 I enjoy very much the meeting, the discussions, talking to the colleagues. It's also a very special moment for me because it's a personal one. I met in three years, thank you, for organizing. I met in three years my teacher who taught me math when I was in college. There was not just a teacher who taught us math. There was a group of Jewish kids who didn't accept it to the university, went to a metallurgical college, and then the teacher from Moscow University taught us for three years, absolutely voluntarily. Every week we would go to the university, and we would be taught math by Yusuf Brovin, who is actually sitting there. So I'm very nervous now. His professor in three years, I met him after 30 years of not seeing him. And when I was there, I was kind of okay. It was obvious for me, we were studying math, but it actually affected me a lot. And I only realized later how it was important for me these three years of studying with Boris Anatolyevich to Brovin. So thank you very much. So I dedicate this talk to him. Sorry for the personal note, but I could not miss it. Okay, let's talk about our function. And so if I would be nervous, please excuse me. So our visual system usually performs very easy computation when we recognize faces independent of angles and lightning and sizes. We can do the task with very few pixels, actually. We can recognize the same face. This problem has been studied in the visual law, and this was the same problem in our function, and we kind of tried to attempt to solve this problem in our function. And in our function, we always need to recognize objects. We need to identify the smells. And the crucial parameter of this is intensity. So we can recognize odor identity independently of the strengths of the smell. Let it be, you know, kofia. I'm sorry for this picture. It's a very weak kofia. It shouldn't be shown in Italy especially. And very strong kofia near the nose. It's always kofia. You call it kofia, independent on intensity. What is going on in our brain during this process is how we sort it. We actually have some ideas and some... But we don't know how we identify different modes, different odors, independent concentration. Of course, there is always a big other problem that we need to identify odors independently of the presence of the background odor. And I'm not going to talk about it. If I will have a few minutes, I can speculate it, but I don't have any evidence and I just only can bring my speculation. So we're going to talk about odor identity. I give a few presentations. There are very basic slides, maybe to explain how alfaltini works. So everybody will go on the same page. Odor gets to the nose, interacts with the receptor. The receptor located in the cell of alfaltini sensory neurons in the alfaltini epithelium. All the neurons that express the same receptor type converge to one small area in the alfaltini bulb called glomeruli. So we have a lot of integration on this stage. All glomeruli kind of collects information for the same receptor type and we can call it as a channel. All receptors converging to one glomeruli can be called one channel. So if you present an odor, it excites multiple glomeruli and different odors excite different sets of glomeruli. The system is very well set up for the imaging study. If you open the alfaltini bulb, if you open the bulb here and get an image with your favorite indicator of the activity, you get a picture like that. And here's the first problem. So here's an image that corresponds to two odors, the welder height and hexanone. There are different patterns and we can spend a lot of time solving how odors are identified. Now, these two patterns are actually different from these two patterns because these two patterns belong to the same order of different concentrations. So the brain needs to put all these patterns into one category, these patterns in different categories and also identify the concentration. So how it may do this? So I propose a very simple model. So the plan for the talk will be the following. I propose a very simple model, how the brain can do it. I bring some experimental evidence. I'll do a lot of speculations. And I'm apologizing for some people who heard this part, but it's kind of important. I want a discussion, so please. I probably don't need to ask you to interrupt me. And then, as time, because it's very hard to predict how the talk will go, I have two subjects to approach this. Two topics to approach this. I'm presenting the model of the concentration brain coding. I can present two stories, either our attempt to approach this problem from a very basic perspective, just coding problem using after genetics, or how our model may be implemented in the neuronal code of the alpha-3-bomb. So let's make this in the middle of the talk when we come to this. And I'll try to keep the time. So here's the problem. So receptor interact with the ligands usually in a very... So receptor got activated about some concentration, and we can explain it in a very simple fashion that it's a very steep function compared to the whole range of concentration that we perceive orders. In one order, we present multiple... If we plot here multiple receptors, we can see that different receptors activate a different concentration. And that's kind of often forgotten, because when we heard talks about this is a ligand, this is not ligand, I think that this may be often ill-posed because as you see at this concentration, these receptors are ligand and these are not ligands. And as you increase concentration, you recruit more and more receptors. What's going on in the brain? Imagine we have a factory bulb