 Florian Berger, our speaker today, who gives the talk from Utrecht in Netherlands. Florian now is head of the theoretical biophysics laboratory in the University of Utrecht. He is working on biophysical modeling, various processes, and in particular he is interested. A big part of his research is devoted to studies of molecular motors. Recently he also started to apply neural network techniques in his research. Yes, so today's talks is about molecular motors and I welcome Florian for you to speak. Yeah, thank you. Thank you, Oman. I will share my screen. Do you see this? Yes. Okay, thank you. Yeah, thank you for organizing the seminar and giving the opportunity to me to introduce you a little bit to our efforts to understand active processes in biology that are driven by molecular motors. And as far as I understand, the audience here is quite diverse, probably with different backgrounds in physics, mathematics, maybe also biology, and also on different career levels, I guess. So there are maybe students, also professors. It's always a little bit challenging if you do these virtual seminars because you don't really see the audience and you don't, you also don't really see if they get bored or if they don't understand something. So please, if there is something you want to know further or if there's something not clear, raise your hand if that's possible in Zoom or just shout or ask me. So because the audience is very diverse, I thought also it would be nice to have a like a little bit of broader introduction that also is always good for me to motivate myself why we are trying actually to introduce these quantitative approaches to understand biological systems. And then after this little broader introduction, I will give you two examples how we think or how we study these molecular motors and how they drive active processes. So one remarkable feature of, I would say, almost all life forms is motion or activity. And already Schrödinger said that living matter evades the decay to equilibrium. And it is doing so by these remarkable active processes. So what you see here, you see two videos. So in the first video here, you see in red, a T cell that attacks a pathogen cell in blue. And here in the lower video, you see cargo transport in a neuron. So these are vesicles that are transported actively by molecular motors. So what we are, what my lab is trying to understand is to develop and quantitative by a physical understanding of these active processes. And the next step, of course, is to define a little bit better what do we mean by a quantitative, biophysical understanding. So for this, we can go very back in history and time and realize that actually physics comes from philosophy, right? And in the beginning, most physicists were philosophers and they started to describe what they saw in nature just by words. And here I give an example for such a physical law that is described by only by words. So any solid lighter than a fluid, well, if placed in the fluid, be so far immersed that the weight of the solid will be equal to the weight of the fluid displaced. So this rather basic physical principle, you can, of course, formulate in words, but you can also formulate it in mathematical equations. And this is what we are trained in or what is more common in our field now to phrase these laws in mathematical equations. And the advantage of these mathematical equations, of course, are that they are very precise and you cannot argue much about it. Because I would say with words, it's a little bit easier to argue what they really mean. Was that a raising hand to someone in question? Okay. Sorry. No, I haven't seen any questions. Because there was something, there was a beep on my Yes, I'm going to many tour of the questions. But if you want, you can also close chats in your screen, which is shared by the way. Yeah, that's I see. Then I can also, sorry. Okay, now it's maybe better. Okay. So going from these description and words to description equations, I would say the really the fathers of modern science were Galileo and Kepler because they really started to connect natural phenomena with mathematical equations. And also as Galileo stated that the book of nature is written in the language of mathematics. And this actually gives us an opportunity to measure things. And then we can compare things. And we can also make very precise predictions. And then we can see if experiments actually fit our, our models or our understanding of the system. And as Bill Bialek actually introduced these concepts of the wiggle room, which I really like because he says, only through these quantitative understanding, we can reduce the wiggle room. So there's not much room for arguing about phenomena anymore. But on the other hand, it's interesting when you think about these biological system that somehow they have escaped that kind of understanding in the last, in the last years. And this is actually what we try to develop. So my personal opinion on this is why these biological systems somehow escaped this kind of understanding that we developed in physics is because they are poised at a very interesting and interesting position if you think about the complexity of the system. So this goes back to Weaver from a very interesting paper he wrote. So there are different classes of complexity and physical systems. And before the 1900s, hundreds in physics, the most systems, you can describe as problems of simplicity, because it was only a few degrees of freedom. So here I sketch the, as an example, this, this pool table with only one ball on it. So if you exactly know the initial conditions, you can calculate the trajectories. And then on the other hand, after the 1900s, in physics, in thermodynamics, and statistical physics, researchers concentrated on on systems, which we were called disorganized complexity. So this is a complexity where you have a lot of degrees of freedom, but somehow they are kind of similar. And then you can use statistics to analyze them. And living systems are actually poised somewhere in between. So they show this kind of organized complexity. So this means they have a maybe hundreds or thousands of degrees of