 And it is known that different people show different immune responses to the same immunological perturbations, such as pollen or vaccines. For example, only a fraction of people is allergic to pollen, or some people respond well to vaccines, but others don't respond so well. So the question is, how different they are? How different are they? And why are they different? And the approach of population immunology that measures a lot of people and immune responses over the course of time can potentially answer this question. So it's an immediate question, because people say it's an immediate response to allergies. They're instantaneous, but immune response takes about three weeks, while it's immunological. These allergies are instantaneous, and we will have the first exposure. So do you have time to develop an immune reaction? Why is it called immune? I agree that these two types of responses are different, immunologically different, but they share some characteristics, because they are related to some genetic background related to... Maybe I can answer the general questions later. No, no, it's still in this answer, because it's confusing. It's not immunological. Yes, it is. It is immunological. It's a different immune pathway. It's still immunological. So it's an immune system which always reacts to anything, yeah? It's not learned. It's ready to go, or it has to be built, and then it goes. Thank you. So I'm interested in the approach of population immunology. This time we examined the cohort of 300 volunteers who cooperated with us and received seasonal influenza vaccine in 2011. The cohort consisted of 100 males and 200 females, age 32 to 66, so they are middle-aged people. And we used trivalent inactivated influenza vaccine containing hemagglutinin proteins of these virus stocks, so they are very usual ones. And we collected peripheral blood on four time occasions, namely before vaccination and one day after and one week after, or three months after vaccinations. And we analyzed these blood samples. And individual blood samples were split into two tubes. One tube was used for measuring six B cell markers like this. And the other tube was used for measuring T cell markers like this. So as you see, they are very general markers classifying the basic subsets of B cells and T cells. And in flow cytometry, each of these markers is measured for each single cell in the blood. So what we get is a point cloud of cells in a multidimensional space. So seven axes, seven dimensions for T cells and six dimensions in B cells. And usually in immunology studies, we select only two axes and show two dimensional plots like this. But please note that this is only a fraction of a much bigger picture. So to analyze these multidimensional space, we devised a new method called lavender. Lavender is intended to uncover the latent axes that can explain the variability of the dataset. Suppose we have four samples showing these kinds of immunological distributions. And looking at these pictures, you might think that top two pictures might be somewhat similar because they belong to the same participant number three. But they might be a little bit different from the bottom two pictures which belong to the participant number six. So we want to quantify these differences. And for that, we follow the four steps. In step one, we perform density estimation of this point clouds. So these raw data are very complicated. So we determine a grid and calculate the density of cells using K-nearest events method. That amounts to basically smoothing the distribution. And this method is nonparametric. That means it doesn't presuppose any distributions, be it normal or Poisson or uniform. So it can deal with any complex shape of point clouds in an unbiased manner. And in step two, we measure distances between distributions using the concept of Kaubeck-Leibler Diabetes in information science. And we give it a little trick to make it a genuine bona fide metric. And we use, actually we use the distance. And in that nutshell, it tends to emphasize the accumulated peaks and focus on accumulated peaks and try to measure the intensity or the positions of these accumulated peaks. But other distances, if you like, can be used. And in step three, based on the distances we just measured, we reconstruct samples in a new coordinate space that we call lavender space. That uses the algorithm of multidimensional scaling. And like this. So the distances between points reflect the distances between original samples. So this is a huge dimensionality reduction from the very complicated raw data to just two or three dimensions. And then we can analyze access in this lavender space. So this is a result of dimensionality reduction for B-cell samples and T-cell samples. So as you can see, each dot corresponds to each sample taken from participants. And different colors denote different days. Day zero, day one, day seven, day ninety, in which samples were taken. And it might be difficult to see, but you see that the same color, the same day samples form a cluster, but they are somewhat dispersed in a certain direction. And this trend is same for T-cells. The same day samples form a cluster, like black, but they are dispersed. And this is a two-dimensional projection that is easier to see. And we see that intuitively, and for B-cell samples, horizontal axis corresponds to the time-dependent axis that shows the difference between days. And the horizontal, no, vertical axis. The vertical axis corresponds to the individuality that shows the variability between each samples, same day samples. This applies to T-cells as well. This