with 12 glomeruli. At a low concentration, we activate 2 glomeruli, then 4 glomeruli, then 6 glomeruli. And these all patterns belong to the same order. So what's interesting is the intensity of these patterns is not always informative because we often can see that the most intense glomeruli are not the ones activated at the lowest concentration. It may be multiple factors for that, but let's talk about it a little bit later. So if you know the identity of the first 2 glomeruli, you know what order is. So you know you can... This is a common feature across all three patterns. This is your order, this is your cost signature. But if you don't know how you solve this problem, how you put this pattern into that category. And I propose a very simple solution, is that that happens in time. So the most sensitive glomeruli is activated first in the SNF cycle. So at the lowest concentration you activate glomeruli HNF, HF AG, HF AG, LK. So if you know the first 2, you know the order. Very simple, but very simple model, but has a lot of consequences. So we call this model primacy coding. And basic parameter of the primacy coding is the number of first activated glomeruli, P, or the temporal window glomeruli activation. It's, as I said, oh, okay, first explainer, all models are wrong, some of them are useful. So this may be useful model. Why it's useful model? Because it's actually, first of all, it's explained concentration... Oops, I'm sorry. It explains concentration variance. Just by... It has been built to explain concentration variance. But there's an interesting prediction of this model. First of all, it's 2 glomeruli, the small number of glomeruli is enough to identify the order. And second, you don't need all of that. And this is kind of maybe disturbing, because when you look at the pattern, you think that all glomeruli are methods. Maybe not. So that's what I'm trying to discuss. So it's kind of... It's a special code with very first 2 glomeruli. The first few methods, the rest not. So the plan for the discussion of this model will be the following. First, what are the neural mechanisms responsible to creation of the code? How the code can be read? And then, what are the basic coding features that correspond to our psychophysical knowledge about alfactory psychophysics? And then, experimental data about the window of relevant information that is related to this model. So if... As we inhale the order of the concentration of the nose, this very simple model of how the code can be shaped. As we inhale the order of the concentration of the nose raises gradually, and excites glomeruli depending on its intensity. So the most sensitive glomeruli will be excited earlier. The less sensitive will be excited later. So this is exosensitivity or sensitivity of the acceptor to this order. Now what happens if you increase the concentration of the external order, you do the same thing, and it will move the glomeruli activation earlier. It will activate another glomeruli, but it will preserve more or less the sequence of the code. And Andreas Schescher talked about it. It was a great paper, first theoretical by Hallfield. And then Andreas actually did the work with some evidence for the latency code. So the latency shifts with the increase of intensity, but the sequence remains the same. But here we're saying that only the first of what is meant. So actually, it's indeed if we change the concentration, the mitral cells that are listening to the glomerulus shift their response first. So we should remember that this shifts for about 10 milliseconds before it changes. So how the code can be read? The code can be read in a very simple way. So glomeruli activate mitral cells in a sequential order, and mitral cells project to perform cortex. And some mitral cells converge to some of the preformed cortical cells and create some kind of symphony code. And those who are activated first, actually, the winner takes over. The first kind of the first who is activated, we can establish simple inhibitory network and only those who are activated first can define the order. The rest can be suppressed. What's interesting here that if you increase significantly concentration, you squeeze more glomeruli in the small window, and you cannot discriminate between first and second activated glomeruli because they activated very fast. And then you create lorax. You create something with your high concentration of very many mixtures. It's not that you have activated too many glomerulus. You've activated too many glomerulus in the first temporal window. And you change the identity of the order often. So some order smells differently at very low concentration and high concentration. It's over simplistic model but yes, kind of for given order, you think you have a projection. But it may be plastic. It may be long. I'm not talking about it. Now, actually, there was a nice paper published this year by Kevin Franks and complimentary codes of order identity intensity from Alex they actually measure timing. And they show that the order identity this is discriminant analysis of the cells recorded in the cortex. Order identity in the cortex arises very, very quickly, much faster than anything else. They did discriminant analysis and show the first wave of activity from the bulb what is metaphonic the order identity, not concentration. So how the code deals with mixture? Imagine we have the sequence of glomerulus C, A, T, M, W and based on our idea if prime is Cp equal 3 this is order ket. The second order is order dog. When we mix them together we get actually new percept that is reminiscent of the ket and dog but it's different