 Great, thank you. Thanks, Lisa, for that very kind introduction. And it's a true pleasure and a privilege to be here. So in addition to the cool project names you heard about, I also co-directed the artificial intelligence for a healthy living project with Dilip Jesty, who's here today as well. And we're really excited about this potential in terms of what AI can do for the microbiome. I've been sequencing my own microbiome every day for the last almost 10 years now. And I still don't understand it. So clearly, this is a case where human intelligence has failed. And we need to, instead, enlist the aid of artificial intelligence. So microbiome research is really changing how we see ourselves as human beings. And I'd like you to take a moment to consider what you saw when you looked in the mirror this morning when you got up. I saw an organism that's just 43% human, and not just because of debt lag. But when we think of what makes us human, very frequently our thoughts tend to our human genome, in which NIH and other agencies, DOE, and many companies, have invested $3.8 billion in deciphering the human genome. And when President Obama launched the Brain Initiative in 2013, already at that point, the ROI had been 140 to 1 in terms of return of investment and sequencing the human genome to the economy. But what we are doesn't end with those 20,000 also human genes that are inside us, because if we think about bodies at the level of cells, each of us has about 30 trillion human cells, but we have about 39 trillion microbial cells on average. And that's where that 43% human number comes from. You might be thinking, well, wait a minute, it's the 21st century. Do we really care about cells anymore? Let's think about it in terms of our DNA. So if we think about that for a moment at the level of our genes, each of us has about 20,000 human genes, depending on what exactly you count as a gene. But amazingly, the size of our microbial gene catalog ranges from two to 20 million microbial genes. Okay, so by that measure, we're at best 1% human. And there's a lot of excitement right now about doing systems biology, even systems medicine. It's hard to do systems anything if you're neglecting 99% of that system. And most shockingly, the 99% we're neglecting are the genes that we can actually change, not the ones that we're fixed with at birth, but the ones that have changed profoundly from the time that we're born to the present, and that we continue to change through things like what we eat, how much exercise we get, even who we associate with. So just by being in this room, we're all exchanging microbes with one another right now. So this was really a big data problem too. So remarkably, each teaspoon of your stool holds in its microbial DNA the amount of data it would take literally a ton of DVDs to store. So you can think about that the next time you're taking a data dump, is it where? And this data is really important for understanding a range of health conditions. So for example, it's now well established that your human microbiome in your gut affects how you metabolize different kinds of drugs, including painkillers, heart drugs, anti-cancer agents, including the latest checkpoint inhibitors for cancer. And what's more, we're increasingly understanding how changes in your microbiome affect changes throughout the rest of your body, including your brain, and we're seeing this as being particularly important for a healthy aging. So a really important question, there's a lot of different things that we're doing today that are depleting our microbiomes, including overuse of antibiotics, overuse of C-sections, not enough fiber in the diet, not enough diversity in the diet, and so on. And a really important question is whether our efforts to control diseases caused by single pathogens have led to a situation analogous to what Rachel Carson documented in Silent Spring and the environment only happening inside our inner environment. Because over the 20th century is one disease after another caused by single pathogens, everything from measles to TB declined precipitously in frequency. At the same time, there was an incredible rise in chronic diseases like multiple sclerosis, Crohn's disease, Type 1 diabetes, and asthma, all of which have exploded over the same time span. And what amazes me the most is that in 2002, when this graph was published in the New England General of Medicine, none of those diseases were known to be linked to the microbiome, whereas today we know that all four of those and dozens of others are associated with the microbiome in humans and causally linked to the microbiome in animal studies. And so if you want to hear more about how we're depleting our microbiomes, Marty Blazer at NYU has written a wonderful book that gets into this in a lot more detail than I can today. So you might wonder, well, does all this diversity in the microbiome matter? And I mentioned that there are a lot of diseases that have now been linked to the microbiome. And one thing I've been working on for about the last 10 years with Jeff Gordon, so I have watched you and a number of others, is obesity. And today, for example, I can tell you with 90% accuracy if you're lean or obese, simply by sequencing the DNA of the microbes in your gut. So on the one hand, this is a cool trick from a technical perspective. On the other hand, we don't think it has a lot of commercial significance