our targets of interest. I will focus first on the protein engineering line, okay?

[00:04:41]So, we identify our protein of interest, and it can be the case that we have already a structure, an experimentally structure for that protein, but sometimes, we all know that we don't have it. So, if that is the case, the first step, of course, will be to predict the structure of this protein.

[00:05:01]And once we have it, or if we already have it, our next step will be to design the sequences to introduce these mutations that we are interested to confer on a specific property.

[00:05:13]Once we have the sequences, and normally in this step, nowadays, we are working with uh 8,000 of sequences, we do the structure prediction for the most promising ones, and then we have to compare this predicted structure with the original structure of our reference, let's say our wild type. And this is something that in the area is known as a self-consistency criteria, to make sure that the sequences we are designing remain folded the same way as our our original target.

[00:05:46]Then, uh applying different criterias, we select our best candidates and we go to the wet lab to do the experimental validation and characterization.

[00:05:56]So, we are normally working, as I just said, in the range of uh thousands of sequencings and then, after filtering, we end up trying in the lab, let's say, dozens of uh designs.

[00:06:10]So, this would be for the general or standard protocol for the protein engineering line.

[00:06:15]And when we are, you will see later on on the examples, when we are working in uh designing the novel proteins, the beginning is a little bit different, right? Because we have first to create these new folds. We will see later on how do we do that. And once we have different folds that we like, we join into the third step, design the sequence, again predict the structure, uh select designs and validate in the lab.

[00:06:43]So, that would be the general picture.

[00:06:46]And now, uh I selected three different projects in which we apply some of these tools for, as I said at the beginning, very different purposes. So, we have uh the first one that I will present now uh with to design proteins for being used for protein replacement therapy, but also we have some example for develop biolo- biotechnological tools or for being used in precision immunothera- So, the first one would be the thermodynamic stabilization of human frataxin.

[00:07:20]Um but, why frataxin? Well, this protein is involved in an autosomal recessive disorder, Friedreich's ataxia. And this disease is characterized by a systematic failure in the mitochondrial function and this have uh different effects in different organs.

[00:07:39]So, as it core as it core, uh the patient suffer from a critical deficiency of functional frataxin. So, less amount of functional protein.

[00:07:50]While the most common cause of this disease is the transcriptional repression of the frataxin gene, we also see a specific mis- missense mutation such as these two shown in the screen.

[00:08:03]These mutations are particularly interesting from us because they lead to a um loss of a stable frataxin and therefore we have less functional protein that is available.

[00:08:15]But why frataxin is so important? Why do we care so much about this protein?

[00:08:19]Well, this protein it acts as an iron storage protein within the mitochondria and beyond that it also serve as a key allosteric regulator in this iron-sulfur cluster assemblies. In particular, frataxin protein would be this red small protein within the cluster.

[00:08:39]Currently, there is no approved cure for this disease. However, the field is shifting toward protein um therapy and um yeah, and replacement therapy.

[00:08:51]Um And this is because it has been seen that even a modest increase in this functional frataxin protein can reduce or even mitigate some of the associated symptoms. So, there has been lot of efforts into this line. In fact, there is already this construct that is in phase um in phase two uh trials. So, our goal here was to engineer human frataxin to increase the stability of both the wild type and the pathologic mutants that I uh just introduced to you.

[00:09:28]So, the protocol we followed for this was following the steps that I showed you at the beginning, but before going into the uh sequence design the sequences, I missed one step. First, we have to identify which are the hotspots, the positions we are interested to mutate because we don't want to change the entire sequence of the protein, right?

[00:09:51]So, for do that, we use both evolutionary and geometric information. You will see now. We define a series of hotspots and then we do sequence designs into these positions.

[00:10:04]And we did the sequence design by using a consensus approach, but also using protein MPNN.

[00:10:11]Protein MPNN is a deep learning tool that solves the inverse folding problem.

[00:10:17]So, given a protein structure, it will design the best sequence that will fold into that 3D structure.

[00:10:27]So, once we have obtained the sequences, we predict the structure with them and we evaluate their stability because it was the main goal of this project. And for that, we use a method developed in the lab that is called multi-alphafold that combines alphafold structure prediction and Rosetta energetic evaluation. And you will see later on that is very good in predicting stability. So, if you are working in this field or in any field related, I invite you to give it a try.