from Cad and this has a different it's a different object. So basically you mix the order and you forget the tail you actually pay attention from the beginning and you create a new object. This is your new gift was interesting that it actually depends on the relative concentration so by changing concentration of the components you can actually change from Cad to COd and this is different orders Well I mean do animals have concentration variance? Humans there is a lot of literature and controversy about some people we definitely have concentration variance because we can identify coefficient in the different concentrations and we can say that identity may change but anyway for the identification of some object we have concentration variance do animals have this? Yes they do there was beautiful, yes I don't so we can argue about it but we do have some concentration in variance. Now do animals have it? Now the exact meaning is a very beautiful experiment they train an animal to discriminate mixtures of order A and order B and this is a component of the mixtures that they put the decision boundary here this was 95.5 and 5.95 all the mixtures was along this line and this was this line so if animal maintain identity ratio between what this kind of ratio is if you proportionally change concentration of two components you think that the object is preserved so if animal paying attention to the order of the relative ratio between component when they in a probe change concentration they half make the half of overall concentration but they reserve their ratios then the decision line should be exactly the same exactly I don't want to go into details but they show that animal doing decision based on that line but not any other so if animal would be paying attention only to order A then it will be doing decision based on the green line but animal were doing decision based on the the yellow line that is correspond to the same ratio and the same thing happened when they increased concentration so they could actually change it also 10 times sorry I didn't show the data but it's actually in some range animal maintain concentration variance and it perfectly corresponds animal paying attention to the first few orders there are other mechanism but at least this is consistent okay what is the information capacity it's a provocative slide it's actually the description is absolutely provocative I tell you why so if you have N number of glomeruli and P number of relevant glomeruli then number of combination of P and P will be N to the power of P divided by P factorial will be if P equals 6 I have no idea why 6 but if P equals 6 for humans it will be 3 million orders I brought it just on purpose I was actually very excited about this paper hope to submit the P but this computation only tells us the code is you can convey a lot of information in reality I believe that not all the combinations are possible and Alex actually will talk tomorrow and that's why I actually appreciate that we switched and Alex will talk tomorrow talking about the prediction of how many orders can be encoded and what is actually the limitation for such a code yes said again sure I don't know this parameter I'm just giving that small number of 6 is give you huge combinatorial capacity I give it it's definitely not one but it definitely much less than total number of active glomeruli that many tens of hundreds of glomeruli that is my point basically the combinatorial capacity of this code is very high so now it's important part about the relevant temporal window and to do this we basically try to identify relevant temporal window of the information person this experiment has been done in my lab and the key figure was Chris Wilson he did the following experiment he trained the mice to discriminate two orders A and B and leaking by left and right and head fixed setup very easy to train the mouse the key point was we don't know how mouse is making decisions so maybe intensity maybe something else so we remove the intensity factor from here with scramble concentration two orders of magnitude presenting five samples of A five samples of B but mouse needs to make a decision only based on identity during the task we recorded the pressure in the nasal cavity and we very precisely timed the order delivery so we know that we order delivery time by onset of exhalation when mouse start inhaling we know that order was delivered in a precise concentration there you know very controllable expand the interesting part is that we use mouse that has channel adoption in the all factor receptors mouse leak left or right on the order A and B we use mouse that has channel adoption and we put in both nostril two optical fiber and we blast the light like that at a specific time so the idea we try to master order with the light and what happens is that we know when mouse start inhaling so the order processing starts here sometime after we start blasting the light so basically it's as if I do the following experiment show you one pattern and another pattern I may make it a little bit difficult I rotate the pattern so I ask you is it left hand or right hand and then I blast the light I try to blind you I don't want I don't activate some random subset of receptors it's not that I activate it all but I activate the substantial subset of receptors but I do it in a very time precise manner because this is a very controllable parameter I cannot do it with others maybe Andreas can I cannot so the idea is if I blast the light very late mouse can easily do decision but if I blast the light very early mouse wouldn't be able to make a decision and by moving this time