Scientist Stories: Helder Nakaya, Systems Immunology

Welcome to the influenza webinar series hosted by the Human Vaccines Project. The Human Vaccines Project is a nonprofit public-private partnership with the goal of decoding the human immune system to accelerate the development of vaccines and immunotherapies. In order to make real progress toward the development of a universal influenza vaccine, collaboration among researchers around the world is essential. We hope this monthly webinar will foster the scientific foundation needed to advance influenza research. This series of webinars will be broadcast on the third Tuesday of each month and posted online. For a copy of today's webinar, please visit the Human Vaccines Project website at www.humanvaccinesproject.org/talks. Following the presentation, we will have time for questions from the audience. Please type your questions into the questions box in the GoTo interface. I will read these questions following the presentation. And now it is my pleasure to introduce Dr. Helder Nakaya. Dr. Nakaya is an assistant professor in the Department of Clinical Analysis and Toxicology at the University of São Paulo, Brazil, and an adjunct professor in the Department of Pathology at Emory School of Medicine. He is an expert in systems vaccinology, an interdisciplinary field which combines system-wide measurements, networks, and predictive modeling in the context of vaccines and infectious disease. Dr. Nakaya's laboratory applies computational systems biology approaches to understand and predict the mechanisms of vaccines induced— vaccines induced immunity for yellow fever, seasonal influenza, meningococcal, and tularemia vaccines. Today, Dr. Nakaya will tell us about unraveling the mechanism of influenza vaccine-induced immunity through systems vaccinology. Go ahead, Dr. Nakaya.

Thank you so much, Jillian. Hello, everyone. Uh, today I'm going to talk about systems vaccinology, and I will show some published data from the beginning when I start working with this, and also some unpublished data that we have been doing here in University of São Paulo in Brazil. So before I start the presentation, I would like to show a little bit of what our laboratory does, and I like to show by using the logo of the lab. So the laboratory is all dry, so there are only computers, no bench, no centrifuge, no freezers. So this motherboard here represents what we do with the computers. And we also work a lot with systems. So we analyze complex systems as networks, so gene regulatory networks, metabolites. But what really interests us, what is behind all this, is the biology. So this laboratory is a computational systems biology laboratory where we use bioinformatics to try to study and understand more diseases, infectious diseases such as chikungunya, dengue, Zika, yellow fever, and also several vaccines that are relevant for countries like Brazil, for example. So I'm not going to talk about this, but this is just to show that we have been using bioinformatics and systems biology to understand why chikungunya infection causes arthritis, or why Zika could cause microcephaly. So this is the data from brain of babies that die with Zika and microcephaly. But I would like to focus on the vaccine knowledge part. And as you all know, biology moved from cellular to molecular a few decades ago. And what happens is that in this past 25, 20 years, there were major revolutions in robotics, nanotechnology, computers, internet, lasers. And if you take, for example, the human genome sequencing, It took us 14 years and $2.7 billion to accomplish, and now we can do it in less than 4 days for around $1,000. So the molecular biology now went to the high-throughput biology. So everyone is doing omics now— transcriptomics, proteomics, metabolomics. And if you take, for example, immunology, if you look, all the conventional tools we use in immunology, which are still really useful, they have now an equivalent which is more high throughput, or at least almost high throughput. So for instance, what you can do with ELISA, you could, you could now use Luminex or mass spectrometry to do proteomics, metabolomics. Same thing with knockout mice that took years to develop, now people can do in a few months. You can measure the the expression profile of all cells or even single cells using next-generation sequencing. You can assess the whole immune repertoire of someone. And also, instead of 16 colors, you can use dozens of colors using CyTOF, for example. And now we seem to move to a more integrative biology, which is the systems biology part, when you are now not looking to one layer of information, which is, for example, transcriptomics, but also different layers and trying to integrate all of this. And the consequence of all this is that we have now projects such as the Precision Medicine Initiative, Human Cell Atlas, the Earth Microbiome Project. And it was clear that data, that biology became big data science. So the data generated across these phase increased exponentially. And the question is, what about the knowledge? Did the knowledge