[00:10:59]And then based on the results of after applying this method, we select our best candidate and we go for the experimental validation.

[00:11:08]So, in this case, we started from a multiple sequence alignment of 2,000 sequences, but we reduce this in three different groups. First, those containing sequences with 60% of identity or more to our ferredoxin wild type, then 17, and then increasing until 80.

[00:11:27]So, in this way, you can see that here in blue or in black depending on your screen, you see the number of hotspots we have for each series. So, we are increasing these mutational hotspots in a in a very controllable manner.

[00:11:45]So, we did the sequence design for those hotspots, the selection of our best candidates. We went to the Lava Express, purified them, and measured the melting temperature.

[00:11:57]That will be our proxy for thermostability.

[00:12:01]And you can see here the result for the 10 tested designs. And we were very happy to see that we were able to obtain, for instance, in the variant number 10, an increase in the melting temperature of more than 20° in comparison with the wild type.

[00:12:17]And very importantly, all these 10 designs, they were done in a single design step and with a computational cost of less than an hour, so in a very efficient manner.

[00:12:29]So, once we thermostabilized frataxin wild type, our next goal was trying to rescue rescue the effect of the pathogen pathogenic mutations.

[00:12:43]So, here you have the effect they cause in the melting temperature, the bottom mutation at position 154 and 198.

[00:12:52]And we did exactly the same as before.

[00:12:55]We followed two different strategies. So, first we transferred the mutations we already knew that they stabilized the wild type, but we also resampled the same positions but starting from the pathologic mutants.

[00:13:09]And we were very excited to see that we were not even not only able to rescue this effect, but even increasing melting temperature in comparison again to wild type.

[00:13:22]Then we wanted to make sure that the mutations we were introducing were not affecting the fold of the protein. So, for that we use circular dichroism. Here in the screen, you can see the spectra for both the wild type and our base candidate frataxin 10. You can see that at 25° the shape of the spectra are the same. And very interestingly, you see that at even at 75° for frataxin variant 10, the fold remains almost intact, whereas for the wild type the structure was completely lost.

[00:13:59]So, yeah, if we superimpose both spectra, they maintain the same shape. So, we can confirm that these mutations we incorporate in the design, they were not altering the the native fold of the structure.

[00:14:14]Here there are some of the mutations we have introduced in these designs. You can find many changes from hydrophobic to charged or polar residues at the protein surface. Also, this mutation to proline is a very prevalent mutations that we found in different designs. And what it was very interesting was to see that this the most promising variant where it in it it appears this new network of salt bridges that was absent in in the wild type.

[00:14:50]We also evaluate the thermostability curve for the most promising variants and we found out that a frataxin 8 and 10 increased thermodynamic stabilization of approximately 3 kcal per mole.

[00:15:04]And here in the middle of the screen, we have experiment I really like because it show us that we were not only improving stability in terms of temperature, but also in terms of proteolytic degradation. So, resistance to proteolytic degradation. In this plot, you have the result for a trypsin digestion at different times. So, in green, you see the results for wild-type frataxin and in blue for our variant, design variant, and you see that after 1 hour, the number of peptides after digestion that appear for the variants are almost the same that at the beginning of the digestion, whereas for the wild-type, the protein is almost completely digested.

[00:15:46]Um uh finally, thanks to the help of our collaborator, the Precision Medicine and Metabolism Lab here at Bio Wageningen, we could confirm by uh CSP NMR experiments that this new variant maintained the ability to bind metal ions and interact in the iron-sulfur cluster assembly, which is vital for its function.

[00:16:07]So, uh to sum up, I just showed you at the beginning this pipeline we followed for the design. Very importantly, we used the criteria uh of uh this energetic structural and energetic evaluation using uh MultiAlphaFold mean to select the candidates we went to test in the wet lab. And after measuring the melting temperature for all of them, we we obtained a correlation of 0.86, which is quite high and very remarkable in a range of almost 40°. So, in a very wide temperature range.