time, the vertical axis is more related to time, and the horizontal axis shows the variability or individuality between samples. So this is a very simple intuitive argument, but we can make it rigorous by using T-cell decomposition. So T-cells are basically, in this context, just three-dimensional versions of matrices, having dimensions and participants or different days, and loving the coordinates. And this tensor can be decomposing to more simpler rank-1 tensors like this, actually the sum of rank-1 tensors. It's called Cp decomposition. So if you look at each component and see the days component, so this is an ideal case, but if the days component is time-dependent, this component can be considered to be time-dependent, and if it is not dependent on time, it can be considered to be time-independent. And actually, we were able to separate the d'Arabenda coordinates into time-dependent and the individuality axis using tensors decomposition. So we analyzed the individuality axis and followed it over time. And first, we separated the participants into two groups based on the value of the individuality axis on day zero. So group one has a lower value of the individuality axis, whereas group two people have higher values. And as we follow over time, the value of the axis is relatively stable, which means that group one people tend to stay low, whereas group two people tend to stay high, even but with occasional switching orders for b-cells. And this trend is more conspicuous in t-cells, where the group under the individuality axis is more stable than the b-cell samples. And we further looked at... We tried the biological characterization of the axis at day zero, and we found that when we looked at white blood cell differential counts, group one people had more lymphocytes that is related to adaptive immunity or producing antibodies. And group two people had more neutrophils that were related to innate immunity or inflammation. So group one seems to be more ready to respond to vaccine even before vaccination. And this trend was verified when we looked at b-cell subsets. And this is the lineage of b-cell differentiation from immature to naïve to memory or antibody producing plasma cells. And we found that group one people had a larger fraction of plasma cells than group two, supporting that group one group one people are more ready to respond. So to conclude, we analyzed variability in the immune system in a cohort of 300 volunteers who received seasonal influenza vaccine and our lavender analysis enables us to extract critical axis of individuality in a supervised and unbiased manner. In fact, in our dataset it uncovered the baseline immunological characteristics under the new response to the vaccine, i.e. adaptive, that is adaptive immunity dominant or innate immunity dominant. So this kind of answers the how different question. But I think that to answer the why different question we need to look at both specific genes or specific cell subsets that were discovered in our dataset. So I would like to acknowledge co-authors, especially Daigo Kada, who is a graduate student in Yabadalab, Ryo Yabadalab, and Dr. Seto, who is an immunologist who performed experiments in Matsudalab. Thank you very much. So I don't think I understood because what you showed is that there are different maybe populations in Europe, but basically they have different immune profiles, but I didn't see the connection between that and how they are responding to the vaccination. It also ignores that the patients would have been exposed in different ways to the flu, historically, as there were childs, etc. So how to evaluate the output is an important question, and we measure the output of this system using the antibody titers, but unfortunately the antibody titers were not in clear linear correlations with these coordinates we measured. But as you mentioned, these small antibody titers are more correlated with vaccination history. That is reported in other studies as well. So the people who have never been vaccinated show the larger response to the vaccination. And I think in my opinion, the antibody titers are accumulation of workings of plasma cells. So that's kind of an integration. So I don't think it's so unnatural that if the antibody titers don't show linear relationships with... So our coordinates are correlated with the fraction of plasma cells, but we're not directly correlated with antibody titers. So that's maybe, as you say, there may be a difference in concepts or what we... So, yeah, and we need to be more specific in discussing these results, I think. How do I define which populations? T and B cells. T and B cells. Yeah, so there are a large amount of studies related to defining various subsets of B cells and T cells, which should I use? And there are just very basic ones. And for example, in a recent study, it was shown that vaccination efficacy is related to the specific subset called follicle helper follicle T cells. That was not measured in this study. So I agree that this study is kind of simplistic in terms of immunology, but we think that the same method can be used for analyzing more detail data. One more question. You have group one and two. Do they correlate with age? So that could be a bit correlated. No, there was no... at least statistically no difference between group one and group two. And I found that one... I didn't remember which was that, but in one very volatile... in one curve showed a very volatile movement. There was a young female. My delusional idea is young females are more prone to... so immunological... more active maybe, immunologically.