 I can start from the second lecture of Dr. David Baker, who is a professor of chemistry at University of Washington and an investigator of our youth's medical institute. Yesterday he presented a protein prediction problem and today he's talking about protein design. It's a very important application, so we're going to do what we can teach us about more about proteins and also we can think about what we can do with protein design. One of the things we're going to learn about is how to design enzymes. It's an extremely difficult task, because the proportion of protein design is not the complexity that is required for optimizing a transition state. And so we learn our strategy that includes how to create a combined conformational search that can be successful. I would like to start with this. Thank you, everyone. It's a very simple question. What does it mean to Europeans to be able to walk on the lake and all of you who managed to scare off everybody yesterday, I guess. So, as I said yesterday, I spoke about all the sequences that we have in all the wonderful biological organisms that exist in trying to be kind of compute from those with help from experimental data and where it's available to the structures and functions and interactions of the proteins that basically carry out all of biology. But today I'm going to talk instead about the inverse problem of trying to make new things that don't exist in nature and generally what I'll talk about today is some new function that doesn't exist. How do we work backwards to come up with an amino acid sequence or an RNA sequence that will pull it up to give that new function? I showed this slide yesterday, too. Before we were going, we were using this model of the interactive, the energetics of interactions to give the amino acid sequence, for example, find the lowest energy state. Now we're going to use the same model. And now we're going to, rather than starting with the sequence, we're going to start with a function or a structure that we like and find the sequence that has very low energy when it folds up to a structure that carries out that function. So we'll be working backwards from various types of functions to structures using the same energy function. I think that's really important that I really need to emphasize is that most of the time when you try to predict a structure, it doesn't work. You get information there that's very valuable about how your model is wrong, likewise with design. So by doing it, it's very important, I think, when you're doing computations for the computation, or that it's possible to complicate the new problem. That's really where you learn that model, if your model isn't right. And I think one thing that we do know for certain is that complication of biology is that current models are not very good. So it's very important. I feel like there's too much emphasis on trying to show how good models are. I think really important points to show how bad models are is that's the way to make them better. And fortunately, traditional design are very good ways of showing how bad models are because you fail very frequently. And every time we fail, we go in and we improve the model a little bit. So I think that's sort of how we look forward. So this is basically the workflow for all the things I'm going to be talking about in this talk. We started off with computer calculation of a design protein. And in some cases, and I'll talk about this at the end, for fun, we're trying to see if we can get some human input into this. But basically, you have a calculation on a computer where you're trying to design an infold or a new end sign. Once you have done this calculation, you have your protein model. You can read off the amino acid sequence from your computer model. And then, of course, we know the codes and we can backfence that into a DNA sequence. And then it's a very difficult step. You have to fill out a gene order form, which you then email to a gene synthesis company. Or if you are in a hurry, you buy all of those and you assemble the gene yourself. But basically, this part is what makes protein design very much easier than it used to be. And then, if you assemble all of those, you have your gene in a week. If you are lazy and you want any weights for the company, you do it. It comes in two to three weeks. And you get the gene back and we always make out these things in, usually, the coli expression vectors with his tags on them. So you put them in the coli. You break the cells. You run them over a nipple call and you have your protein. And then you assay. And then you find out how weird your design was. And I wasn't really sure where to include this, but this we just talked about a few days ago. And you can see that... I would say the opinion is not unanimous about whether protein design is a worthwhile endeavor. So here's how the public calculation works. One starts with a protein backbone, in this case. And now instead of... Now we need to search through all possible sequences or a sequence which is very low in energy when it's on this protein backbone. This is for the problem of designing a new structure. And basically, what's happening is that each step we're going through and putting in a new confirmation of a new amino acid or an alternative combination of an alternative amino acid. And we can do this very rapidly. Again, this moving, this batch of calculation doesn't take much longer than what you're seeing here. So we can very quickly come up with amino acid sequences that pack very well in any arbitrary structure. You can see here that the blue things are nonpolar so we end up with a completely nonpolar