as a test for obesity because I bet you can tell which of these people is obese without doing any DNA sequencing at all, right? But if you try that same classification task, lean or obese, using random forest classifier on every human gene ever linked to obesity by GWAS, you can only get 57% accuracy, whereas we can do it with 90% accuracy based on the microbiome. And what's more, we can even transmit that obesity from humans into germ-free mice by transferring the microbiome into an animal model. So in the Human Microbiome Project, this was one of the first large-scale efforts to figure this out, including, so this was a large NIH-funded effort, including the Broad Institute, just a few blocks from where we are at the moment. And so with about 400 researchers around the country, we looked at about 250 subjects, all of whom were rigorously screened as healthy, at up to 18 sites in the body, and collected four and a half trillion bases of DNA, all of which we released in 2012, and then an update in 2017 as open data, so you can download all this and work on it yourself. And so the great thing about this project is we had an unprecedented amount of DNA sequence data about the microbiome. That was also the terrible thing about it. So to illustrate this, I'm showing you the first file of data from the HMP. This is actually the first 0.1% of this file, and it's pretty hard to tell who lives where in the environment from this, right? Like, you probably can't even tell that's an oral sample, let alone the sequence signatures that let us determine that. And this is really a problem at the moment because both through citizen science projects like American Gut and also various companies now, you have the opportunity to get your microbiome sequenced. And so more and more often, you have patients coming into their doctor's offices, always with a huge grin on their faces, saying, hey, doc, I've got some great news for you, which is that I had my microbiome done, and now I have this list of 1,000 species they found in my gut, or maybe a list of two million genes they found in my gut. With all this data, you can tell me what's wrong with me, right? And I mean, what's your doctor gonna do? Refer you to colleagues in psychiatry for being crazy enough to think they could do that, and the 12 beautiful minutes you have together, right? And so our goal is to make it not crazy anymore and to figure out how from a snapshot of your profile or from a movie of it changing over time, you could actually extract that kind of actionable information from it. So a lot of what we do in my lab is build pipelines, including lab processing pipeline and the software processing pipeline called CHIME that basically takes all that data and turns it into usable conclusions by doing things like making sequence alignments, grouping the sequences together, building phylogenetic trees, and then using the phylogenetic trees to relate at the whole microbiome level those communities to one another. And when we apply these tools to the HMP data out of all those sequences, we get this map of the microbiome. So what this is is a principle coordinates analysis of unifract distances, which is a phylogenetically based method we introduced in 2005. And essentially each point on this map represents all the complexity of a microbiome read out from its DNA. So for example, this oral sample here and two points are close together on the map if their sequences are more similar in terms of their evolutionary history and further apart if they're more dissimilar. So I told you these were all healthy subjects but you could imagine all kinds of factors that might affect your microbiome, like whether you're a man or a woman, whether you're old or young, whether you're fat or thin. I'm getting a color on this unsupervised ordination, the main factor structuring this data set. And it's immediately obvious what's going on, the different parts of the body are almost like different continents on this map with their different microbiomes from one another. And to highlight this here, I'm showing you the mouth and the gut of the first person in the HMP, which are in completely different locations on this map. And so you might think, well, you might think, well, what does this really mean? And it wasn't until we did the Earth Microbiome Project looking at tens of thousands of samples contributed by researchers across the planet that we understood because we could go out on the Earth's surface and ask what two points on the Earth's surface are just as different as the mouth and the gut of this one individual. And if you think of your mouth as being kind of like a coral reef with complex mineralized structures covered with biofilms that perhaps your dentist bugs you about, amazingly, your mouth is as far from your gut in terms of its microbiome as the water in that reef is from the dirt in this prairie, right? So essentially, they're completely non-overlapping ecosystems and a few feet along the length of your body makes as much of a difference to microbial ecology as thousands of miles across the planet's surface. And we never expected that. So I told you that the microbiome affects individual response to drugs. What about individual responses to food? So studies of nutrition have been incredibly frustrating because even the largest studies have tended to reveal very small effect sizes and a lot of inconsistency between