I would be able to identify the minimum time okay good so okay what's important here that I do this experiments only on probe trials so I don't let mouse to learn the mask I do it only on a subset more subset of trials on the intermediate concentrations so without mask mouse perform at the level of 90% it's a little bit higher at a higher concentration at low concentration now when I present the mask the performance goes like this at a low concentration so I cannot completely mask the order at about 100 plus millisecond the performance goes to the level of unmask order if I do it at a higher concentration the whole mask shifts to the left and this timing is approximately 13 millisecond that slightly less than what we see in the in the mitral cell responses with the shift of concentration but order of magnitude more or less the same so in the first maybe 100 millisecond is enough for us for the mouse to do concentration-dependent decision of the order and the shift can be explained by the concentration shift of the mask but where we are located with this 100 millisecond maybe during this 100 millisecond all receptors got activated that's not the case mitral cell group publish the result on set of calcium activation of the glomeruli from the from the sniffon set for many orders and many glomeruli and you see the distribution with the peak is about 100 millisecond and some of glomerulus activated much much late we are here on the distribution we activated this small portion of glomeruli about maybe 10 or 20 percent yes yeah say it again, above 50 percent yeah that's above 50 percent yeah that's the we cannot mask completely so basically there is some leak of information probably we don't activate all the receptors we don't know but there is some speculation where this may be so but definitely in 100 millisecond we don't activate all the glomeruli and only small portion of glomeruli is the element for this experiment what we do also on the unmasked trials we measure reaction time how fast mouse leak left or right and what we found that concentration there is some speed accuracy so if mouse leaks spontaneously early the performance chance level and then performance goes up but the two curves shift more or less on the same time on the same time shift so everything so if you think that first processing depends on the order delivery or processing and the next is included in the decision time these two interval are the same tells us that maybe processing is concentration invariant the timing from the if mask tells us when the enough information in the nose and the decision time tells us when the enough information when the mouse starts moving the tongue so these intervals are the same tells us that processing is concentration invariant yes yes yes yes I agree I don't think that I will shift glomeruli activity at 300 milliseconds into the window of 100 milliseconds I'm not claiming any absolute numbers here but you right and I'm actually a very important point I don't know how to measure actual actual number glomeruli this is the but I'm sure that it's not there so it's not there so that's my point thank you thank you very much so what else we did what happens if we significantly increase the complexity of the task and we asked the master perform 6040 60 make sure we do exactly the same thing with we with trouble concentration and master from a little bit slowly you need a little bit longer mask but this timing also maybe 120 millisecond it still does not cover all the range of the glomeruli so we still located in this corner so this is kind of the main message the temporal window is relatively small we can argue it's 100 millisecond is 100 millisecond but it's smaller it's much smaller than all the activity in the bound and that brings us to the notion what we measured before we measure activity of many many mitral cells this is recording from one mitral cells in response to four orders the mitral cells activity is locked to the same cycle some of them excited some of them inhibitory this is multiple in mitral cells to one order and on and on and on so we collected all the data in the following graph so we recorded 460 cell order pairs third was excited third was inhibitory so it didn't respond to the order and we plot the activity of individual mitral cells it's cell order pairs and in the color diagram so red is excitation blue is inhibition and we order them in the sequence of the of the peak it's only excitedly cells and it was it's a difference from the spontaneous rate so that's why you have a blue blue side so we recorded 460 cells and if you believe that we sample the cells absolutely randomly and orders also randomly this is strong assumption then for one order that is exposed to 50,000 mitral cells 15,000 will be excited to respond so it changes the axis so that is what the bulk tells the brain and one inhalation of a new order you have this activity and it's actually over 400 also almost 400 millisecond so that is the flux of information one sniff going to the brain and it's a lot of information you can encode way way more than trillion orders here sorry I'm using this just as a number that is kind of it's a lot of information but what I'm trying to say is that the information for the order identity only here and this is very very important what the rest of the sniff cycle I don't know and this is the main message so I'm very curious I would be very interested to discuss it but until we find the behavioral relevance of the latest of the sniff cycle I will consider it's kind of it's not used the beginning of the sniff cycle is important for the identification we don't know and