solution. Now in practice, when we want to make a new protein with a new pool, we can't just draw a backbone on the back of the envelope and then use this method to find an amino acid sequence which attacks the core well because there's no guarantee that there is any amino acid sequence which will really fold up to this structure. So because of that, we actually need to iterate after we do this first calculation to get the sequence. Then we go into structure prediction of what I described yesterday in that pie resolution movie and we let the backbone and then we repeat the sequence optimization and we let this go into we have a sequence and structure combination that really fit well together. So we used this a number of years ago now to see if we could make a protein with a topology that didn't occur at nature and this is before the days of gene synthesis so we've sampled all of those and when the protein was purified it was found to be extremely stable and we were able to solve the crystal structure and it was very, very close to the structure that we were trying to make. So we had the computer law model which is shown in blue and the x-ray structure in red and they are very, very close to each other. So this showed that we could make from scratch from scratch we could design amino acid sequences that fold up to new topologies and we're very, very stable when they fold it up. Now there was a big problem going actually before I get through them so since then what we've done is we've been trying to make new folds and the reason why we're trying to make more very stable scaffolds the thing about proteins that we're designing if you notice them they look much more ideal than most native proteins do they don't have long loops the secondary structures are regular and you might think of them as sort of being like a platonic ideal this is a very common protein called fold called a paradoxical fold design protein is very much similar so again we use the same protocol this protein is very much more stable than naturally occurring proteins again these proteins typically are penetrating in five or six molar quantities so they're very, very stable and then there's another one a Roslin fold again another very, very common fold and you can see why we're interested in these because of course Roslin folds use these loops to do things so we can make sequences that fold up to new structures either new typologies or similar versions of existing ones and we can do so with quite high accuracy and they can be very, very stable but they don't do anything at all and so where we were a few years ago we knew we could design new protein structures but they weren't very useful unless you wanted to rock because they really didn't do anything so what we've really been focusing on since then is trying to design proteins which actually do things and the problems I'm going to talk about today are designing new DNA cutting enzymes and designing new enzymes new enzymes generally I thought that would be my talk we have been working on designing HIV vaccines and there the idea is to pick epitopes from GP120 which are recognized by neutralizing antibodies but for some reason our bodies don't frequently make those type of antibodies because the epitopes are probably obscured on the virus anyways design small proteins sort of like the ones I showed you which present the epitopes in a very upfront and center sort of way to hopefully elicit a response and their hope is that those would then be neutralizing so this has worked out with my colleague Bill Sheath in Seattle and the result so far is we've been able to design proteins which interact by strongly to be neutralizing antibodies with crystal structures so the epitopes look right they elicit and new responses that bind to these epitopes even bind to the virus but they're not neutralizing so I guess if they were I would probably only talk about that so this is a mystery I think part of the big mystery of immunology of HIV infection okay so now to talk about DNA I think so what are the types of things you might want to do well long term we're working with class of endonucleases which recognize very extended recognition sequence 20 base pairs and if we could redesign these enzymes because any arbitrary 20 base pair sequence then there are a lot of interesting applications the most obvious one is in gene parity so if we go in and design an enzyme that would leave near the site of a mutation and then the idea is we would introduce that enzyme along with the wild pet copy and double strength rate then hopefully it'd be corrected by by copying on the wild pet template there's also possibilities for some antipathogen applications and being able to design a very specific entity basis so in this case now we don't just have a protein we have a protein here on top of DNA but it's basically the same calculation that I showed in that earlier where we made the change in the DNA so we could change the sequence of the DNA here and optimize the protein to recognize that new sequence and so here's an example here's a here's one, this is an extended site I'm just showing you a glow up part of that site here's the wild type amino acids at that position in the new base recognizing the wild type base pair and you can see that if you change this from a GC to a CG the wild type amino acids no longer can make good interactions but if you redesign using that movie I just showed you can get a very nice non-cooler interaction here and a nice hydrogen bonding interaction here and in fact when you make this protein it specifically leaves this sequence this CG whereas the wild type doesn't believe