also increase exponentially? Sidney Brenner, he said that we are drowning in a sea of data and thirsting for knowledge. Most biology today is low input, high throughput, no output biology. So that's the, the motto of the lab, because we are really afraid of just generating data without trying to give any knowledge. So what the summary is, that data itself does not mean anything if it's not to bring knowledge, and knowledge to be transformed into understanding. And what do I mean by that? So for example, if we take eggs and put in a boiled pan, we are gonna have— we know that we are gonna have boiled eggs. But how many of us understand the molecular mechanism, the molecular aspects of this, how the, the, the bioattacks are born at molecular level. Same thing with vaccines. If we immunize millions of people, we know that they're going to be protected against infectious disease. We know that this has to do with antibodies, innate immunity, CD8 cell response, but we really need to know more about the mechanism of vaccine-induced immunity or why people respond differently to the same vaccine. And that's how I think that systems biology can be applied to not only vaccinology but also to immunology in general. Now, if you look at systems immunology, there are like two ways of, for example, investigating the molecular mechanism of something. Imagine that you have a perturbation, in this case ligand binding, and then you want to know what happens here that gives you this output here. The top— the bottom-up systems immunology, the bottom-up way, is what we have been doing the, the several decades of immunology. So which is doing mechanistic studies, trying to understand every little piece here, and then putting together here to try to understand it and, and what's happening in in this process here. What we do in the lab is more a top-down approach, which is you start from this black box, you measure the whole thing, and then you try to reduce to using computational and mathematical models that will be able to find places where you can test and validate later. And this perturbation, there is no other better perturbation to study immunology than vaccines. Because you know exactly when you protect the system, the dose that you are using to protect the system, and that's the, the goal of systems vaccinology. And that's why systems vaccinology is an important field. When I joined Bali Pulendran's lab at Emory University, when I became his postdoc at Emory, he has this dream of, can we unravel the molecular mechanism of vaccine-induced immunity? So he told me that his dream was actually to later, once we do, we study several diseases and identify the signatures of each disease, we could put these signatures into a commercial kit or a chip, vaccine chip, and then use this to predict vaccine-induced immunity. So you will not need to wait 30 days or even worse, you will not need to wait people to get it, to die from the infections to see if the vaccine worked or not. So that was the ultimate goal when I joined his lab. And the principles of systems vaccinology is the same as systems biology but applied to vaccines. So for instance, we use the vaccination as a perturbation of the system, and then we monitor the response at the systems level using omics technology. Such as CyTOF, next-generation sequencing, mass spectrometry. And then these data, these big data actually, is going to be integrated into computational and mathematical modeling that are going to be used to understand and to predict the system. And with the hypothesis and the insights that are generated in this phase, we can think, for example, in a new perturbation of the system. We can think, for example, to increase the dose, add an adjuvant, or make heterologous vaccines and then do it again all over. Or we can try to validate and test these hypotheses using animal models, mechanistic studies with in vitro systems, and that may lead to better vaccine design and development. So that's the goal of systems vaccinology. And that's what we did in 2009 in this paper led by Troy in Bali Pulendo's group. So we study people vaccinated with yellow fever vaccine and then measure the antibody response that happens several weeks later. So you measure the expression that happens in the first 7 days after vaccination, so the innate response. So you measure the expression of 25,000 genes in parallel in the blood, and then you try to predict, you try to find genes that could predict what's going to happen with the CD8 T cell response and the antibody response several weeks later. And then we found some genes that could predict, for instance, CD8 T cell response. We knew that if you have the expression, high expression of this gene in the early phase, in the early innate response, you could predict what's happened, what's going to happen to CD8 T cell response. And one of the genes was GCN2. GCN2 is a kinase that senses amino acid deficiency in the cell. Well, it is related to an ancient mechanism where, for example, amino acid starvation can trigger GCN2, then the metabolic stress pathway— it's