here's my analogy so I actually want nobody to ask the question yet here's my analogy you know maybe the rest of the sniff cycle has nothing to do with the order coding so it has a lot of correlation with orders but it's not relevant for the behavior and my analogy is the following if you would be observing the tennis surf you can extract a lot of information about where the ball meets by the end of the trajectory well only the beginning matters so the trajectory can be very different than the ball will go in different directions this part of the trajectory has nothing to do with the direction of the ball and if you put here if you ablate the movement here the ball will still fly to the step direction yes we have a lot of cells we have a lot of cells so I don't know it's a good question but there's this I don't know how to do envelope type calculations like that because we have a lot of responsive cells so look first of all it may be not 4 it may be a whole 10 15 and Alex will comment on the numbers actually he has some way to estimate the numbers from the information perspective but while we are normally carrying information cells 30 cells has some temporary situation in this time window you have a lot of capacity hold on a second what you're talking about you're talking about information capacity in the other space or in the concentration space that is all depends on the inhibitory network so the idea is how quickly the inhibitory network kicks in so my idea that you build up the activity and the inhibitory network kicks in actually happens in the ball and if I have time I will talk about the ball with that but defining the temporal window is super important so let me conclude the first story so what we're proposing is a primacy coding model that is a concentration variant we can use for forming the code we can use for reading the code some code is consistent with non-behavioral phenomena and small temporal window is relevant for other identification so here's a question that you may want to ask so if you want to ask the question that's the time you can just tell the number so how this model is different from latency coding or how this concentration encoded or the rest of the sniff cycle encode intensity concentration so is a primacy coding model consistent with the animal ability to extract order from the mixture is an animal using information in the rest of the sniff cycle or can it use whatever so this is the kind of question that I got during my presentation if you're interested in any of this I would be very happy to answer this question and what is the role of Fakti Babu network 5 ok what is the role of Fakti Babu network ok we change the pace anybody is interested in any of these questions please formulate but it's basically I can speculate on that and we can discuss it but I wanted to kind of see the what the most interesting here absolutely absolutely and so I like the exceptions because when the model fails we can learn much more so whatever exceptions you can bring in I would be very happy absolutely but it's a more general phenomena and you may ask when we perceive the wine the percept stuff stays along and we sense additional notes later in the sniff cycle of course it's all true it's a much more primitive model to describe all this phenomena and I'm super interesting how the model can be expanded for the more complex complex stuff so I it's an interesting question so as I understand that except the question 7 there were no questions more so how is the model different from the latency coding model the basic difference between these two models in their predictions we can both of them rely on the timing and one is saying basically the sequence tells us about other identity and you cannot derive the predictions that we derived from the primacy model that's the main difference that the temporal window number of timing is important we don't know if the sequence of the initial granular matters we don't know it maybe it's true maybe it's not my feeling that sequence of the first granular matters but the main difference is even not in that it's a redoubt model and it's small how the code can be read and it's model that has many many predictions and Alex actually builds up the theory based on this very simple assumption that other identity can define by small number of orders one interesting point that I wanted to make is the following I always was puzzled by the notion about the diversity of factors receptors if order is encoded by 100 or other receptors then if you knock out one you would not change any difference in the percept so what keeps the order what keeps the genome more or less stable if the order is defined by small number of receptors then the role of individual receptors is much higher actually it has been shown already that knocking out one receptor tar-4 it was shown by Michael it's actually make a behavioral phenotype so each receptor counts that means that each receptor is important for the coding if there is a multiple combinatorial coding if it's multiple by any any I'm recording from Michael's cell from TARS I'm recording by any physiological difference we cannot find the difference projections maybe somebody can comment in the cortex and OAN the property of the glomerally it's not OR but it's we don't see the difference from the from the circuitry perspective so in my opinion it's as an order as a receptor as other but the role of individual receptors is much much higher on the primacy model than any other model that has been proposed I agree with you it's not OR I know but it's the front end is not OR but the rest we don't see the difference yes