it the sort of data one generates this just shows a number of different designs but this is that new sequence the wild type enzyme the designed enzyme in green cuts this in a little concentration whereas the the wild type enzyme doesn't cut the site and we've gone on and now made multiple simultaneous changes and so I think we can we'll start working our way with a couple different enzymes towards more general recognition but I wanted to tell you about you know in the course of doing all this design work sometimes one comes across really unusual biochemistry which I think is going to be very interesting on its own right and I just wanted to illustrate that here so this is this is what we had to do another enzyme from this family which they had these two domains and they recognized this again here quite extended sequence and it had been known for some time that mutations throughout this site could make space changes to be throughout this extended site you knock out cleavage so this is just a this is just a cat or a ham the effect of the cleavage for all what you're seeing here is substitutions up to each of the four bases at each position in the site from minus 10 to plus 10 and you can see that throughout the site there are mutations which knock out binding and so there's some down on this side which is the internal side of the of the complex and there's others down here so it's very specific throughout so we got our first inkling that something funny was going on when we looked at the effect of these mutations on binding we found that mutations on this side substitutions in this site on this side had an effect on binding which seemed to affect binding at all so to go on on summer time we decided to carry out the K. Lisbon kinetics based on all of these different sites which was a endeavor and what she found was really quite remarkable so this shows for each substitution change in K. Am be the substrate in terms of substrate binding for the K. Lisbon complex and again here's the minus side here, this is the minus side you see that all these substitutions are affecting K. Am on the minus side and this is a very symmetric interface you can see that there's pretty much a two fold axis here but in spite of that there's a real asymmetry of how the interactions are used but not here seem to be used in binding the substrate whereas the mutations here again have very little effect on K. Am but you look at where the effects on K. Kat are, substitutions down here in fact, on this side actually in some cases even make the reactions faster but mutations down on this side almost universally slow the reaction so there's this really strong partitioning of what the binding energy is being used for on this side it's being used to bind the substrate and here it's being used to utilize the transition state and the simplest way to think about what will attract a picture is that this end time needs to scan through a genome and bind this right condition site it binds in the transition state complex which probably resembles the crystal structure to this bent confirmation which is probably not sampled very often in a cell so instead what it can do is it can bind to B form just through this side, it binds in that way and then this interaction, this stabilization of this bent state basically is used to stabilize fluctuations of the DNA that then put into an confirmation that's heavily confident but we can now take advantage of this and we can make specific designs then that either effect K. Kat over 4 K. Am so these are just designs that are designed on the minus side, designed on the plus side so this is one that's on the minus side and again the color coding is, the different colors represent the different bases so in this design see so the wild type shown in dotted lines and the the design is the solid line so this design increases the K. Am makes it worse for this face, this face this face but makes it better rocks the K. Am for this one so it achieves, this design is changing the specificity of wild type by modeling K. Am for the different bases but it has pretty much no effect on the K. Kat for these different substitutions. On the other hand we have this design on the plus side which is not having much effect on the K. Am's for the different possibilities of that position but it's drastically reducing the K. Kat's for the wild type pretty much has a uniform K. Kat's are very similar but this design they're lowered for three of the four base pairs leaving only the blue one highly active so this shows that we can now sort of go one level beyond sort of this great design specificity we can actually go in and now modulate the parameters independently which you can imagine could be very useful and depending on what your bio-athletal application is okay so now I'm going to switch gears and talk not about redesigning in type specificity but about the no going side design and so if you can come to me and say design reactant enzyme which capitalizes this chemical reaction the first step is to compute what the intermediate and transition states in that chemical reaction are then the next step and this is sort of the fun part is to design a disembodied active site around say the transition state for the reaction so you might say well I have a hydroponic group here a positive charge here, a negative charge here but these are completely disembodied and as Alessandro mentioned we can use on-mechanics calculations in collaboration with K. Kat's group to sort of disturb the optimal positions of these you can also use chemical intuition about where where would be optimal how would that best upload and let out