going to phosphorylate IGF-2 alpha, it's going to shut down translation, and it's going to promote autophagy. And the question is, how GCN2 could be related to CD8 T cell response? And I don't have time to show the result. That was a beautiful study led by Rajesh and Nur that shows that the yellow fever virus could not only lead to this starvation but also activate several genes related to innate immunity in dendritic cells, and that was related to antigen presentation to T cells. So that was published 5 years later, showing that an unappreciated link between virus-induced integrated stress response in dendritic cells and the adaptive immune response. So we only found this GCM2 because of this Systems Biology paper on yellow fever. And then there was a link how this could be related to higher CD8 T cell response. And then Rajesh was analyzing, was studying this GCN2 knockout mice, and he noticed that there were some problems with the inflammation of those mice. So he asked me, I was already here in Brazil, to check if the expression of GCN2 was also changing in a disease, the chronic inflammatory disease in the bowel, such as ulcerative colitis and Crohn's disease. And then, yes, we did not need to start a new cohort. We just got public available data, and we showed that the GCN2 was actually changing in this inflammatory disease compared to healthy controls. And that also was part of his paper, that he shows a lot of mouse work that shows that GCN2 also controls gut inflammation by inhibiting inflammation activation. And all this was done for the yellow fever vaccine, which is one of the best vaccines in the world. One dose can give you protection for the whole life. And you know that this is very different than influenza vaccines. Influenza vaccines, you have to be vaccinated every year, no matter how important you are. And the influenza vaccine depends on the strains, of course, that were chosen. It works relatively well for young adults, but it does not work that well for elderly people. So you have to increase the dose, for example. And also does not work the same when you are vaccinating very young children. And my goal in Bob Lefkowitz's lab was to try to understand why the same vaccine give you like different antibody response and what are the molecular signatures that are involved in these different antibody responses. So all of these were published before and I'm not going to talk about each one of them because they are already public, but I want to talk about one result that was found in this first 2011 paper here, which is this network here. Here are the genes that were correlated with antibody response. So means that if you take the expression of the genes 3 days or 7 days after vaccination, it is correlated with the antibody response that happens 30 days after vaccination. And one of the genes that I want to bring your attention is TLR5, Toll-like receptor 5. The expression of Toll-like receptor 5 was positively correlated to antibody response against influenza following vaccination. So that means that if you have a higher expression of TLR5 at day 3, you would have higher antibody response 30 days later. But that did not make any sense, you know, because the ligand of Toll-like receptor 5 is actually flagellin, and flagellin is a bacterial component. So why a bacterial component would be important for influenza vaccine, inactivated actually influenza vaccine. So the first thing we tried to do was to see if Toll-like receptor 5 was really important for influenza vaccination, and that became Jason O's project. So what Jason showed was if you immunize wild-type mice, the immune response to the vaccine increased. To the flu inactivated vaccines. But if you immunize toll-like receptor 5 knockout mice, the immune response diminish here. And then the second question was, so where the ligands is coming from? We check for contaminants in the vaccine and there was no contaminants, bacterial contaminants in the flu vaccine, the commercial flu vaccine. So where the laganin was coming from? The hypothesis is that they were coming from the commensals in the guts. And remember, that was in 2010. The microbiome story was not like a hype in that time. And how do you check if the commensals in the gut are important, for example, for influenza response? You can treat mice with antibiotics. So if you treat mice with antibiotics, the amount of bacteria in the stool decreases 26-fold. And if you buy germ-free mice, the amounts decrease even more. So when you decrease the amount of bacteria in the gut, you can then immunize them with flu vaccine and measure the antibody response. As you can see, the antibody response of antibiotic antibiotics-treated mice and germ-free mice were comparable to toll-like receptor 5, meaning that the TLR5-mediated sensing of gut microbiota is necessary for antibody response to seasonal influenza vaccination. And now Bali Pulendro is doing experiments in humans where he is treating humans with antibiotics and then checking what happens with the influenza vaccine response. What was clear to me, like, in