but it's respond to other orders and it's activated as combinatorial features it's has timing the glomeral is exactly the same thing I don't see the difference from molecular biology perspective I completely agree with you and it's very nice model to study because we know the the ligands okay so how much time I have so I will just mention briefly so we try to with up to genetics we try to actually simulate the code and simulate the we try to create a special temporal cord with up to genetic pattern simulation find which features are important and by manipulating the artificial patterns but the work is still in progress and probably I will skip this part and go to what Andreas asked because nobody else asked so I'll go to that so the story is the following so actually so how much of the process this information in an ideal situation what we want to do we want to present an order no identity of each excited receptor and measure from the mitral cell that is connected to the first to the second to the third and so on so forth and see what's the differences but that's it's very hard experiment we don't know how to do the main problem how to preserve the identity how to know the identity of mitral cells glomerally it's just that's the hardest problem so what we decided to do we decided to switch the problem not for one other mini glomerular for one glomerular mini-order it's a little bit tricky interpretation but stay with me I try to make a case so what we did we focused on M72 glomerulus and we have a general adoption in this glomerulus so with what we can do we identify the receptors and the glomerulus and we can identify mitral cells connected to the glomerulus so we're working with one channel of information first thing what we did actually Tom Lab did they measured the responses of the receptors and recorded activity to the bunch of orders and this is at CETA Finan sorry I will be using just abbreviation and he took the concentration series and recorded these receptors and we have the whole battery of orders two heterocets at CETA Finan it tilt a glade the height I don't remember all of them I'm very bad with chemistry sorry but on the concentration axis they have different EC50 so we sequenced them in the activation profile the next thing we did and I always will be presenting two heterocets at CETA Finan the only thing you need to remember this is the strongest ligand so they activate the receptor in the lowest concentration next thing what we did we also conformed via imaging we created the mouse that has GFP in all receptors and RFP in M72 glomeruli and so we can sorry and we can sorry we can identify the response of this glomeruli so we can conform it in Viva so we put electrode in the mitral cell layer and we blasted this is the work of three students and two postdocs Nils now in the he left NYU to DeepMind and is a killer near the now in UCSD so we we take the mouse that has general adoption M72 glomeruli and we blast the light on the glomeruli and we find the responses with a short latency so the mitral cell responding to the light with short latency we will call light cells the one that will respond to the light versus all other cells either M72 cells or light cells this is a cell that connected to M72 glomeruli and Nils did a lot of analysis and this is sorry for not showing graspers because we have a lot of variability in the sniff cycle he removed the variability of the sniff cycle and this is typical response of three light cells or M72 cells to a bunch of odors so the typical response like that this is exactly the response the black line is the spontaneous ray the blue line is the response to the odor that in this case it was 4-meteor, etc. and the same cell response to the Zaldehyde was inhibited the black is inhalation no no no this is one sniff this is about 100 it all has been put in the sniff coordinate so black thick line is inhalation and the rest of the sniff so it's about 400 per second so it's all locked to the sniff side now at the same time we recorded activity for so this is all M72 cells collected together from many many it's more than three this is three typical but we also recorded not M72 cells but a lot of diversity of the responses ok so what you can see here that everything but the very right column it's quite a mess and here you see some dense stuff so this is a special case this cell responded only initially inhibited but this cell responded in different way so let me quantify this this result all the response can be initially inhibited and what we did for the beginning we just counted the number of inhibited responses for M72 cells and not M72 cells for given order exactly so green is excited blue is inhibited so at set if and on M72 cells have approximately third inhibited and third no responses all other cells recorded the same electrode but didn't respond to the light so we don't we call it not M72 has more or less the same distribution we have much more cells like that but the distribution was more or less the same ok this is a generic ligand at set if and on what happens with other ligands we see all of them so left column is M72 right column is the generic cell all of them are very very similar to each other except to hydroxyl set if and on the strongest ligand has a very different structure has much more excited responses ok now what happens with the latency of the responses the latency of the both actually excited and inhibited responses this is a cumulative latency for M72 and others they are more or less identical if you measure the time of deviation of the response from the background it goes more or less you know the distribution