flow in a bond rate in a bond making reaction this is really just hypothesis this is like what you see in a biochemistry textbook you know they first show you the whole thing that it's doing on the active site and we just see it as having a triad in the transition state we start with just that low enough view but we don't know what an ideal active site would necessarily look like we just have hypotheses we need to have a number of different hypotheses because we don't know which one is actually going to be able to catalyze the reaction so once we have this ideal active site we design a protein that contains this ideal active site so we have this ideal active site we want to make it now we have to build a scaffold which has that ideal active site and that's why we went back out the top seven to start kind of making bolds these are basically two ways to build a scaffold one is you could construct a noble scaffold that is just perfectly for all the back but it's just perfectly poised to hold the catalytic residues and the other thing you could do is just go through naturally occurring scaffolds and so what I'm going to be telling you now we're going through a naturally occurring scaffold of thermohella proteins that are very very stable so where we've removed large numbers of side chains put in the ones that create this site but this isn't really ideal because these proteins really evolve to do something else so we'd like to be able to do it from scratch so given the description of ideal active site then we need to next up is to well let me illustrate this first part we need to find we need to figure out what scaffold we can use to actually hold that ideal active site so here's a really simple example this is called the chemical elimination reaction where we abstract this proton and then it's going to hopefully lead to the breakage of this bond here so this is the product so the first thing we need to do like I said is to choose a chemical peak this is like a disembodied active site so here's an example here we have a negatively charged group that's pulling off the proton that's forming a space here and we have a positively charged group that's stabilizing the negative charge that's appearing here and here's another scheme where we have a histeme that's pulling off the proton and a proxylate group that's backing up and holding the histeme in place so we have a hydrogen bond in this case that will stay close to the negative charge so you see there are a lot of different options that you can pour a very simple reaction on this so the next thing we do is we take this large set of staples and we ask where can we geometrically we constitute this site so the backbone we need some backbone here, here and here and we need to have room for the substrate so we fill a geometric caching algorithm for that after we found these places where you can support these residues the next step is design all the surrounding of the elastic that's shown here it's a little hard to see and the final step is to select the best designs and make them and this is now an illustration of the two designs using two of the motifs the two motifs I have on the proceeding slide so if you remember there was one with the carboxylate here and there was the one then we had see you can't see the lysine I think coming in in this case you see the substrate which is in yellow this is being in a nice hydrophobic pocket that's been designed that's part of the second step of the optimization now we have an aromatic residue on top that's sort of planting the substrate in place now the other option to call was having an isthene which is now pulling out the proton out here and this is backed up by a proboscillate group in this case an aromatic group that's planting it in so the the neat thing is that these designs they're very simple but they do have enzymatic activity this is just an out of product form versus time and when you mutate the residues which are like this is the glutamate this is the one that was on me you make the glutamate into a glutamine you completely lose activity and then use the isthene in the sparking if you would move the isthene you are down to here and if you would move the sparking you're down here so it's not as absolute a requirement so that's a very simple reaction I'll just take you through a couple of other reactions and then I'll make some more general comments about where we are with this and so here's another reaction and this one is interesting because it's not catalyzed by naturally occurring enzymes or at least there are no well documented cases of enzyme catalyzed reactions it's called the yield-solver reaction so here we have this group of blue, this group of green and they're going to come together to form this ring here so we're forming two carbon-carbon bonds in order to catalyze this reaction one has to bring these two groups next to each other and that's shown here and also we're pleased to perturb the orbital energies of this guy so that they overlap more to increase the rate of reaction and we do that by introducing hydrogen body groups which an electron-domaining group which will raise the energy levels on this side and an electron-crawling group here which will lower the lumo energy here to increase the overlap again this is basically the same thing I showed before so this shows an example of such a design here we have the small substrate and the big substrate we have a hydroponic group here and a hydroponic group here those are what are doing the orbital derivations and again this is the case where we go through a big gene and this shows a product versus time here notice that this is not