subjects. And this is still one of the largest of this type of study that's been done. We're a group from Harvard tracked 120,000 people over 20 years looking at the effect of each food item on weight gain or weight loss. And if you think about your stereotypes of what fat people eat and what thin people eat, those stereotypes I have to say are very much reinforced by the study. So the food most associated with weight loss is yogurt but the effect size is pretty small. You lose four-fifths of a pound per year per serving of yogurt per day. And then in contrast, the food most associated with weight gain was our friend the French fry, America's leading source of vitamin C, by the way. There's not much in each fry but if you eat a lot of them, that's where it comes from. But again, the effect size is relatively small so each serving of fries per day led to a weight gain of about 1.7 pounds per year. And so what this means is if every single day, 365 times in a row, you were going to eat a serving of fries and you'd virtuously give it up and eat a serving of yogurt instead. At the end of that year, the difference in your weight, you'd expect to be two and a half pounds if you're typical of the population, which is not a very large impact, right, given the lifestyle change it involves. But on the other hand, there's very high individual variability so with that switch, some people will lose a lot of weight, others will gain a lot of weight, no one knows who and the effects basically balance out. So not in the context of weight but in the context of blood glucose control. A group from Israel led by Aaron Alenev and Aaron Segal were able to show in a groundbreaking paper at the end of 2015 that a lot of these personalized responses are due to the microbiome. So they got 800 people, hooked them up to a continuous glucose monitor, measured a whole lot of things about them, including the microbiome and then predicted the effects of each food item after feeding them a defined sequence of diet. And what they found is that on average, the glycemic index of different foods were perfectly recaptured but the individual glycemic index values for individual people were all over the map. And so for some people, it was actually better for them, for example, to eat ice cream than it was to eat rice in terms of their blood glucose. And what they found was that essentially all of the explanation was in the microbiome, not in the blood tests, the anthropometrics or the various other things they measured. And so this gives us hope that through machine learning, we can start to understand the effects of even very complex things like diet on our microbiomes and on our health. So in the artificial intelligence for Healthy Living Center that I co-direct with DELIP, we're taking a two-pronged approach to address these problems. So one is microbiome, which I'm primarily going to focus on here. And the other is healthy aging, which you'll also see a number of posters about out at the poster session. And this has been a really exciting collaboration together with a number of IBM researchers. And this is just part of the larger landscape of research into aging at UC San Diego, including a lot of work going on at the Center for Healthy Aging, the Moors Cancer Center, the Alzheimer's Disease Research Center, and so forth. So one particular piece of this that's been in the news lately is the study that we're doing together of centenarians in the Italian villages of Cilento. And Cilento is one of these blue zones where people live much longer than in the surrounding areas. And what's amazing is that these individuals at age 100 are still foraging a lot of their heaps from the wild. And so one thing that there's been a lot of interest in is the specific contribution of rosemary as potentially having a lot of bioactive compounds that promote healthy aging. So unfortunately, we didn't get to go and sample the rosemary in Cilento. We looked a little closer to home between my building and Peter D'Arrestines, who's the closest collaborator on this project, where what we did is we got the plant, chopped it up into little pieces, ran each piece through the mass spec, and then reconstructed it into a 3D scanned image of the plant so we can see where every molecule and also where every microbe goes throughout the whole plant. And so what's cool about this is that we can see a lot of known bioactive compounds. So the way to read these diagrams is blue means the least of a compound, red means the most. And so for something like rosemary and acid, you can see it's mostly on the older leaves where it's concentrated. Whereas, for example, methoxycardosic acid is mostly on the flowers and on the younger sleeves. And then we also see a huge number of unidentified metabolites that are unique to particular parts of the plant. So here I'm sharing you some that are unique, either to young auto old foliage. And so there's a lot of interest in going into the ethnobotany of looking at how people in different parts of the world use plants with now we can reveal different levels of bioactives in the parts that they're being used and understand how that relates to living to extreme old age in a healthy fashion. So this was really the beginning of the global food omics project, led by Julia Gauglitz and Peter Durstine's lab. And essentially what we're doing here is running thousands of food substances through the mass spec