that time is that the immune response is actually composed of several different networks. We have a cell population network, epigenomics, transcriptomics. Each layer also talks to each other so that they connect, they influence each other. And when you are dealing with perturbation such as infection or vaccine and measure immune response, you have to take into taking into account not only these different networks, but also past immunizations, previous infections, family history, diet, stress, nutrition status. All this taken into account on top of the technical noise of the high-throughput technologies. And since then, we have been working with different vaccines datasets, so herpes, malaria, influenza, yellow fever, meningococcus, pneumococcus typhoid fever, and also a lot of datasets, a lot of studies on infection, acute infection: chikungunya, Zika, dengue, Chagas, leishmaniasis, tuberculosis, and what I want to show is 3 different projects. One is published and 2 are not. This one was published at the end of last year, you know, it was about the Inflammation induced by influenza virus impair human innate immune control of pneumococcus. And that was involved Daniela Ferreira and her group in Liverpool School of Tropical Medicine that did all the experimental, the wet lab parts, and we did the dry lab parts. Pneumonia is the leading cause of death in children under 5 years old, 1 out of 6 childhood deaths were due to pneumonia. And also pneumococcal nasal colonization is the primary reservoir and the first step for the disease. Influenza infection, this is already known as well, is a major risk factor for pneumococcal pneumonia. So you, as you see, there are like a seasonal influenza increasing pneumonia Streptococcus pneumoniae increasing after. So that was already known already. What is not so clear is the nasopharyngeal mucosal correlates of protection for this pneumonia or the clearance of the bacteria. So what we tried to address was, are there other key players involved in this clearance of, or even in the colonization by the bacteria, and which cytokines and genes are associated with protection. So to address this first part here, which is the colonization and also the influenza infection, how influenza infection is related to pneumococcus pneumoniae, Daniela's group did the human challenge model. So we simulates influenza infection by using live attenuated influenza virus vaccine. So basically you use cold-adapted virus that replicates in the nasal mucosa as a— to simulate the influenza infection, and then you challenge, you inoculate Streptococcus pneumoniae and see if it's colonized or not. And then what we did here in Brazil was to do the analysis, try to find correlates of protection in other key players and cytokines and genes. So what Daniela did there in Liverpool was amazing because they have a facility that collects samples before the challenge. So they collect nasal wash, they collect blood, they collect DNA, then they inoculate the bacteria and then they collect post-challenge samples and do the culture to see who is carriage positive and who is carriage negative. And they found around 40 to 50% colonization. And most important, they did this in over 700 people, volunteers, and no serious adverse events were found. So this is a short video just to show like they inoculating the the bacteria in the nose, and then they wash, and that's what we did the RNA-seq in the analysis. And here they are showing the colonization. So I wanted to talk a little bit about this part here, but it's, uh, the design was Very simple. They separate the group of volunteers into two. One received the live attenuated influenza vaccine 3 days before the inoculation of the bacteria. So that simulates the infection with the influenza virus and then the colonization or not with the bacteria. And as a control group, we use the volunteers immunized with the TIV vaccine, the inactivated vaccine given intramuscularly. The first thing we noticed, so we collect, they collect samples for immunophenotyping, RNA-seq, Luminex, and the first thing that was clear is that people that received first the influenza vaccine, which is the influenza infection, LIV, had higher colonization carriage density compared to the healthy controls, to the people that received the TIV. And if you look at the monocytes, if you look at the monocytes in the nasal mucosa, you see that people that have, that are carriage negative and received the inactivated flu vaccines Basically, the monocytes do not go to the mucosa 2 days, 9 days, or 29 days after the colonization. And when you do the— when they are colonized, the monocytes go there and the peak is on day 9. But when you use the— when you infect first with the influenza virus, the monocytes since you do not go there at the same rate as if you do not have the influenza. So you see that there is a lower level here compared to here. And then we did the differential expression analysis, the network analysis, the co-expression network analysis, where we try to identify the genes and the pathways that are involved in people