freedom. And it's very complicated to find a unifying descriptions for them. And often that right now, I think often what what the only thing we can do often is starting doing simulations and see if the models that we put in, if they make sense and if we can make consistent, consistent prediction with experimental data. So there's this other field in physics, where people try to underpin or developed a lot of theory and models, how actually order comes about or order is derived from from active processes. And this goes back, of course, to free regime. So there are two examples that I illustrate here. So one is the Lusov-Cherwood-Zinsky reaction and the Rayleigh Van Aakom action that you're probably familiar with. So the idea here is that you although you're far away from equilibrium, you see an order appearing on larger on larger scales. And this order is quite interesting because we see order in living systems. And these this order is often driven by active processes. So this brings us back to the two movies here. So we have the T cell on the top and here the cargo transport and neurons and the underlying molecules of these activity, these are always and are always molecules to transuse one kind of energy into the other energy, right, because this is kind of a typical example of an engine. And one big class of these molecules are molecular motors. And here's a movie how we think about these molecular motors. I will introduce them a little bit to you. I don't know if you have a background in molecular motors or if you know these molecules. So essentially, what these molecules do is they transuse chemical energy into mechanical energy by hydrolyzing ATP into ADP and inorganic phosphate. And by this energy transaction, they power intercellular transport. So they walk along these filaments in this step or step fashion. You see they walk along this way. And then they are also there are also other kind of motors they they rotate, or they do other things. So but they are usually the underlying proteins for activity. So what our group is mostly interesting is interested in is to understand if we can build up different or if we can, we can develop a quantitative description on different scales by building up on these molecules. So if we so the idea is if we understand how they work on a single molecule level, can we actually combine them to understand how they how they function in a collective small ensemble and then from the ensemble go up until we understand how they work in larger scale systems and try to organization and function themselves. But lately, I also thought or I started to think about that. I think the other approach is maybe also quite valuable to start to think about the top down approach where we start with a phenomenological description. So we start with an experiment and see what's going on there and then think about what are the control parameters that we can control, and then try to describe the system from this perspective and then match it back to the underlying molecular mechanisms. So we develop these description and base them on physical principles from non equilibrium physics and statistical thermodynamics also. But what I want to emphasize and what I will also talk about in the next 10 to 15 minutes is that I'm also very convinced that it's very important to develop tools and methods actually to analyze and interpret data. Because only if we have a better understanding how we can also connect experimental results from different groups and also from different molecules, then we can reduce this wiggle room. So we can really start to quantitatively analyzing biological systems. So this is one project I want to give you a little bit of background in this, how we buy or how we started to characterize the single molecules, these molecular motors. Again, what are the basic observations? So to understand what we are trying to develop here, it is quite easy if you just think about these motors that they walk along a filament, they can bind to this filament and they can unbind from this filament stochastically. And these processes of course depend on the force. So under which force these motors are, if you can think naively about it, if you rip the motors off the filaments, they will unbind very fast, of course, if they are under a high load. And also, you can think that probably the velocity will depend on the force. So researchers in the last 10, 15, or even longer now, probably from the starting, they started in the 90s actually to measuring, measure how these molecules react to forces. And the typical experiment is done with optical traps because with these optical traps, you can apply forces on the range of piconewtons and this is exactly what these molecules also produce. So the idea here is that you have a molecular motor here, it's a dining motor depicted with these two great donuts. And this motor attaches to this microtubule filament, and then it's connected to this polystyrene bead, which is then placed into an optical trap here, depicted with this orange laser beam. And now this beam actually stays constant. And the motor walks along this filament, by walking along this filament, it pulls this bead out of the trap center. And because of the potential that this optical trap creates the force on this bead increases, also the force on these two bonds, and then the motor will unbind from the filament. So there are two basic observations. Of course, one is that the velocity actually changes with the force, and you can assume that this is just linear. So this is usually the most simple or the easiest assumption you can do is you can say, okay, the force now decreases linearly. I can actually this year. So the force actually decreases linearly until it hits the stall force, and then the velocity is zero. And the other thing is that the unbinding rate increases exponentially with the force. And this is just a factor which is the characteristic force, and we call it detachment force. So this velocity, you can also just sketch it briefly. So this is the force. And this is the velocity. And then this just goes down linear until it hits the stall force here of the motor. And then it's then the motor