a very good enzyme if this were an apple, they aren't actually bringing in enzymes to catalyze this reaction but the word this would probably be second but that's just where things are right now so we have a crystal structure of this design as we did we do also of the preceding ones and in general the crystal structures are quite close to what we to the goal we've been trying to make the interesting thing about this reaction is that these substrates can come together in several different ways and so this is the substrate this is the other and so there are various anti-mersite that can be produced and you can separate these as shown here and what's neat is we only form one of them and it's the one that we were actually trying to make this one here so there's some specificity here so alright so to summarize where we are so far then I showed you an example of how that uses general acid-based catalysis pulling on and on putting one on on the other side we can't eliminate it I showed you a bi-molecular reaction that deals all the reaction and they talked a little bit about the elbow reaction which involves covalent catalysis where the enzyme becomes covalently bound to the protein and knock and knock on ester hydrolysis so we've been able to use these are pretty much the strategies that native enzymes use to catalyze reactions but there's a big difference which is that our designs are much less effective catalysts than native enzymes now I'll come back to that just to show you what the diversity of chemical reactions are here's the camp I showed you I talked about this one already here is the reaction I'll talk about in just a moment forming this carbon bond between these two substrates there's ester hydrolysis so these are quite different and this just shows different types of what the designs look like okay so now these really aren't very good catalysts but we've been collaborating with several groups who are expert at laboratory evolution including the antics group here and the designs I showed you those pictures of are called K78 K59 but also the numbers are important because it gives you an idea of how many times we had to try we don't have all the numbers because some of them work solid some of them we decided not to make but we're trying quite a few designs so it's like a one out of ten sort of thing that we get one that works so that's an important caveat to what I'm telling you but we always keep the numbers on so you can see in his lab and as a student although constantly very talented a student took them and started evolving them just making mutations screening for once that have more activity even though this is a fairly simple reaction there isn't anything in an equal act which catalyzes it so you can just rake over the cells and do the assay and she's now up to about here I don't have her by slide but the best designs now the best evolved version of this K59 design that's the one that uses the the glutamate is now about 5 times 10 to 5 which is starting to get into the range of naturally occurring enzymes so the maximum diffusion control will be 10 to the 9 so dark so it's starting to get respectful and the rain investment is on the order of 10 to the 8 a little bit about that so these are still not as those naturally occurring enzymes but they're getting better of course we'd like to be able to compute what these these are the sequence changes which are spread out throughout the protein and it's been possible to get crystal structures of a number of these evolved variants this is actually for the first design we had activity I haven't shown you a picture of that's the one through which we have the most structures now it's called A7 and we were really really silly when we made the design it had a glutamate that was attracted to the proton and this lysine if you remember it had a negative charge but if you put an awesome next to a negative then they can decide to do other things and that's what you see in this crystal structure here now once we saw this crystal structure of course we got rid of the lysine and that increased the activity considerably but evolution itself follows in unusual ways sometimes so during the direct evolution what happened is a negative chart, a negative cluster from evolved down here which distracted the lysine down and thought it out of the way and what you see happening by the final round of evolution is that this catalytic residue is really held quite a bit into place, really better positioned to lock that proton up so this is this is a the reaction I mentioned before this is the retroalval reaction in this direction we're breaking this carbon-carbon bond here and this reaction is a little bit different as I said we're forming covalent bond between the protein and the substrate and that's shown here this is a complicated reaction these are many many steps you form this enzyme substrate intermediate then you break the bond here and then you need to recycle the the free enzyme well we've been able to do here, well actually you can see so we've had to try as I said we always have to try multiple mechanisms and we tried it in a number of things and the one that worked really well was one which where we used a water molecule that was positioned by two hydrogen bonding groups that we designed did and this water molecule helps facilitate proton transfer between these two oxygens this is the bond that's going to get the lead criteria so what's been exciting recently about this is a couple of things first of all we've been able to using this motif we've been able to design this active sites and catalyzes reactions very reproducibly so we're not at 1 