[00:16:43]So, we were very excited, actually, with these results. And now, we are very happy to see that uh in fact, the project continue evolving to a more clinic point of view.

[00:16:56]Um so, this project has been recently granted with a proof of concept in collaboration with Macarena Sanchez uh Navarro. They have um they been working in developing peptide shuttles to help frataxin cross the blood-brain barrier.

[00:17:11]So, now what we are doing is joining forces together, and we are combining their peptides and our stable variants.

[00:17:18]Um so, so we've been already we designed new constructs that they have been tested in patient cells and in mouse model with already very promising results in terms of the ability of functional protein also with the help of Elisa Gabiskol. And in particular, Alejandro Benitez has been in charge of running all these experiments.

[00:17:42]So that would be the first example I wanted to share. Now I will go for a uh the optimization of TEV protease which is in term of methodology kind of similar but the objective or the application is very different. So in this case we are interested in improving a biotechnological And as Ilenia said at the beginning, this is the result of my stay at the IPD during my PhD.

[00:18:10]So probably I don't know I don't need to introduce TEV protease within this audience but very briefly, this is a very specific protease enzyme that is a valuable tool for protein engineering and recombinant protein purification.

[00:18:26]Probably most of you have already used it in the lab but it has some drawbacks or suboptimal properties let's say. So it has low expression levels, low thermostability and kind of poor catalytic activity. So the idea here was to use these techniques to improve the expression levels and if possible the catalytic properties of this TEV protease.

[00:18:51]So we follow a similar strategy. We define different sets of hotspots and then we use protein engineering for design the sequences and we end up testing in the lab 144 designs.

[00:19:05]And the first objective was to improve the expression yields.

[00:19:10]So I don't know if we you are able to see, but this dashed line would be the value of the yield for our starting point, this TEV protease, and here you have an Instagram distribution of the yield for all the designs. So, almost 130 of them were outperforming the the parent variant of two. That was great.

[00:19:33]But, of course, this is an enzyme, so we have to make sure that, apart from improving expression yields, we retain the activity, cuz otherwise has no sense.

[00:19:45]So, to assess this catalytic activity of the designs, we used a previously described assay based on the reaction with a coumarin derivative. So, after the purification, the these designs were exposed to this peptide coumarin, and we followed the fluorescent signal. And what we observed was that 64 out of the total displayed fluorescent growth exceeding the background, so indicating some degree of activity, but we selected the three most candidate most promising candidates to characterize in in more detail.

[00:20:23]So, we measured the catalytic parameters for them, and we observed that they were displaying higher kcat values, lower km values, and then, of course, higher catalytic efficiency. In particular, the design labeled as 60 showed a 26-fold enhanced in in catalytic efficiency than our TEV wild type.

[00:20:48]We also evaluate the thermostability of these designs. You can see that our starting point was around 40°, and the new variant 80°, so this very very stable. And also, something I find very practical is that what we did was to incubate both the wild type and the most promising design at 30° at different interval of times, and then we repeat the activity assay. And what we found out it was that after 4 hours, the design remains almost the same activity, so 90% of the original activity, whereas the reference variant only 15. This means that if you forget your TEV protease and the bench go for lunch these summer days and you come back, your protein will continuous working, which is something very good for us.

[00:21:44]Um, yeah. So, in this part, we wanted to explore a little bit more the effect that these mutations were carrying out cuz it's it's widely known that these mutations that are far away from the active site can modulate the activity in different manners. So, it can be that they stabilize the catalytic productive conformations or that they induce some global conformations that are good for activity. So, we use molecular dynamic simulation to explore that. First, we evaluate the key distances of the catalytic triad residues. However, we did not see any significant difference between our design and the wild type.

[00:22:28]But, what we did observe was an increase in the rigidity in the rigidity in certain regions that I show right now.

[00:22:39]Um, that they are not directly involved in the substrate binding.

[00:22:45]So, to further validate this this observation, and because it's becoming popular that the PLDT score from AlphaFold also can recapitulate somehow this flexibility, we also compared the results from the molecular dynamic simulation with AlphaFold ensemble of structures. And indeed, we found consistent patterns of structural rigidity and flexibility similar to the ones that we observe in the MDs.