where what we can do through global food omics is characterize what's going in at a chemical in a microbial level and then relate that to what's going out in the American gut project and then integrate the data to understand the impact of food on health. And all of this we're producing a huge open data resource both on the DNA sequencing side and on the metabolomic side. So if anyone's interested in looking at that data it's all available now. So our journey's through the microbiome map that I showed you track and possibly predict our health. And so there's a tremendous scope for new collaborations and applying advanced AI techniques there. And so in American gut in particular we launched this in part as a response to the human microbiome project where in the HMP you had to meet very strict exclusion criteria to be allowed to be enrolled. And you could only enroll if you happen to be at one of the two clinical recruitment sites. Whereas we wanted to make it possible for anyone regardless of their age, their health status or their location to claim a pin for themselves on the microbial map. Of course it turns out that not everyone wants to know what's in there. So here in middle school is visiting our lab and learning that we're going to use lasers and robots to read out the DNA of the bacteria in their poop. But at this point it's been a fairly successful project with over $2 million raised, essentially all the $99 increments from members of the public. And over 10,000 individuals, it's over 15,000 samples now sequenced and released with all the de-identified information released for free so that anyone can analyze it. And a lot of people assume that the pipeline is just as easy as this. It turns out that there's a lot of technical detail that separates the production of the sample from the production of your readout of your microbiome. But essentially what we can do is give you information about yourself, how you compare to other people and the kinds of microbes that are common or rare in your sample specifically. And on the scale of 10,000 people, we see all kinds of associations with the microbiome that no one suspected. So you might have expected that your microbiome changes as you age, which it does. But another thing that's just as statistically significant is that it changes with how many hours of sleep you get at night. And that question was considered so crazy that our original institutional review board at Boulder kept crossing it off the questionnaire because they thought there was no way that that was something we should be bothering our subjects about in terms of link to their microbiome. Whereas now it's very well established that there are links between the microbiome and sleep, both in terms of your microbiome being able to determine how much you sleep and in terms of damage to your microbiome through jet lag, through time shifting and through other issues like that. And one really exciting thing about this data set which again is all available as open data is that we can start putting together a common effect size scale looking at how a lot of factors affect the microbiome. So here we're looking at the number of people per group on the x-axis versus the statistical power of the test on the y-axis sub-sample from that data set. And basically things like age and inflammatory bowel disease and antibiotic use have large effects on the microbiome where you need a few dozen people per group to find them out. Whereas other things like BMI, alcohol consumption, exercise frequency have more subtle effects. But amazingly, one of the leading effects on this graph is the number of different species of plants that you ate in the week before you collected your sample. We had no idea that that would have such a large effect and in particular an effect that can even override the effects of age or IBD or antibiotics. So that's a very exciting direction we're following up. But essentially with these massive data sets there's a lot of issues with visualization and representation as well as coming up with good ways to quantify the effects of different parameters and also understand which of those parameters are independently of large effect which are dependent on other variables. We're analyzing that in more detail now. And one thing we're doing is taking it international. So the MicroZeta initiative which we launched back in April is an umbrella project for the American Gut Project, the British Gut Project, the Asian Gut Project and various others around the world. And so what we're trying to do is to understand the large regional variation in the microbiome and understand which of those microbiome differences are important for underpinning health in different parts of the world where there are very different susceptibilities especially to the chronic diseases of old age. So while this is cool, but I'd argue that it's not enough. So what we really need to do is move beyond the kinds of statistical methods that we're using to get better classifiers, to get better prediction. So we want to know not just what's happening with your microbiome right now, but why can we predict about your future from the microbiome? And also we need to understand so advanced is an explainable AI we think are tremendously exciting. And the solution we're turning to with this is artificial intelligence. So our project on the microbiome side is largely focused around the Watson