stops. Of course, I mean, this function goes on like this. Okay, so here's a example of this kind of experiment of the setup. So what you have here is the optical trap is here, then you have the bead, you have the motor, which has a certain elasticity that we assume it's linear. So it's just a hook and spring with us kappa m. And then the motor here, this is a motor that can unbind with a certain rate, and it can work forward with a certain velocity, both of these quantities depend on the force. And then it's held in the optical trap. So the optical trap potential can be also modeled as a linear spring, which is here, the kappa team. So what you will see of kind of trajectories, when you do such an experiment is you see that the bead goes up. So the force actually increases on the bead. And then at some point, the motor actually detaches from the filament, and the bead snaps back to the center of the trap. And the motor is unbonded. So now you can repeat these kind of experiments and you can measure a whole distribution of these forces here. This is shown here. This is a real measurement for kinds of motors. So my question is that I started to ask you is if we say sorry question. Okay, no, no, no question. Right. So how is the macroscopic probability density functions or something that we can maybe fit here? How is this actually connected to these single motor properties? But the problem here is that this force actually, so the force that is applied on the motor in the trap is actually time dependent. And in these quantities, the unbinding rate is for a constant force here. So this is our constant. So the idea is how can we connect these things? So how can we derive this probability density function connected to the single motor properties? So what do we expect? So if we have a motor that is slow, and that unbinds fast, what we expect is actually that the forces that the motor creates so this depicted here. So it can only go a little bit forward and then it will unbind. So you will only measure small forces and maybe a probability distribution like this on the side. If you have a motor that is fast or that slowly or sorry fast motor or slow unbinding motors or the motor actually travels very fast forward and then or it does not really unbind so this means that it creates or generates a lot of force. So it will go up until it hits stall force where it cannot produce more force and then it will unbind. So the probability density of the measured unbinding forces in these kinds of experiments maybe look like this. So here you already see that these distributions they should somehow or the shape of the distribution should somehow depend on the on the molecular on the dynamics of the molecular motor. There is a question. So Raphael Petrosyan is asking why is force time dependent? The force is time dependent the force that you measure because what you measure here you in the experiment you measure the the distance of you measure the position of the beat and this motor works forward with the velocity. So this actually this delta x changes in time. Is that clear? I believe so. So in other words the force depends on time through the position of the motor which is exactly right. And then usually what you have you know the you know the you know the stiffness of the trap and then what you measure is the position of the motor and yeah on on time because it works forward and this is the force that you that you then measure. I think it's clear. Okay I have another question sorry hi Florian. So my question is so there are different experimental setups in which in some of them you are not able to measure the force but only the scatter light intensity. So for example in in our experiments in the past we we were incurring the force from a previous calibration of the trap in which the the beat was always in the linear regime. So we were saying okay if the particle is the beat you see next we can infer the force by a simple Hooke's law. However the trace that you're showing you're you're plotting directly the force. So you will be able to measure the force down on the beat directly or is it an indirect measurement based on a calibration of the trap? Yeah thank you for pointing it out it's actually yeah so this is this is actually what we measure is x and then we calibrated with a with a with a stiffness of the trap because we know we can also measure that it's in the linear regime the trap. Okay so what is actually measured is the distance actually what is measured is actually the distance of the beat from the center of the trap. No because I know that the beat is going sometimes you can also move out of the linear regime of the trap. Yeah that can happen yeah I think but usually yeah and there's a other setup I will talk about this maybe later yeah I will briefly mention it where you actually you can also use a feedback system and there you can apply a constant force on the motor by always adjusting the position of the trap while the motor walks. Okay thank you. It also works. Okay so now I will come because I thought there are also some people who like equation in the audience so I thought I'd bring some equations here so now what we actually do is we combine or we derive this probability density for these unbinding forces and this is related to a binding probability here and then we assume that this binding probability or the time derivative of the binding probability can be assumed to be a first order equation with this unbinding rate here as a function of force and then if we combine these two equations we actually end up with this one which gives the unbinding rate as a function of this probability density and then we can also invert this and solve for the probability density it's actually not too complicated the whole thing but here you again already see something very interesting I found is actually that here this equation always depends on the unbinding rate divided by this loading loading rate and the same here so it actually depends only on the ratio so that also limits actually then our inference that we want to do because from this for measuring this probability we can only get the ratio of these two quantities