[00:23:13]So, this would be two examples regarding protein engineering, and now I will shift towards the de novo design field.

[00:23:22]And in particular, in a project that its main objective was to design de novo mini proteins for cancer immunotherapy.

[00:23:30]But, before going into this project in particular, just as a reminder, what do we mean when we say de novo design? It's just creating new folds, new proteins that they did not exist in nature previously, and probably the next question would be, how do we do that?

[00:23:48]So, there are many different manners of doing this, but most of the programs that have been developed recently, what they do is they use diffusion models, and I will explain it in a very general manner, and I use my cat for that. So, what these models do is starting from a from data that already exists, let's imagine in this case a picture, they start adding noise in each step, so at the end of the day, what you have is this random distribution of noise. And the model, what they do is to learn to do the reverse process. So, starting from the random distribution, being able to generate realistic pictures.

[00:24:32]And thanks and because we have now many structures, experimentally solved structures, generally through X-ray crystallography of many proteins, we are now being able to do the same with proteins. So, starting from a random distribution of noise, we are able to generate folds, protein folds that did not exist previously.

[00:24:57]So, these have a lot of applications, as you can imagine, because proteins are involved in many different processes, and they range from a condition unconditional protein design, just for the fun, but you can also design symmetric oligomers, you can design protein binders against any specific target. This is what I will explain now.

[00:25:21]You can even, if you are interested in a functional motif you know that it exists, you can create a protein containing it, but also apply symmetry to this.

[00:25:31]And what I will explain in the in with this research project is how can we use this technique to design binders, protein binders.

[00:25:42]I'm not going to go into this today, but just because I think it can be interested for the audience. Already this technology is also available for nucleic acid design, and we are actually in the lab working on this to design RNA linkers.

[00:25:59]So, what is the motivation of of designing these mini proteins in this particular case? Well, this project emerged as as a result of this clinical study from 2018, in which they were doing a following up on clinical patients that were treated with CAR-T cell therapy in order to identify the T cell clones that were associated with better outcomes.

[00:26:27]And what they found out was that these cells were presenting a reducing or even elevated activity of TET2 enzyme.

[00:26:36]>> [clears throat] >> So, but we also know that if we disrupt the entirely the activity of this enzyme can have some bad consequence, like expansion or leukomogenesis, so the idea is to design inhibitors of TET2 by in a controllable manner.

[00:26:57]And for that, we use mini proteins. So, here you have in green or yellow the structure of the receptor, the TET2. You saw at the beginning of the animation and probably now where the DNA binds. And the idea here is to design the novel mini proteins that will bind into this DNA binding site so that DNA will be shifted.

[00:27:21]So, the way we did that was using RosettaFold Diffusion. It's a program that it used the methodology I explained in at the beginning of this diffusion for generating proteins.

[00:27:34]And we define the binding site where we want these mini proteins to bind, in this case the DNA binding site. And in this way, we generate different folds.

[00:27:45]Once we have folds that we like, we go for the design sequences of these structures. For that, we use again protein MPNN as in the previous examples. We generated thousands of sequences per each backbone. And then for the best sequences, we predict the structure by using AlphaFold in the monomer shape.

[00:28:07]Then, we select the designs that were predicted with higher building score and lower RMSD. And we confirm the structure in with AlphaFold in the multimer manner. So, we make sure that they are predicted to bind in the position that we designed at the beginning. And also, we use RosettaDock to do an evaluation of the energy of this interaction.

[00:28:31]So, after following the whole protocol and selecting our best candidate, we went for the experimental validation and characterization.

[00:28:39]These are how these five selected designs look like. So, you can see that the first design is only this four helix bundle, whereas the other four has this beta hairpin tail.

[00:28:55]We evaluate the thermostability of them.

[00:28:58]Here you can find for all of them their circular dichroism spectra and their melting temperature curve. However, for most of them, we were not even able to measure the TM because it was over 90°, so very, very stable. And for the design number four, we obtained a melting temperature of 80.

[00:29:19]Then, in order to assess whether these designs were suppressing these Tet2 catalytic activity, we measured the recombinant Tet2 activity using a chemiluminescent enzymatic assay. Paloma Velasco carried out these experiments.