that are— that receive the influenza infection first and were carriage positive, or the ones that are carriage positive but did not receive the— were not infected with influenza first. And then we see like that the symptoms are actually not the same, they change a lot. But the take-home results could be summarized here. So the hypothesis mechanism was that at day 0, there is no infection. This is the control group. When you inoculate the bacteria, neutrophils go there, degranulate, activate, the monocytes migrate to the nose and that clears the bacteria several days later post-colonization. And when you infect first with the virus, the influenza virus, you trigger inflammation, increase pro-inflammatory cytokines, IP10, TNF, and the neutrophil function is impaired. And then when you inoculate the bacteria, there is an increased carriage, reduced monocytes migration to the nose, and then there is a delayed clearance because of that. Georgios in our lab, he's working with long non-coding RNAs involved in response to vaccination. So this is a manuscript under review now, and it talks about long non-coding RNAs. So long non-coding RNAs are transcripts longer than 200 nucleotides that lack significant protein coding potential. There are actually several types of long non-coding RNAs. This is only one of them, which has 10,000 different long non-coding RNAs in the human genome, but there are like several others that exist that are also long non-coding. And the functions as gene regulators are also diverse. As you see, like, they can recruit proteins to, to, to do epigenetic modifications on histones or DNA. They can serve as a sponge for microRNA. They can work on alternative splicing. They can impair or promote transcription. And if you look at immunology, there are like several papers describing long non-coding involved with different immunological processes, but there was no paper that systematically looked at long non-coding associated vaccine-induced immunity. So that was his goal, try to see if they are there, if they are important for vaccine-induced immunity. What we did was to collect blood transcriptomic datasets that are publicly available. So for example, we have blood from people at day 0 before vaccination and also people after vaccination, the blood transcriptome at day 1, day 3, day 7, day 14 after vaccination. And we did this, for example, for 12 different cohorts of influenza vaccines. So each one of these are different seasons: 2007 season, 2008 season, 2009 season, 2011. And these are the number of subjects and the time points that were collected for each one of these different cohorts. And we did also the same for the yellow fever vaccine, where we collect hundreds of different samples for the yellow fever vaccine as well. And the first analysis that we did was to find if there are any differentially expressed genes that is consistently differentially expressed in all the cohorts. So for example, if you take an influenza cohort, the 12 cohort, the 11 cohorts, you don't see any gene that is upregulated or downregulated in all 11 cohorts, but you do see some genes, these 7 genes here, that were differentially expressed in 10 out of 11 cohorts. And as you can see, each line here is a time point. And when you treat only one dataset, if you ask the gene that should be differentially expressed in at least one dataset, you see thousands of genes. But when you start increasing the stringency by asking genes to be consistently differentially expressed, the number drops really quick. And when you do this for long non-coding RNA, it's the same thing. You do see hundreds of long non-coding if you consider only one cohort, but if you consider several, the number decreases a lot. There is only one long non-coding that was differentially expressed in 7 out of 11 cohorts, which is this 138 long non-coding RNA. And here is the forest plot that shows the meta-analysis for different, 4 different long non-coding RNAs at day 1, day 3, day 7. And this one is a protein coding gene, which is almost a positive control that we know that it is increasing after flu vaccination at day 7. So we wanted to show here them, to show the pattern of expressions. So we do see that some genes, some long non-coding are changing at day 1, day 3, and some on day 7. And then we use an expression atlas to show like where, which immune cell type probably express these long non-coding the most. For instance, if you look at this long non-coding here, the antisense one, it seems to be highly expressing in the case CD4 T cells. If you look at the positive control, the TNFRSF17, highly expressed in germinal center B cells and plasma B cells. And what I want to bring your attention is this gene here, FAM38. FAM38, when you look at the expression on different cell types, we see that it is highly expressed in B cells, especially in plasma B cells, cells that secrete antibodies. And when you look at the location, the genomic location of FAM38, it is actually very close by to immunoglobulin genes. You see, this is the genome region of the gene. And then the question