is more or less the same for all others except to hydroxyl set if and on to hydroxyl set if and on excited vital ligand cells or M72 cells early in average and we checked maybe this effect can be explained just by the level of activation of the glomerulus maybe to hydroxyl set if and on activated glomerulus stronger than others we actually did this imaging and we see that yeah it fluctuates this is low activation but it's nothing special to hydroxyl set if and on level like other orders so it cannot be explained by amplitude of activation of the glomerulus yes say it again I sorry I didn't understand the question yeah possible um ok let's I have an answer to this question thank you ok I'll answer in the next it's not in the amplitude effect yeah I agree that may be the case ok yeah that will be your decision that will be your decision I just give you the data and you decide what is good what is bad it depends on the experiment um so ok maybe this effect of the concentration maybe effective concentration of two hydroxyl set if and on is significantly higher than any others again we don't have infinite way of presenting concentration but we take two two orders two hydroxyl set if and on and main phone that is medium level ligand and we shift concentration of menton down and up and from two hydroxyl set if and on down and compare the responses and actually nothing change so again we always compare to the to the genetic population two hydroxyl set if and on behave in the same way independently on the concentration and menton behave in the same way independently on the concentration you you have less less differences but still well maybe let's look at the individual cells that was a big surprise for us look at not on the population but the individual cells yes yes no this is taking on this concentration yes this is actually strong correspond so we actually correspond to these concentrations exactly so it's not explaining so it's not saturated you see it's not saturated but the behavior the same this is what has been done in tom lab I I don't want to make state no no no it's actually was a little bit harder they moved the window on the site you know the M-72 so I apologize I should put the it's they cut the part of this they basically try to show the activity in this area it's not whole bulb it's probably on the level few tenths of the glomule sorry that's a good point I will put the M-72 we we know this is this is an arrow here this is M-72 this is M-72 that's the bar that shows the intensity of M-72 the bar nearby is just intensity of M-72 well this glomule is not saturated the second concentration exactly that's the most point maybe we saturate no we don't and we don't here so but the response is the same it's not an amplitude effect it's not a saturation effect well we decided to look at the individual self and we trace one cell for three concentration and if you increase concentration for two heteroxet and on the cell behavior in a very consistent way they shift their responses towards the beginning of this cycle now if you do it for menton it's a mess the responses disappear with higher concentration sometimes they flip the sign and the distribution the following so the number of flipped signs responses half drop responses and the consistent contribute very small portion of the responses versus most of the responses for M-72 is either consistent like this or one drop there were no inhibitors responses very very different behavior absolutely first of all my take on it this is all extraneous information we don't need so much information second thing what I'm trying to say the distribution of the latency for the strong ligand is much more narrow than for the weak ligand that's all what I'm saying but let's I will give the explanation for this result in a few minutes I cannot describe actually we recently did we measured the correlation with the depths of the recording and we found no correlation I cannot say exactly we didn't analyze relatively to flip not flip but the depth the kind of the distributions the latency we didn't find any significant correlations and we look carefully but we didn't look for example if that is correlated maybe I can imagine tafted as more consistent than my talk quite possible but I don't want to do this because it will be too speculative without identifying yourselves so how can we explain this result this result can be explained if we go back to our original idea and see glomerular for one order not one glomerular many orders one order many glomerular and when one order is presented glomerular activated in the following sequence for one order yellow green blue as this is more sensitive this is less sensitive and they will be activated in this sequence independently of the concentration so this glomerular project to mitral cells there's a lot of inhibition inhibiting network and I'm not I don't want to speculate which specific inhibition plays here all but what happening and I just this is a cartoon it may be PG cells it may be whatever cells whatever favorable cells but what happening then when the signals comes yellow glomerular activated first and create more or less stereotypic early response I don't want even to call it synchronous it's stereotypic and early synchronous means actually they fire synchronously and I could not measure it on the same experiment but there's a good chance that they kind of create an activity together when the green glomerular got activated it's actually got active so mitral cells green mitral cells got activated by this glomerular but they also feel they're very heterogeneous inhibitory network so they result their