[00:29:38]She did this experiment, of course, first for a negative control that, as I suspected, was not showing any inhibition.

[00:29:46]However, the five Tet2 binders designs, they were reducing the enzymatic activity in a dose-dependent manner.

[00:29:55]In particular, design number four was the most potent with a half-maximal inhibitory concentration of approximately 2 micromolar.

[00:30:08]We also evaluate the binding to Tet2 by using Tet2, sorry, by using biolayer interferometry with the with most of the designs binding in nanomolar with nanomolar equilibrium dissociation constant.

[00:30:25]And >> [clears throat and snorts] >> we still cannot believe this, but we were so glad and thanks to Maria Elena, we were able we were able to obtain the crystallographic structure of one of the designs. Um we were so happy to see for the first time the facing real life of one of the de novo design. In fact, here you can see the beautiful crystal she obtained and how the structure resembles almost the same as the prediction. So, in gray you have the crystal structure, the experimental one, and in light blue the model of AlphaFold. And you can see that for the core of the protein, the structure are identical even at the side chain level.

[00:31:09]But we saw some dis- discrepancies in the beta hairpin tail, which is expected because this is a very flexible region.

[00:31:19]So, we were very very glad with these results.

[00:31:24]And that would be the three examples I wanted to share with you today. So, we went from some applications from protein replacement therapy, also for biotechnological tools, also for immunotherapies. So, just to sum up, it reminds you that we have these two different lines.

[00:31:45]Something I really like is this plot from the protein design archive. So, this is those are the de novo designed proteins that has been already deposited in the protein data bank. So, only the one that has been crystallized.

[00:32:00]Probably there are much more, and you can see the trend, how is increasing every every year. So, yeah, just from a personal perspective, I think we are in a very exciting moment for us that we are working with proteins. So, if you are interested in any of these techniques, I I highly invite you to try because they are very fun.

[00:32:24]So, before finishing, of course, I wanted to thank my group, the Computational Chemistry Lab led by Gonzalo at Biounité, that is been always a pleasure to be part of this team, and all our collaborators. Of course, all of you for your attention, the organizer for inviting me and I'm happy to take any questions.

[00:32:51]>> Okay, thank you so much, Reyes.

[00:32:55]Thank you, Reyes.

[00:32:59]Okay, so I Maybe I can start asking while people start standing. Yeah, I just have [clears throat] two questions. Okay, really, really nice presentation, Reyes.

[00:33:11]I really like and I find that's really interesting, also.

[00:33:14]Um, so uh, what I am what um what I understood, maybe I missed something. In the protein engineering part, >> Mhm.

[00:33:26]>> uh, for um frataxin, right? There are some mutations that leads to uh, disease, >> Yes.

[00:33:37]>> But then, uh, when you uh try to increase the stability of some mutated forms and you say that they gain function and so on, these mutations are the ones that appear in disease or other ones that you introduce to improve the activity of that protein. I don't know that part I missed.

[00:34:03]>> I don't know if I understood properly. I think so, if not, you correct me. So, in this project, I use mutations many times, but there are like two set of mutations. So, we have the mutations, single mutants, that have been reported to be linked with the disease.

[00:34:22]>> Okay.

[00:34:23]>> And then, we have the other set that are the ones that we have designed >> All right.

[00:34:28]>> the thermostability.

[00:34:30]>> Okay.

[00:34:31]So, with those ones are the ones that you introduce or you design and then you have them with a new activity that will be like to restore the damage and to protect from this, right? So, I think this is really really interesting and it's a really nice field for a lot of different diseases, right?

[00:34:55]>> Mhm.

[00:34:56]Yeah.

[00:34:56]>> Okay, I don't have more questions. Thank you.

[00:34:59]>> Thank you. Okay, so this is a question from the audience.

[00:35:03]The first one is how do you see the evolution of the protein design field in the short term? And the second one, what do you think that are the main challenges to be solved?

[00:35:14]>> Mhm. Yeah, probably they are linked.

[00:35:17]>> [laughter] >> So, how do I see the field?