right so here for very simple for the very simple system that I introduced earlier where the unbinding rate depends exponentially on the force and where the velocity depends linearly on the force and with these linear springs here which connects actually this gives how the force is actually transduced here or i would say i should say transmitted sorry not transuse so the how the force is transmitted onto these bonds you can readily write this down and then you can use these equations and plug them in here and then as i said before because this is a ratio you will get a ratio of these two quantities in this expression which exactly gives you the run length so if you think about it it's it's it's not um i mean i did the calculation and afterwards i thought about it but i could have thought about it earlier that this should not give you any actually any information about the dynamics because somehow their time is not relevant anymore here right because you you measure the the forces and not about nothing about the dynamics may ask but you have a time derivative of the force inside so this thing is as you actually anticipated in your examples depends on the protocol exactly it depends on the protocol here yeah that's right but this equation does not depend on the yeah yeah okay all yes thanks for pointing this out all what i wanted to say here is that i cannot determine either the velocity or the unbinding rate from this equation i can only determine the ratio of both from this one right yeah okay so and then you can do this actually we use this experimental data here and then we fitted this simple model to it and fits the data very well because here probably you underestimate in your experiment because these are very small forces here so it's in the noise more or less but if you if you use this equation to fit the experiment data what you get out is actually that the stall force is almost 15 piconewtons then here this is the run length this x0 is about 500 nanometers and this detachment force is 2.2 and then if you um so i i compare this to to to results from the literature and then you see okay we are a way a factor of two or a little bit more here also a factor of two but the detachment force actually works quite well so yeah i think it's a it's a interesting way but probably there are too many fit parameters here for only this data set but then what we use this equation in the next step which i think is quite interesting is actually to to determine now the mean because we have the distribution so we can also calculate the mean unbinding force here and we can excuse me florian there is again a question what was the form for probability of the force f the functional form yes i believe the expression maybe do you have anywhere the yeah what is explicit expression for f oh that's a long expression i didn't put it on the slide because it's too long all right i mean but you can actually solve it analytically because in this in this case you can you can solve it here it involves some um do you have a paper where this result has been yeah it's published i give the reference in the summary all right very good so yeah but you can actually do here if you calculate the mean force and you assume that the unbinding rate is actually constant it's not it is not dependent on the force anymore this is a crude assumption that you can make and then you can derive actually analytically a very simple equation so for me that was quite shocking because you have complicated integrals and then you have this very simple equation here in the end and this is now very useful because you can use this now in experiments because this equation has one parameter here which is the um the the stiffness of the trap and the stiffness of the trap you can change in your experiments and then from a fit of this equation as a function of the stiffness of the trap you can infer in principle the stall force and the run length and this is this is quite powerful now because there are actually molecular motors human dining motors they are very low processives this means if you have them in an optical trap like this what you always measure are very low forces and it's very difficult to determine the stall force of this model because it almost never reaches the stall force in the experiment because they unbind before they generate large forces so we did this with our experimental data with on again a rich together and what you see nicely here in this in this graph so the green line is actually the force measurement you see that we can almost perfectly fit this result with these two parameters here with the with the stall force and the run length and this is these are the results from the fit which are very good and because we because we still needed to do this very crude assumption that the unbinding rate is actually constant we did another measurement with a force feedback trap so with the force feedback trap what you can do is you can apply a certain constant force on the motor so the trap actually all in this kind of experiment what you measure or what you have you you you use a feedback to adjust the distance between the bead and the trap center so while the bead is pulled by the motor you adjust the trap so you also went on a piezo stage you actually you actually keep the distance between the bead and the center of the trap constant and because you can help you can keep this constant you exactly know the force that is exerted on the motor and then from this you can also measure the the stall force and the singer run length which is a very good treatment with the experiments so in this part i i wanted to show you a little bit that we are trying to develop these these concepts to reduce the bigger room and to introduce frameworks to understand these motors better and to describe them and i want to acknowledge here the collaborators from which i got the data so one is Arne Genelich i have a long collaboration along yeah long and very productive years with him and then the other collaborators Paul seven from the University of Illinois and here are the publications of here there's also this quite cumbersome expression for this distribution of the