technology where essentially the idea of the first three years is to educate Watson and then the last two years to apply it to diseases, especially those in aging. So this year we've been teaching it the names of microbes. Next year we'll be teaching it about microbial genes. The year after we'll be teaching it about microbial metabolism. And then in years four and five apply it to diseases of aging including things like Parkinson's, possibly Alzheimer's and a range of other diseases that are more prevalent in old age. And of course this is where it's especially exciting to be part of this network. So I'm just gonna show you briefly really advertisements for the posters. So we've been going a year and we've already made a tremendous amount of progress in producing novel research with both UCSD and IBM co-authors. So one thing we've been doing just to see what the baseline is is looking at a whole range of different regression methods applied to microbiome data. This is led by Patrick McGrath where essentially we're looking at about 20 different algorithms including classical machine learning algorithms for understanding what gives us the best regression characteristics in a range of different tasks. Haidang Lee has been leading a project on selecting taxa that are associated with disease. We've been starting with inflammatory bowel disease because we have a number of very productive collaborations including one with rhomnexaver at the brode looking at the role of the microbiome and IBD. And so the goal here is to use basically an additive regression trees to get better classifiers than we've been doing with random forests and other techniques that we're more familiar with. Another really important question is what features should you use? So typically we've been using a clustering based method to look at groups of sequences that are related to each other. But using the phylogenetic tree has some important advantages. And Effan Sayari is going to present a poster that goes into detail about how much more effective a classifier you can get using these tree-based features rather than using sequence cluster-based features. And then finally, we're going to talk about the NLP side of the project. And Dustin Wright has been leading this project looking at disease normalization so that we can construct the biomedical knowledge base. And a lot of the strategies for normalizing diseases we're also going to be using for normalizing groups of bacteria and groups of genes that perform the same kinds of interchangeable functions so that we can associate extractions between bacteria and diseases, both from abstracts and from full-text articles. So you might be wondering what's our vision for this in the long term. And just to give you the best example we have for why you care where you are on this microbial map, what I'm showing you here is these orange spikes are fecal samples, but they're fecal samples from people with Clostridium difficile, which is a very nasty hospital-acquired infection that kills about 14,000 people a year in the US alone. And you can see that they look nothing like healthy fecal samples. And what's going to happen is that four of these patients are going to get a fecal transplant from one donor who, as you can see, is in the healthy range defined by the Human Microbiome Project. This is what we did with Alex Droots, Mike Sadaski, the University of Minnesota. In case you're wondering what a fecal transplant is, here's Bill Sanborn, who's our chair of GI, about to deliver one using hospital-grade stool that he got from a non-profit called Open Biome. That's right here in Cambridge because you have to remember that the FDA regulates stool as a drug. So anyway, four of those patients are going to get a fecal transplant from that one donor. And you might wonder, are you going to see any change at all at the level of their whole microbiome ecology? And each frame in this animation is just going to trace their journey on the microbiome map one day at a time. And what you can see is that immediately within one day, all of them move down into the healthy state and then they stay there for months of follow-up. And this is coupled to remission of all their clinical symptoms. So it's incredible how much you can transform someone's health literally with the gift of healthy human feces. And so what's challenging at the moment is that for all of these different medical conditions that are now linked to the microbiome, we have to figure out whether it's a very unsubtle change like this one or a more subtle change like an aging or obesity. What is the difference in where you are on the map? And what do you need to do, so in other words, how do you find, for all of these different conditions, how do you find the good places and the bad places on the microbiome map? And then how do you guide it back into health, whether it's something as extreme as fecal transplant or phage therapy or whether it's better drugs or whether it's something as gentle as probiotics or probiotics or even diet? And so a lot of our dream is to go beyond this kind of mapping exercise and develop a microbiome GPS where the concept is that maybe as soon as you flush, you get an instant readout of your microbiome telling you on your smartphone, which let's face it, I bet you're using in there anyway, am I going in a good direction or a bad direction and what should I do in order to maintain my