[00:35:21]It is exciting and overwhelming at the same time because there are many tools that have been developed every week, so it is difficult to keep on track.

[00:35:36]Um, but because of that, I also think that now the main, not problem, but question could be what to design, not that much how that was in the previous years.

[00:35:52]Um, for me the most challenge goal are enzyme design. It is much more tricky Well, the novel enzyme design is much more tricky that for instance binders.

[00:36:07]So, because you have many factors that you have to control and it's very easy that only with some mutations you ruin your enzymes and it is very difficult to create a new one from a scratch.

[00:36:22]>> So, basically continuing with that, the diffusion models right now has a limitation in terms of size. So, you can only apply for a small molecule. You cannot extend the application to complex or multi-domain proteins.

[00:36:37]>> Now you can in terms of technology, but if your system is too big, it's really that it's going to be It's more of a limitation a technical limitation, not about the model, but about the graphic cards you need to run them.

[00:36:55]>> Mhm.

[00:36:55]>> So, they have to evolve in parallel and that is not always the case.

[00:37:03]>> Okay.

[00:37:04]I have another question. It's related with the one that Patricia asked you before.

[00:37:10]How do you define the hot spots? Because sometimes they are not related into the catalytic sites. So, how do you define that the specific part of the protein that could have an effect on the catalytic site? Because this is I think that the key point because this is great to split you from the rational design because >> Totally.

[00:37:29]>> is the the the key part. So, how do you define that?

[00:37:33]>> That's the key part and each group has its internal rules.

[00:37:39]>> Okay. [laughter] >> We mainly use evolutionary information to try not to touch a position that by evolution has been defined as important or conserved.

[00:37:51]And also geometric for instance in the case of frataxin because this protein is part of a cluster.

[00:38:00]We have to make sure that we are not touching any position that is important for the interaction with the rest of the proteins.

[00:38:07]So, it's really case by case. Depend on the protein, you have to first study a lot your system and and be sure that >> is not a prediction. So, it is let's say handmade.

[00:38:22]>> A combination.

[00:38:23]>> A combination.

[00:38:25]>> [clears throat] >> Okay, just to you have done an affirmation in which you say that the PLDTT score [clears throat] that it is one the code that define the color let's say of the alpha alpha fold recapitulate the flexibility of the of the protein. Can you explain a little bit on this this affirmation?

[00:38:47]>> This is I mean it's not that the PLD score is defined like that, but it has been found in different cases that if you run multiple jobs for the same protein of alpha fold you can recapitulate somehow the flexibility the dynamic depending on the time scale of your protein. So and in fact in this case it correlated very nicely.

[00:39:15]More [clears throat] in in particular more using alpha fold two than alpha fold three.

[00:39:22]>> Okay, we have another question from Oh, there's so many. Okay, so the first of all is very nice presentation.

[00:39:31]Are there any consideration you should take when testing the novel design mini protein in wet lab? And the second one any practical or theoretical challenges or any specifics to consider related with the the novel design mini proteins?

[00:39:47]>> I didn't listen the second survey linear.

[00:39:50]>> Oh, don't worry. Any practical or theoretical challenges or any specifics to consider >> Um in terms of the de novo designs >> Mhm.

[00:40:01]>> um In fact, they are very in general very good behaved for expressing and purified because since you design completely the sequence and you in our case we use this program that optimize the sequence for a fall.

[00:40:21]At the end you have a sequences that are very stable and express a lot. So they are very easy to work with them.

[00:40:30]Maybe as a funny fact is that sometimes for instance some time ago I when I started working on that, I didn't realize I generate a lot of sequences with protein MBNN and then a I couldn't follow them in the lab. I didn't see them in nanodrop. I couldn't quantify them, but I did the CDS spectrum and they were there. So I was not understanding what was going on and it was as silly as they were not containing any tryptophan residue. So >> Okay.

[00:41:04]>> [laughter] >> I couldn't measure the absorbance spectrum. So those are little things that maybe are nice to consider, but in general they are very good to work with them. Very easy.

[00:41:19]>> Okay. So from Victor Barbacedillo, how are mini proteins delivered to the target if this intracellular?