forces this one is inside of the first one here excuse me there is a question in the chat yes so Matteo Marcily is asking i'm confused your theory gave the force that is a factor of two away from the experimental one but the feed works is the reason for this that's uh that's that's quite interesting but we used if that's also different data sets sorry maybe that was not clear so here this is a different data set this is for for kinasein one data it's a different motor protein and here we we did it for human dynein and the thing is for this experiment what you do what you have to do in the experiment you have to change the trap stiffness so you have to do one experiment at a certain trap stiffness measure a lot of these forces and then you determine the mean force here and then you change the trap stiffness you do a lot of measurements maybe 500 and then you determine the mean force again and this kind of data set is actually not available for kinasein for this one here because here i only have one distribution at a certain trap stiffness so in principle it would be super interesting if i can convince someone to measure but it's also a lot of work because you have to yeah you have to do a single molecule i hope this would answer the question is it all right yes thank you okay yes all right let's keep going okay and then the next last part of the seminar here i want to talk about a new collaboration now in ütrecht with Anna Akmanova and this is mostly work that has been done by a master student that i supervise on our gross and other students peter jahn and hücho and also in a collaboration with lukas kapitain and here we look actually at the role of molecular motors during t-cell activation so this is again this movie that i showed before so we have a t-cell here in red that attacks a peter jahn cell so a lot of things are going on during this process actually and here's a very simplified cartoon so you have this t-cell and it looks like this kind of and then it attacks this peter jahn cell so it's a yeah a cell that needs to be destroyed or your immune system decides that the cell is not a good cell and it needs to be destroyed so what is happening here is actually that you have in your in cells you have a cytoskeleton the other cytoskeleton here is is depicted in in in plek and consists of these microtubule fibers they are connected at the central zone which is here in red and it's also called a microtubule here's an mtoc it's a microtubule organization center so what happens is when the t-cell docks on this peter jahn it forms synapse this is this interface here and then sorry yeah this interface it forms the synapse and then this microtubule organization center is actually pulled over close to the synapse here to and and and this process is called polarization and this is done because then the secretary lysosomes can actually start eating up the other cells or attacking the this this other cell here yeah and we are mostly interesting how this polarization actually works and what are the molecular players during this polarization so in the experiment what they found in in the akmanova lab is that when they knocked out a certain molecular motor which is called kiff 21 b they found that the cells cannot polarize that efficiently anymore so here what you see is a side view i hope this is clear this is a side view of the cell so here they form this interface and in blue you see the nucleus and then here in purple is an actin so this is the cortex of the cell and then this green marker here it should be here or you can see it here better this is actually where the centrosome is so it means that the centrosome actually reload or and relocalized from behind the nucleus to the synapse and then if you do a knockout of this molecular motor this kiff 21 b then you don't see this anymore so the centrosome is actually above this synapse here and you can also quantify this say the here is a distance from the cover slip so this is the wild type here so it's very close to the cover slip and these are the knockout experiments so the idea is that in the absence of the kiff 21 b motor polarization the polarization is impaired so this what usually should be the normal process that the micro tube organization center is actually translocated to the synapse it's not happening anymore this motor is depleted so to understand a little bit what's going on here to understand this mechanism we started to do some simulations and we wanted to understand if we can actually reproduce this phenomenon with a with a agent-based simulation so we used the large event simulation package was which was developed by Francois Nadelic to to simulate these different processes that are going on so what are the key players of such a simulation so we have a we we need a shape of course that's the cell and we need a nucleus and microtubule organization center so this is all modeled with elastic interactions and then also some steric interactions but what is most important is now how the microtubules actually are modeled so here they are elastic fibers so they have a bending elasticity and then they can they can they can grow and then they can shrink so the growing and shrinking are different states of the fiber and there is a certain rate it's usually called a catastrophe rate it goes from the growing state into the shrinking one and then it can start growing again and then we have these kiff 21 b motors here so what's happening here this is a motor so the motors they can bind to the filament and then they walk along this filament and this is very interesting because it has been known before that these motors actually they interact with the tip of the microtubule they introduce a pause in this growing dynamic and then they introduce shrinking of the fiber and then the next key player that we needed is a cortical dynein so this is also a molecular motor that is embedded in the synapse here and when it when it's binding a microtubule because it walks in this direction but it's anchored here in the membrane or in the cortex the microtubule is actually pulled in this direction and this actually this creates or generates the force that in the end pulls the