health over my lifetime? And when my daughter was two, which is when this photo was taken, we couldn't do this sort of thing on a smartphone, but we can now. And this is her view of her own microbiome that she developed over the first three years of life. And even to a six year old, which she is now, you can explain how the microbiome is moving in these different directions that make it healthier or sicker. And so we think this is the kind of thing where with good enough visualizations and with the user interface for your microbiome, we might be able to guide you into a place of health throughout your lifetime. So this is really the science fiction part of the talk, but a lot of these pieces exist already. Our dream is that microbiome analysis is going to not be exotic and require million dollar instruments in the lab, but be routine and you can do it as you examine yourself in the mirror in the morning. So the idea is that as you look in the mirror, perhaps as you breathe in your mirror, your breath will be whisked away for some kind of instant chemical analysis that you can see right there. Obviously we're still working on the user interface for this, because that's probably not exactly what you want to see. But then translating that using some of the same deep learning technology that back to Google translate into the kind of microbiome report you would get from American Gut. And then ultimately placing you on the microbiome map, identifying any risk factors that you might have based on where you are on that map, and then giving you any guidance on what you should do, either that you could do yourself or that you might need to consult a physician about, depending on the risk level, in order to stay in the healthy region. And then you could also imagine that smart mirror communicating with your smartphone. So when you're faced with a thousand yogurts in the supermarket, you can use augmented reality to zoom in on the one you're looking for on the shelf, then scan the barcode to confirm that you got it. And then finally, you could imagine that smartphone recording what you did during the day, what you ate, your activity, and so on. And reporting all of that back to your smart mirror to show yourself visions of yourself. Five, 10, 20, maybe even 50 years in the future about what you're going to look like if you behave the same way that you did today. Or what you might look like if you did a little better or if you did a little worse. And a lot of these individual technologies exist, including the real time video editing and including doing things like video editing for affect. So it could make you happy or a sadder depending on your projected future microbes. A lot of it is gathering the data that we need to fit those models. So that's the kind of crazy thing we're up to at the Center for Microbiome Innovation that I directed UC San Diego with an increasing list of corporate partners, including IBM. And it should be over 130 UCSD faculty members now on a number of projects as crazy as this one. So with that, I'd like to thank everyone involved on the IBM UCSD artificial intelligence for a healthy living center, both on the IBM side and on the UCSD side. Here I've told you mostly about the microbiome part, but you'll be able to see a lot about the healthy aging part at the post-accession as well. I'd also like to thank the many members of my lab and our various other funding sources for some of the projects that I showed you here, including tens of thousands of members of the general public on the American Gut Project. And finally, thanks for your attention. I'd be delighted to answer any questions you have. And I think the goal was to enter the questions on the app and I'll read them out from the screen. Thank you and thanks again for this opportunity. Right, great. Okay, so Rob, actually we're having, okay, I will read the questions for you. Okay, okay. So what advances in AI and machine learning methodology are needed most for the microbiome and the health projects? Well, well, there's a lot. We could really use better NLP to get things like microbiome disease associations and especially as we get down to the strain level disease associations. For structural biology, which I didn't really talk about here, we need better ways of predicting protein structure from the genes and then relating those structures to one another and to their functions. And so that's somewhere where 2D and 3D spatially resolved image analysis protocols like convolutional neural nets could really come in handy. And then just in terms of the, just in terms of classifiers for the microbiome, there's a lot of boosting strategies that I think are gonna be exciting for developing better classifiers and better regression that have not been applied in this field yet, but could reasonably easily be applied. So those would be some of the big ones, I think. Okay, fantastic. And actually we do, we can talk afterwards. We have some projects in terms of this better structure where we're using more of like techniques that were founded out of the NLP field to try to add structure for protein sequences and so on. Oh, that's great. I'd love to chat more about that. Okay, great. And like I said, almost like all these data sets have shown you were open, so it's really a playground to play around in. Okay, fantastic. So I think we'll have to go to the next speaker. So thank you. Great, thanks again.