out of 10 anymore we're better than 1 out of 2 so we have a very high frequency of the designs the new designs that we make for this reaction actually work and this just shows the range of protein scaffolds and the ranges of positions of the catalytic lysine that we have now active designs so this we like this one because we actually understand how to reproducibly produce designs that catalyze this reaction and they're quite pretty this is what a design model looks like here's the two residues which are formating the water here's the lysine that's coming in and forming a covalent interaction with the substrate and here's the binding pocket for the substrate here so what's neat is we just have gotten a crystal structure of this and I think I'll just skip to here so in the crystal structure we let's see we've been able to trap the intermediates on the lysine this is the lysine here by using our hydrohypride production we can actually trap both the substrate and product on the enzyme and so we have pictures of what these substrate and product complexes look like one thing that's quite interesting which I'm not showing you on the lysine this one accumulates the substrate there are other designs which accumulate the product so again there are multiple steps in this reaction and it's hard to do them all well it seems like different enzymes did some extent block the different steps what's neat here is these two water molecules which we had designed in we actually see in the crystal structure they're interacting here and they're positioned by all these hydro body groups so this is our we're excited about this because this is our first co-construction of a enzyme and a enzyme with a substrate the substrate lies pretty much in the pocket that we designed in ok so where are we after all of this first of all we can't design active enzymes from scratch I've shown you that now we've done this now before these are low they're going to be increased by directed evolution there are clearly some things are happening that are on their competing reactions the product of the reaction goes back on slows down the activity it's clear that we need more precision positioning and the catalytic groups are a level of control and it's not so high and I think one of the bottom lines is that real enzymes are masters of the art of compromise they have to bind substrate and carry out the chemistry they have to release product they can't get hung up at any of them and if we're producing this it's really not very easy and so from a practical point of view one of the key questions is not really about protein design but about evolution we can make things that catalyze these chemical reactions we can make them from scratch but they're again not very active catalyst and we can also make quite a few of them for the alkali we can make many tens 40 or 50 by now but none of them are very active so the real question is I think we can turn back the clock 3 billion years and look at the ancestors of modern day enzymes how active were they and how many different starting points did they just need to be able to get one that was really really good is naturally occurring in enzymes so I just ordered for that to be better than these so it's not clear whether can we take any one of these designs and do we not by evolving it doing enough rounds of direct resolution to really turn against enzymes or are there many large fraction of dead ends so sort of a question about dead ends and evolution and this is the practical question I don't think it's likely that in the short term we'll be able to design from scratch catalysts that are very very highly active so I think from a practical point of view we'll be making new enzymes we'll be left with sort of making things that have starting activity then optimizing them experimentally one of the areas that we're very interested in here is there are a lot of interesting applications obviously in the energy area because it's, for example, is some you can capture energy using oscillating solar panels very effectively but using that energy to make molecules you don't really have a way to do that but if you could design a design sort of a pathway that could take electrons off an electrode of a solar panel or from electricity or would take a voltage difference that generates at least a solar power an electrode, use it to reduce some compound and then get electrons from there into a cell through some sort of assigned pathway so there are a lot of interesting applications that I've designed but we kind of need to know the answer to this to know how many starting points do we need to have and so forth so that's one of the real key questions so just to sort of summarize what I told you in these two talks then today is that for the structure calculation problem the problem was really a search problem that on the lowest energy state per cookie is a very, very good hole for anything but the smallest per cookie structure we didn't really have a problem with accuracy though because once we got close enough to the native structure we really fell into this deep hole and because of that the solution I was advocating was to get experimental data to sort of indicate what the position of the of the native structure was and then do computation to work out all the detail that's for example what we were doing in the structure in the case with NR data we used the NR data to guide to lowest energy search and then we would but we wouldn't use the data in terms of fine detail to the structure because there was a steep, evolved energy gap because that solution had really optimized that sequence to be very low in energy