microtubule organization organization centered towards the synapse so here's a now a simulation that we run and one on the left the simulation here on the left side is the wild type and sorry that's actually wrong the wild type is here and this is the knockout so what you see in the simulation are these black fibers these are the microtubule fibers you see in red here is the microtubule organization center and now we introduce the synapse and the blue dots are the dynein motors and they start now pulling on the fibers and with this the microtubule organization center should actually move towards the synapse this is what you how can i start this again this is what you see here in the wild in the wild type and then in the the orange dots here in the wild type situation are the kiff 21b motors they interact with the microtubules and shrink them a little bit you see they can somehow regulate the length of this microtubule network of these microtubule fibers and the length of the microtubules fibers here are essential because if they are overgrown what we see here in the knockout then the cell got somehow stuck so the not the cell but the microtubule organization center got stuck behind the nucleus and the cell cannot efficiently polarize anymore and this is exactly what we see in the experiment you can now also quantify this a little bit better now with the with the simulations we can look at the mean microtubule length so this is just a transient so this is how we initialize the simulations and then after they reach a steady state here we can measure the microtubule length and assimilation and then we see if we add now these molecular motors that the length is actually decreasing so the motors they regulate the length of the microtubules and the microtubules need to have a certain length because if they are too long then the cell cannot polarize anymore and this is also what we see in the polarization time if we start measuring this polarization time so if we increase the number of these motors the polarization time actually decreases and this is quite interesting because we only need to add a few of these motors there's only a hand few of motors and this is also consistent with the with the experiments because in the experiments it's very hard to to stain the kiff 21b motors and this is probably because there are only a few of them in the cell so we can also look at the trajectories of these centrosomes so how do the centrosome actually polarizes and it's quite interesting because at first it fluctuates a lot behind the nucleus and then at some point it starts to repolarize so this is the distance from the microtubule organizations and the to the synapse and here it seems when we add the motors it just takes longer but it's not that the speed is different but it's more that it wiggles around and then it just stays in this metastable or in this stabilized state here on top for longer until it polarizes so we also quantified this a little bit better and look at the at the forces that are produced so we introduce here this factor which we call the force imbalance by just projecting the forces onto the synapse here on this plane and then we have either a force from this side or a force from this side and if this is what is r and so left and right and then if these forces are the same then this factor should be zero so this is happening here so you have on average you have the same force from this side of the nucleus and from this side of the nucleus so the system is stuck in this position and if we increase now the number of kiff 21b molecules this force balance is actually broken and it goes towards a value of one of course because it's normalized and then the center zone actually can translocate to the synapse and here it's actually an interesting system that i also want to explore a little bit further is that this is probably in fact of the small numbers of microtubules because if you have a lot of microtubules it's very likely that all of them are somehow bound and then you have this stuck and stall or restricted situation but if you only have a few microtubules it's more likely that you have an unbalance and this is what we quantified here actually so this is the total number of microtubules so if we increase sorry this is the total number of microtubules that are bound to to to dynein in the synapse at the synapse so this actually decreases with a number of kiff 21b molecules but this difference between left and right microtubules actually increases and this is probably only an effect because you have a small number of of microtubules and here you can also you look here at the trace which is quite interesting so the solid lines are the central zone distance and these dotted lines are the difference between the number of microtubules on the left on the right and sometimes you see actually these attempts where the system actually tries to polarize but it cannot finish it and then so it goes actually back into the stalled position and then here it seems that if this so if the dotted line actually reaches a certain threshold then a polarization can start and the cell polarizes yeah so with this I already want to summarize so what I'm trying to do or what we try to do is to to develop these quantitative understanding by building a biophysical model between scales so what I told you today about was a project where we where we are looking at these single molecule behaviors of these molecular motors and how we can understand data from optical trap experiments and then I also showed you a little bit an approach where you where you jump into the middle so you start with the whole process which is very complicated active process in the cell and then you try to recapitulate what's going on with the simulations by putting in these different agents and how they react but I think in the end so what the long-term goal would be of course to connect both sides here in the middle and yeah that would be actually quite quite cool and then also go to actually larger systems here and think about transport the many neurons and axons and with this I also want to advertise that if you are interested in these things come and work