is, do FAM38 somehow is involved in the regulation of the immunoglobulin genes, how can we check if FANTA-RE is involved or associated with the expression of these immunoglobulin genes? We can use an approach which is not the best one, but it's called the guilt by association analysis, which is basically you correlate the expression of FANTA-RE and each one of these genes and see if they are correlated. So of course correlation is not causal, And we are doing now functional experiments to try to see if FONTA2A regulates actively these genes. But what we can see is that for different cohorts, we see a positive correlation of FONTA2A and this gene. We see positive correlation on this gene and this gene here, and also here, and also here. But when we look at this gene here, there is no positive correlation anymore. So as it's stops to be correlated with this gene here. And what about the epigenetic mechanism or topologically associating domains? So what about long non-coding RNAs that act in trans, which is regulate genes very far away in the genomic regions? So one way we can do this is using co-expression, which is you take one sample, for example, and then you measure the amounts of molecules that for the long non-coding, this yellow one here, and also for this gene here, and also for this gene here. So imagine that these are the proportion of molecules for this gene in sample 1, and then you can plot this here in this graph. And then you do the same for the sample 2, for example, where you have more long non-coding and less of this transcript here, so that is translated like this. And then you do this for sample 3, and the same thing here. And when you do this for hundreds of samples, you can see the patterns that there are of some genes that are coexpressed. So that means that when the genes in this module is up, the other gene is also up. If the gene goes down, the other gene also goes down. And then we can find for long non-coding RNAs that have inverse correlation patterns to this coexpression modules. So in order to identify co-expression modules in a very easy way and also do functionality, we developed this tool called SemiTool, co-expression module identification tool. And because I wanted to make it available to anyone, we later created this web version here that allows anyone without bioinformatics background to run modular analysis. So you can just point and click and go to this website, and then you do co-expression modular analysis with any expression datasets. So the end— the results of this analysis, after you run co-expression analysis on 12 or 11 different cohorts, you end up with modules of genes, communities of genes that have similar expression patterns. So then you can look, for example, of what the genes in each one of these communities are enriched. For example, you see that the community 1 was enriched for monocytes. The community 3 is enriched for neutrophils. And so forth, so on. And then you can also look for the activity of this community in these different time points after flu vaccination. For example, the community 1, which is enriched for monocyte-related genes, they are actually upregulated or highly active on day 1 after vaccination. And if you look at the module community 7, which is related with plasma cells and antibody-secreting cells, it is actually highly active on day 7 after vaccination. And then you can look at the long non-coding that are inside these modules or regulated inversely each one of these genes inside this module. And we create these websites called the Vaccine Database, VaccineDB, that allows you to explore not only influenza but also yellow fever vaccine cohorts. So all these cohorts were there in the database, and then you can look for differential expression gene long non-coding RNA correlations and also gene antibody correlation. So if you put FAM38 gene here, you can look the genomic location and also the forest plots. You can see the gene is upregulated or downregulated or positively correlated or negatively correlated with the antibody response on these different time points. And we are planning to do this with more vaccines to expand this database. And this is the last project from Patricia that works with the Ebola vaccine. So this is a collaboration with the EBOPLUS consortium, which is a European consortium that has generated data on vaccinees to Ebola vaccine. The Ebola vaccine is very immunogenic. It gives 100% protection. It gives you like high antibody titers after vaccination that persist for actually 2 years after vaccination. So it is really good in terms of producing antibody response. However, the vaccine is also very reactogenic. If you look at, for example, the side effects, the adverse events that happen after vaccination, you see that myalgia, chills, And in European people, also some cases of arthritis caused by the Ebola vaccination. So the goal was, let's try to use machine learning to try to find genes that can predict adverse events. So even before you vaccinate someone, can you try to identify genes that can tell you if it's good or not to