[00:41:30]Um how do you deliver the the mini proteins? Because the receptor it is located outside or inside of the cell.

[00:41:43]Reyes, I think that I missed her.

[00:41:46]>> Can you hear me?

[00:41:47]>> Ah now, yes.

[00:41:50]>> Okay. Um that's an entire refill. So there are peptides designed for that as in the case of protaxin.

[00:42:00]Also you can use vehicles for that. So that's like a second step once you have your >> He's He's asking about the mini proteins. The mini proteins I think that yes, the the one of the third application, how do you plan to deliver into the cell?

[00:42:18]>> Yeah.

[00:42:19]Yeah, yeah, yeah. Is the same we with this for 32, we tried directly to see if they were able to internalize.

[00:42:28]We have not We have different results, so I don't want to say something that is wrong.

[00:42:33]Sometimes we achieve it, but yeah, that's something you have to study in detail once you have your designs because sometimes you have to change something.

[00:42:44]>> Okay, and the second question is, do you use the entire target protein for the computational design or only the region where you know or want the binding to bind?

[00:42:55]>> Mhm. If it's That's maybe something very personal. In my particular case, if it's not too big and you have good graphic cards, I use it in the the whole target, but it is also has been recommended in particular for Rosetta for diffusion. There are many other programs that you can cut your system only with the region that you are interested to bind, and then the calculation will be much faster.

[00:43:26]>> Okay.

[00:43:28]Um there are two more Yes.

[00:43:30]Uh do you think that computational tools that are now available for the design of nucleic acid could be extended to ribozymes?

[00:43:40]>> Who >> Or maybe the limited number of available structure is a problem for the for the application.

[00:43:48]>> I don't know. I see very difficult, but also if you ask me 2 years ago if I thought it would be possible to design DNA, maybe I say no, and it's true, so we'll see.

[00:43:59]>> Okay.

[00:44:00]Uh okay, more questions.

[00:44:02]Uh this is something that always seems to ask is how often do your computational design mini binders work experimentally on the first round? So, the yield of success of your design.

[00:44:16]You can give us a percentage.

[00:44:18]>> For binders is quite high for binders.

[00:44:22]And in the case it depends on the case and in the receptor. In this project that I present and I'm not lying, we only tried these five binders and all of them they show inhibition at at different but they all inhibit. It's not always the case but in general for binders is quite high.

[00:44:45]>> Okay, uh, one that I Okay, so basically you started for I don't know uh Have you tried some sometime just to test if some of the candidate that you have eliminated because of the of the prediction, you have tried tested later and see that could be a good candidate even against the numbers?

[00:45:09]>> That's an excellent idea, something we should do to train the models.

[00:45:15]We did in in these cases we didn't but that's something we all should do to make sure that we are filtering the right way.

[00:45:25]>> Okay, there is a question related with the uh let me check with the mini proteins.

[00:45:32]They ask also related with the size if it is the protein size matter to use this application. For example, if you could extend this application to bigger enzyme of of if you know if there is any specific limit for this mini protein because mini protein can you define what is the size of a mini protein?

[00:45:49]>> I know there is not a rule.

[00:45:52]Maybe for us are many and for other people they are too big, for other people they are tiny. So, normally I would say less than or around 100 residues but Um in terms if you can go higher, yes, because this program is not only for designing mini protein. So, I will try to go back. I don't My connection is not the best right now.

[00:46:17]But um to the example here, yes.

[00:46:22]So, in fact, if if you check this reference, you can check that they build huge systems for nanomaterials and so on. So, I As long as you have computational capacity, you can go for bigger systems.

[00:46:44]>> Okay. So, I think I want to check because there are different uh sites with the I think that this is so.

[00:46:53]A uh If there is not any any question, I would like to thank you, Reyes, very very nice presentation. You have guided us towards this uh very exciting field. Thank you so much for your clarity and uh very very nice presentation.

[00:47:12]>> Thank you very much, Reyes. Really nice to >> Thank you. Thank you. It was a pleasure.

[00:47:16]Thank you so much.

[00:47:18]>> Okay.

[00:47:18]>> Thanks.

[00:47:29]>> [laughter]