in that native structure now in the case of designing functions it's really quite different we don't have a search problem we can come up with these motifs and then we have these algorithms for rapidly finding places where we can design proteins new backgrounds and cavities the problem is that our accuracy is very high because unfortunately we no longer are dealing with design the problem is that we're dealing with things that we've designed we've come up with so they aren't really as optimized for the structure we want to have that we have for in the structure prediction case but we also don't we're far from having a complete understanding of the requirements for the policies and I think by designing new enzymes or trying to design new enzymes we're learning a lot about what really is important and this policy can be understanding we we do the computation first because we can, again, we don't have the search problem we can just build any site we want in principle but we need experiment to tell us how minimum our designs are and then to optimize them and we need a lot of experimental feedback to tell us how to both improve our current designs and learn in general how to make better enzymes so I'm going to give you a little bit, I want to show you a little bit more about full net just to be in harmony with yesterday but let me first acknowledge people who've done this work so the design of the pieces of scaffold that I showed is the work of two very talented Japanese postdocs the end of the new phase we designed the work to talk about the work of the summer time and no one has ever signed Justin and Alice through the work on the Neal Solverace Eric and Link have done the work on the Retroal Delays and Daniella through the work on the camp and it's been a collaboration with Danny Toffett over at Ronsky and we've got over so let's see now so now what I want to show you is that the Neal and Full Nets set up to not just predict structures but to actually do design and so what I a puffle that's up now, I'm sort of limited by what you can get on the web, is rather than designing new enzymes this is designing approaching much more quiet this is a sort of model problem we're working on and here's why I said here, we give people a starting structure where we use the same sort of matching algorithms to basically build an ideal act of sight sort of copying the strut that hydrogen bonding groups to bind the base of the monotone and you can sort of see that here but then what people can do, in addition to all the things I showed you yesterday they can go in and they can try and make it better and I'll show you how that works here we go so this residue I'm allowed to be taken so I can now go in and I can change that residue to anything I want so maybe I might try and be on a trip to band there for example and now I have a trip to band there but you know I created that little those red and nasty things that doesn't clash, that's not good my score is not much worse we can again compare my score to everybody else let's see yeah so I really have done that thing but I can now I can do the same kind of things as before I can I can move that trip to band there I can move everything around in this case not like this happening so my score is going up and so I'm very excited about this so these problems we're starting to give people now the option too they can change the sequence but what they can also do maybe also one more thing is we can we can add more residues well that wasn't a very good place to do this but people who hold it to those are very good so what they're doing now is we give them starting points and for example for protein protein interactions we can then decide protein stock up against the influenza virus and then they can throw out loops and they can pull the backbone around the sort of ways that I showed you and change the sequence so we're trying to see if we can get all these great approaches that I showed you yesterday for protein protein design but we're actually starting to test some of these in the lab anyway of course most people don't do things like this that's kind of embarrassing but anyway so I think as soon as we can start making things in the lab and people that design all these people around the world and they work and I think that'll be sort of a new point, a new way sort of step and sort of get listening to the general public and sort of biomedical research and also I think really again you get an appreciation for what the molecular nature of a lot of problems are so anyway I'll stop there well yeah I think the most obvious way to do it is to take again using direct evolution if you can truth the activities if you can compare the dynamics of the ball version of the initial design it's very possible because when you're using these thermophilic scalp they're pretty stable so yeah I think that's a really interesting question, we don't know what's missing so I think the sort of two schools of thought one is that these things are too rigid and they don't conclude enough and the other school of thought is that they're actually not rigid they're not binding the substrates like to get a substrate up against catalytic residue will generally be uncapable so you really need to exert force in a very precise positioning to do that and so that may be missing probably both are missing we don't have precision where we need it we don't have dynamics where we need it I think I think I'm pretty excited I think we'll learn from if I can continue to follow up what about the lines that contribute to do the assays to the function of temperature they have pretty much they have thermophilic