with us because we are quite an interesting team of cell biologists and also physicists here so what we are trying to understand are these transport systems and here what you see is actually a super nice rendering of one of these t-cells the microtubal network and now we want to connect these experimental data of course to biophysical models and I think there's a big need actually to bridge this gap between images and videos and models and I think Oman also mentioned it in the beginning a little bit that I started to investigate how we can bridge this gap with machine learning techniques and also deep learning techniques so if you're interested working with us then please contact me yeah so thank you for your attention and then I of course would like to answer all kinds of questions if there are any thank you thank you Florian thank you I invite anyone from that in this list to ask questions the problem should I stop sharing let me see well you can stop and return upon request if necessary because then I see the chat okay okay so there is no question in the chart but if anyone wants to connect audio and ask please go ahead okay well while we're expecting any incoming questions I would like to ask actually but I was a bit curious about when you mentioned that you started applying this machine learning techniques could you unless there is any information sensitive information to release can you describe what you're trying to study with machine learning in these projects yeah so I mean it's quite complicated actually because machine learning I mean I probably also discussed it with you it's as always what do you want to do with it right and I think for some things we have very specific tasks actually so maybe I can switch back here to this slide so of course one problem is what I think machine learning techniques can be very useful is just to enhance the quality of images and videos because this is nothing else than a non-linear filter for it right so there's always a and then this has been also developed in Dresden at the Max Planck Institute so there's there's always a trade-off when you do these fluorescence microscopy experiments that if you have a if you want to have high resolution you need a high laser intensity but this induces light damage and also photo bleaching so this means you can only record for a few minutes so the idea is that you acquire a data set with high laser intensity so you have a very good signal-to-noise ratio but you do this only for maybe one or two seconds and then you switch to a lower laser intensity and then you record the rest of your experiments with a very low signal-to-noise ratio and what you can do is now you can use the data set that you acquired in the beginning with a high laser you can use this to train a neural network to match the high laser intensity images to the low laser intensity images I see yes yeah so this is nothing else than just training a non-linear filter right you have you have a lot of parameters and you have a non-linear filter well the specialist would know I'm not yeah I think that's it's super cool to do this because it works very well actually and then the other problem is if you if you look and at this other video where you see this micro-tubal network here and tracing is of course interesting because you want to trace single micro-tubals and you want to understand how many of them are there what are the crossings so there's of course we have now people in the lab they actually do this by hand in a virtual reality where you trace these micro-tubals in a 3d system and we are trying to understand if we can use actually also these annotated by humans annotated data sets to train a network to trace and detect certain objects in these microscopy images so this is a little bit what we're trying but it's still still work in progress and maybe I can come back in a few years and give another talk about it all right very good thank you do we have incoming questions may I ask a question yes of course yeah back back to that question about the time dependence of the force so time dependence of the force you meant like not the unbinding force depends on time but just just the force depends on time is that right yes so unbinding force it does not depend on time because it's just like the like force yeah this is how you define it but here the unbinding yeah so the unbinding force what I define it as the force that you measure under which the water like when the unbinding happens yeah exactly this is stochastic variable yeah so on the force depends on time because you pull and also because the motor moves yeah so you don't really pull but the motor pulls itself yeah because the motor moves and then it increases it's a little bit like if you are fixed on a wall with a rubber band and you would now step away from the wall then the force that you feel increases while you step and then you will slow down probably and then at some point you cannot walk anymore this is the stall force yeah I don't know if you know these rubber bands from gyms you know where you do like no this is clear yeah I was confused because I was thinking the unbinding force depends on time and the mean force expression that you showed this this is the okay yeah so this is mean unbinding force exactly and and it is calculated from the this force probability density function is it exactly so you take this big function and then you integrate but the problem is for this you need to plug in some unbinding rate some assumption of the unbinding rate and also of this force acceleration and I could I mean I'm not the best mathematician so I could only solve this under the assumption that the unbinding rate is actually constant I mean I would have liked to solve it for an exponentially increasing unbinding rate because this is what we think is actually going on in the system yeah I mean this exponentially in like force dependence of rate was like Bell's model and this but I mean maybe you can solve it I couldn't but it would be actually nice to see so what I did I did the crucial assumption that I actually approximated an exponent with a constant okay thank you yeah okay thank you there is one more