vaccinate this person because this person may or not develop some serious adverse events. So the way we did this was taking the vaccinees, and we know which ones are, which of these vaccinees had no adverse events or they have several adverse events. In this figure, I'm showing 18, but actually there are more than 200 vaccinees. And then we have the expression, for example, the blood transcriptome at baseline for each one of these people here. Imagine like a table with 20,000 rows, and each column is from a different person here. And the way we want to analyze this was using a machine learning algorithm called random forest. And I want to explain a little bit how this algorithm works so you understand how we found the genes that could predict the adverse events. So if you take, for example, the expression of one gene, let's say the expression of gene A. You can plot, for example, the expression of each one of these vaccines here in this axis, and also the people that have adverse, several adverse events. As you can see, there are some places where you can establish a cutoff that separates most of the people that have several adverse events from the people that did not have adverse events. But of course there are some people that were not supposed to be here and some people that were not supposed to be here. But then you can use another gene, for example. You put this gene here and then you can use another gene, let's say the expression of gene F, and then try to find another cutoff that separates most of the people here and most of the people here. And then you put the gene F here and then you do the same for the gene K and then you put it here. And you keep doing this until you separate people that have several adverse events from those that have no adverse events. But the thing is, this is the decision tree. And the thing is that you don't do this. You don't create the decision tree using the entire dataset. What you do is to randomly select subsets of the subjects and also randomly select a subset of genes. And then you do this You create a random dataset and then you create the decision tree out of this. And you repeat this, this process 1,000 times and then you have 1,000 trees at the end. And how do you call a place where you have 1,000 trees? A forest. So that's the random forest algorithm. And what you can do is to count how many times the gene appeared in this tree to give you the importance of the gene in predicting the adverse events. And when we do this, for example, when you try to find genes at baseline that predict having 2 or more adverse events, we found, for example, several genes, and among the top ones were really interesting genes that, that could predict at baseline what's going to happen after you're vaccinated. So just to conclude, I would like to say that we are drowning in a sea of data. There is a lot of data there, and, and the thing is that there are not enough fishermen. If you put the boats randomly, you may not fish any nice insights or hypotheses. And since vaccine knowledge in bioinformatics cannot at all give you the fish, cannot at all tell you, uh, here it is, that the gene that you should validate. That's unfortunately not possible. There is no program that you press a button and give you this. We need the experts, we need the people that understand the biology and immunology to really help us looking at the data and find the nice insights. So the only thing that the bioinformatics can do is to try to reduce the region where you should put the boats, but it's still going to be a fishing expedition. And with that, I would like to thank a lot of people that did this first part here, including my previous mentor, Bali Pulindra, and also the people from my lab. That's it.

Thank you. Thank you, Helder. That was really interesting. At this point, we can take any questions, so please just type them into the question box and I'll read them. And while we're waiting for everyone else to get questions typed in, I'm curious about your random forest approach. So do you think that the genes that you pulled out of there to predict adverse events are specific just to the Ebola vaccine or to a certain subset of vaccines? How broadly do you think that those can be applied?

Well, that's a really good question because we are now trying to do the same for different vaccines. But the thing is that the other vaccines are not as reactogenic as the Ebola vaccine. We do see some inflammatory-related genes in common to Ebola and also to some live attenuated vaccines, but Ebola vaccine is really much more easier to find predictors because it is a very clear output. But I think that once we increase the sample size, we may be able to find more reliable baseline predictors for flu vaccination, for example.

Okay, yeah, that makes sense. Thank you. Um, do we have any other questions? Okay, it looks like there are no other questions, so thank you again. And like I said at the beginning, this talk will also be available on the Human Vaccines Project website later this afternoon. Thank you, Dr. Nakaya.

Thank you so much.