profiles initially but after you follow them they have more mesophilic lines which is sort of what you'd expect I mean that's the yeah so in the means other reaction that we were trying have you ever thought of putting in I mean the means other transitions state is really high in energy and has got kind of an aromatic headache that's an interesting idea we haven't tried that we've used primarily aromatic groups for satin but that's an example of where this is at there are a lot of different ways of going about trying to solve these problems and you know it could be in some cases we just miss on the optimal idealizer site configuration so yeah yeah so of course what we're doing is we're now going back to the complications we've come up with these substitutions in the first place in fact he was a graduate student here in my lab about a month ago as a postdoc and he's really focusing on that question there are let's see I can't think of a specific case where we know we threw out tonight but it almost must be happening he's only doing the computational design it's going through all possible sequences and if there's something that we find out that it was clearly must have been missed a lot of it is there's a lot of there's a little bit in the companies for example of bigger residues so we sample both ligand confirmations but we can't do that and then we find an example where we find something perfect yeah do you know about designing the loops to make your rise more flexible yeah that's right so that's one of the things we're really focusing on now to go to these and try to design the loops yeah more flexibility maybe to allow the substrate more yeah exactly so we're trying that now you mentioned you need to do the slide changes in the financial state here you clearly can't do that so like the looping and the kind of that looping model that you approach in this case so we have to calculate what that would be yeah well we can also determine it experimentally by looking at the pH and the reaction and actually we kept it since you just started pretty much where you'd expect them to be so the PK's seems like getting a PK shift is not such a big deal but for example you have a very licy that PK dropped so I thought that was a harder problem than it turned out to be we're more limited by other features it seems so you're able to predict I would say that in general designs when we go and look at the PK's there in the range you'd like so the licy tends to have a PK around 6 now for 7 the glue remains around 6 so I'm just saying that now we obviously are able to predict reasonably what the PK's are going to be we're not going to get into turbulence just by burying them when you get them you're shifting what is the number of N? actually it's the address of N what is the N? what's the number? I think the smallest one is about 150 N I don't think you can cut down much because you need to have a pocket at least for if you look at how intense mine substrates if the substrates small generally need to have it looks like that will develop or if you're substrates you can kind of lie on the surface yeah yeah yeah yeah yeah I think that's actually I think that's actually right yeah so I think the water is like you need to get away with prime saver get away with reusing the ends of side chains because they can move around they can easily rotate them that's exactly right well actually what we're trying to do now is replace those with with charged groups because we think that maybe why we have why we're not getting to a certain amount I think it's good for starters but in actually occurring evolution I'll bet that early on there's a lot more use of water and then you gradually replace water with things that you could water's good back in the same piece and it's good and how do you talk about designing the economy for this whole yeah so we are trying now to make design coaching so that we'll be totally solid and organic and so we're trying to basically take the top seven that we've got in version I think I'm trying to figure out how we can make the protein so we can get it to see whether it's actually solid or not then finally the enzyme is in that range they'll come in a common area what about enzymes that can do successive operations so attaching domains, sort of like the polytheta it's interesting yeah well I think that's the end for the future I think we need to understand how to be simple first the retroal law reaction so all the solar reaction pathway we can see even for that simple thing we're getting stuck in different steps so we have to be a lot better than they are and can you in a rational way design allosteric switches something that you could control with light or live in or something yeah well we're trying to do that now actually because it turns out there's a phosphate binding site in this capital core of the back of the company so we're now trying to see if we can put in larger groups of phosphate and see if we can modulate the F2P that way so I think that should be possible finally on the DNA interface aspect of things how frequently do you find a base by base recognition pattern versus multiple bases and affiliated changes coordinated changes of sections that's right so we've now found we have some cases now we've been able to redesign four consecutive base pairs and they clearly are much more more than the individual residue substitution so you need to consider them all at the same time do you see sidechains move in a coordinated way I think that's exactly what happens that you change one and therefore you can change another so we do see exactly that kind of thing going on we do this design