signals are scrambled some of them inhibitory some of them excitatory some of them delete some have first inhibitory then excitory and on and on and on and so the blue glomerular is same so basically inhibition in the bulb plays a very interesting role it lets the primary glomerules go through the signal by synchronous activity of the all mitral cells mitral and tafted maybe mitral cells go very early in synchronous with tafted it's one of the model but this is a channel of information that goes go get green light to the cortex and what Kevin sees this is the important signal the cortex in use the first signal the rest it's actually way expensive to suppress mitral cells you need a lot of inhibition much easier is to scramble them you just basically ruin symphony you still have all dynamics you still have all activity you know about the bulb but it's kind of scramble you can extract information yes you can but maybe it's irrelevant so mitral and tafted cells connected to primary glomerules have serious typical responses and potential of inhibition is to scramble not relevant signals kind of it's a temporal temporal yes so what I'm trying to say is a following let's say we have it shouldn't be the first one it should be one of the first we have 10 primary glomeruli and the 10 primary glomeruli activate let's say 200 mitral cells just random now these 200 mitral cells create the first wave of activation they're more or less synchronous and they'll wire specifically the wiring between these cells is super important I don't think that the wiring to the cortex is random I think that primary glomeruli for given order has a preferential wiring wiring for different mitral cells is different so now these mitral cells may be wired with another one but they're not in a different connection so when you scramble this this ensemble this motif is not activated that's what my idea of course you have a lot of activation but not all of them work together that's all what I'm trying to say I don't know it's a preposition it's a speculative model I cannot make the proof the only thing what we're trying to achieve now and we get the first result sorry I didn't have time to put it on there we did exactly the same thing with the optogenetic pattern simulation we find the mitral cells by optogenetic pattern simulation and we play around and we see if you present the glomeruli early you scramble this one but if you present the second one inhibitor later you don't attack it obviously yes I was waiting for this question so how concentration is encoded look so what is the dynamic range of concentration 6-order of magnitude what is the dynamic what is just noticeable difference maybe a few percent how many bits you need to encode concentration 20 bits if you look at the signal you can convey 20 bits so many different ways you can encode concentration in the latency of the first responses what Kevin for example Frank saw that concentration increased synchrony between cells in the cortex I'm finishing synchrony maybe by more synchronous activity you get more you increase the number of active glomeruli you may get more activity and intensity of first individual glomerulus also may depend on the concentration so you have so many ways to encode intensity we actually published a paper about how many different signals can single glomerulus convey and a single glomerulus can be timing, amplitude and the presence of the glomerulus in the ensemble all of the skews can convey concentration all of the skews are correlated with concentration the question is which out of these skews are actually animals using for making the perceptual judgments this is a big question I doubt it is here I think that animal can judge concentration very, very quickly but I can only speculate so we kind of trying to solve this and anybody has idea how to solve that I'm very open to discussion oops yeah said again the problem is that it's very hard to do it with mice what's interesting here that we did explain to a new stimulated glomerulus here and mouse can detect there's no problem so what I like to tell this is the usual span of attention this is a temporal phobia this is a part of the signal that mouse paying attention by default the rest is your is your peripheral vision if I allow this analogy I can use it you may extract additional feature it may be useful but in other terms it defy the face of 48 and my main notion that we kind of we assume it for granted that in vision the majority of the spikes from the retina completely disappeared you don't the cortex has no idea what's going on we somehow assume that every glomeral counts I propose that not so you get the input from your peripheral vision you don't know what color of the t-shirt Alex is wearing if you look at me but these photons are on your retina it's the same thing you do get this input and it actually reaches the cortex so that's fair enough but it's relevant for the defecation of my face the photo of the K shooting to your peripheral vision from your neighbor I might take on it and in order to understand other coding we need to start thinking in these terms well thank you very much the work has been done the experimental work has been done by Chris Wilson this is the masking strand Edmund Chung did the experiment with the up-to-genetic pattern simulation Zinke Arneada-Christine worked with the light of cells I'm 72-glamorous Romer collected the initial data he is now he was with me in Janelia my collaborators, while Neil Rebinovich analyzed a lot of data for the light of cells project Tom Bozer and his group from the US and funding from different sources thank you very much