question from the chat Luciano Bruno is asking so saying thanks for the talk and how do the forces modify in the you can read all right okay thanks for the talk how do forces modify in the presence of many motors on the bead and if the motors are able to diffuse on the surface of the vesicle that's a very good question and I I actually I always wanted to sit down and do the calculation but I don't know I mean it would be very valuable because I know that people are doing these kind of experiments probably you also know about it so because one problem with this it's not a problem but one disadvantage of the single molecule experiments is that you need to know that you're under single molecule conditions so this is very time consuming actually so what is quite easy is actually that your code beats with motors and you don't know how many motors they are on average you know there's many motors and then you would do the same experiments so I thought about it actually but one one other there's another technical problem is that if you have a lot of motors they can also create larger forces and then the problem can be that you get out of the linear regime of the trap but it's actually a quite this is a quite interesting problem also with the diffusion on the on the vesicle because the diffusion on of the motors are harder as far as I understand if this is correct um or this is what what you mean with this question is that the motors actually diffusively anchored on the cargo right so it diffuses in the membrane of the of the vesicle and this will probably effectively increase the binding rate because it can find the microtubule better because it has a certain range that it or a certain space that it can can explore yeah it's a quite it's quite interesting I think there's a lot of potential just to do the calculations I don't know I hope the answer the answers yeah a question to the best of your ability now all right we may accept any other question we have three minutes still of Florence time yeah I can also if someone wants to talk a little bit or chat I mean a little bit I don't know if this is a nice stream to you on youtube but well this will be a problem I understand to the to the channel of our section and I can also stay a little bit longer here and then we can have also some all right one last question from me maybe this is a superficial resemblance but sort of I see that you have you can apply different protocols to these experiments at least the first experiments where you had different distribution of forces is it possible to find a way for jarsinski equality to be applied in this experiments and then what kind of for the jarsinski but I think I think it's interesting that in this case you probably may because you know the potential of the trap you probably could yeah could know what is the real free energy landscape but then the question is when the motors pulling with the arbitrary protocol does their statistics verify the work that is done verifies the jarsinski equality yeah I think that would be interesting and but I think the so so there are two things to this question so first with the protocol I think it would be super interesting because you can also not related to jarsinski but what you can do is in experiments you can change the the position of the optical trap so you can also ramp that you can you know drive it with a constant velocity or something so you can think about is there an optimal way of measuring biophysical quantities from of these motors by adjusting the trap you know this is this is one way and then with the jarsinski I think the problem there is that for the jarsinski relation but please correct me if I'm wrong you need two different states right or what you usually do is you measure the or you want to have the free energy difference between two different states yes that is correct yeah yeah and then it's not clear with this motor what is the what what are the different states so you can of course say okay when the motor runs into a stall maybe this is a state because it unbinds before and I don't know what would be maybe interesting is to think about the molecule here right a little bit more on a molecular scale what they did also with the dna unsipping and jarsinski that you think the motor provides a random protocol and you want to understand the elasticity of the motor or something you know you want to determine the free energy difference between extended state and more contracted state or something of the protein itself that would be maybe something that would work okay thank you one more question from the chat max moffi did assume that epsilon i'm running where it is constant you mean that it is just time independent yeah so usually the here usually it is here the first equation usually you assume it's it depends exponentially on the force this goes back to battle and also i mean you can probably also derive it from balzmann but what i assumed here to solve the i really assumed it's a constant number so it does not depend on anything it's just a number because otherwise i cannot solve this integral you see this is an this is here an integral over the exponent and the exponent has also an integral in it yeah this is maybe there's a way of doing it but i i just couldn't do it so it would be interesting if someone finds it find out now that looks actually like laplace transform if you introduce the explainer so you should look in the table of laplace transformers perhaps probably in there all right i hope this would answer the question do we it seemed to me that i heard someone's voice maybe uh no no i see no no one is asking any more questions but we also finished our time all right then i think florian for this very interesting talk indeed and uh in the chat there was someone who also complimented your talk all right you see the applause from zoom yeah thank you very much and if anyone is willing to continue conversation yeah so i wouldn't go right in the chat thank you very much yeah thank you very much stop this here and then we can go to uh thanks a lot yeah hello hello yeah oh it's good also to see some people i think it's a little bit what's my first time actually