 Okay, so what we're going to imagine here is that you guys have gone through the analysis. You've been able to do some identification of either interesting pathways, interesting molecules, and at this stage you'd like to try and relate what you're doing with metabolomics to some of the other possible omics systems or other conditions and situations. Remember, metabolomics is only one part of that pyramid and that ultimately if we want to get some greater understanding we'd like to connect proteomics and genomics. So we've seen how we went from the metabolite lists that we learned how to do yesterday to get some information pathways. That's biology. That connects metabolites to proteins and proteins to genes. And if we want to use pathways, we try and work with the databases, some of which we've already talked about last time. So we saw how metabolologists is connected to K and to SMIT DB. And we talked as well the other day about SMIT DB and the number and types of pathway databases and what's in them and how you can use it. And as I say, this was a database designed for metabolomics. Unfortunately it's only for humans so it doesn't cover everything that everyone's interested in. And we saw how you could do things like mapping concentrations and mapping positions and cellular locations and that allows you to generate things. The next step beyond the pathway stuff and that was also the METPA work and you guys saw that as well was to going from pathways and lists to models. This is another element of systems biology and this is an aspect of simulation and this is crunching equations and one that also metabolomics I think has been sort of unique and has been leading the way. So has anyone heard of Bernie Paulson? One. Anyways, his group in San Diego has been developing reconstructions, metabolic reconstructions of entire organisms. So they have modeled the human being, they have modeled E. coli, they have modeled yeast, they've modeled a whole bunch of different cells. And to do a reconstruction they have to connect the genes, the transcripts, the proteins, all the reactions and all the compounds together. So it's non-trivial. And so their first efforts were on smaller organisms like mycoplasma where there's like 500 genes but then they moved up to E. coli where there's 1,000 plus genes. And they, as says, progressed now to humans and then originally it was just a single compartment, E. coli just has a single compartment but humans and more complex organisms have multiple compartments. And the main pathways that they generated for the Paulson reconstruction was 98 or just under 100 different pathways. We call it, Smith DB has 90. What people call a pathway is sort of arbitrary but anyways somewhere between 80 and 100 is where the metabolic pathways are known. Anyways, that information is compiled from resources like the HMDB also from the literature. And that information can be used to create what are called flux balance analyses. So there's the genes, there's the proteins, the metabolites, they have to drive the equations. They might do some testing on subsets to make sure that everything works. They have to deal with balanced equations. So this is chemistry where you have to actually worry about where electrons and protons are involved. You also have to have a charge balance with all of these. And they will have constraints. They have to know roughly how much of the starting compound is there, how much is produced. And each time they run these tests they'll find failures, they adjust and tweak. So this is many hours. But that's got genes, proteins, metabolites, reactions and then they can start doing these flux balance analyses. And this is essentially a way of saying given a starting state of how many metabolites and composition is this a sustainable metabolic reaction. If we in silico knock out a gene or mutate a gene or move a bunch of genes, will the system find an alternative way to create it or will it die? Will the system grow? This is not a dynamic simulation. What it is actually is a whole bunch of linear algebra equations. And if you guys have ever taken linear algebra matrices, you had to calculate cost functions or things that were saying what's the optimal price and optimization. So this is an optimization calculation where time, these are time independent equations and you're essentially working with it. How much of a given product, which is one set of metabolites, and how many inputs have I got? And will these maximize out in a certain way? And are they solvable? So again, if you remember a little bit from linear algebra, sometimes you come up with equations that are not solvable, divide by zero. Others you'll find things zeroing out. So in this way, they can predict whether a given system is functional or whether a given system will thrive. So you can predict essential and nonessential genes, essential and nonessential metabolites, essential and nonessential proteins. You can predict a phenotype. Will this organism or this human grow? So in the case of humans, we know certainly metabolic mutations, phenylketonuria or PKU, knock out this gene. What happens? What's going to accumulate over time? And does this match with what we observe in human studies? You could do this for E. coli, you could do this for yeast. So this is flux balance analysis and you're balancing the flux in and out. So it's a classical set of studies that chemical engineers do and Bernie Paulson is a chemical engineer. Another approach that is not a steady state analysis, but a dynamic analysis is something that we've been trying. It's a tool called SIM cell or simulating a cell. And you can try solving time dependent partial differential equations and dealing with spatial diffusion and stochastic events. And it gets really, really complicated. Or you can do a simple thing. I don't know if anyone has played the games like the Sims or Sim City, some of these old ones where these are called cellular automata or agent based games. And some of these are probably very similar to war games now. If you're battling enemies that the computer creates, it creates automata or agents that have simulated properties like a game player. So you can do agent based models where you're creating enzymes that have functions, genes that have functions, metabolites that have functions and connections to each other. And if a particular enzyme recognizes a particular substrate, you tell it that rule. And now you just tell the enzyme to just bounce around inside a cell until it runs into a substrate. And if it looks at the substrate and says, is this a good substrate or not? And if it is, then you tell it a rule, convert that substrate to its product. And then you can tell it a rule that convert that substrate to that product at this rate or with this frequency or with this speed. So in that regard, you can, and if you've played a game, these Sim type games from Sim City to these war games, they behave pretty realistically. Sometimes it's pretty hard to beat the Sim agent. But you can do the same sort of thing again with cells. And now you create essentially a time dependent, spatially dependent process where you can see events happening over time as signaling happens, as drug metabolism happens, as interactions will happen. Now this is not something that you can measure over eons. These are interactions that are measured over seconds to minutes. Whereas the flux balance analysis is something that says, you know, after a few days, months or years, this is what we'd expect to happen. So you can download Sim cell and you can run it instead of a sketch system where you can draw pathways and cell systems and then run them. So this is one where we created a compartmental model actually of drug metabolism. And we had, I think there's the liver and the heart or kidneys and the heart and then a cell system. And so you could actually show where systems were moving around. So it's not just cells. It could be compartments that allow you to do this. And you can take an existing pathway and then sketch it out onto the Sim cell. And all this is a schematic which tells the computer what it has to do. And then from there the computer actually starts calculating concentrations as they climb up and fall and changes in the appearance of products, some disappearing completely, some rising. And in fact this is a picture of a time dependent change. Actually in oscillation you can make cells oscillate fairly sequentially if you set up the right parameters. And the people may be seeing these chemical reactions which oscillate. So you can do the same thing with some genetic manipulation. And this is one that we had done. And this was the simulation. And people have seen this oscillation in cells. So the system works. And we've played around, this is a couple years ago, trying to get this simulation to go on in the computer and then to take that information that we were mapping and then to put it onto a virtual human. And this is where we're looking at some drug metabolism and where it happens and how things would distribute. And we can match what's been measured metabolically in different compartments. It's still a long ways from being I think something routine. But it's an example where you can take this list of metabolites and actually translate it into something that's biologically meaningful and measurable. So we've seen this picture before as well in terms of, again, you've got your list of metabolites. What can you do with it? And so these are the applications. And so some of them inspire the list of metabolites. But then once you've got your list of metabolites, you need to work with the people who were asking you to do this and tell them what it means to translate into decisions, to translate into protocols, to translate into biomarkers. So about half of the applications I listed there were in clinical medicine. And about half the people here are probably involved in some aspects of clinical or medical research. So I'll talk about some of those applications. One application I guess is something that you guys did in your tutorial yesterday, which was trying to learn a little bit about inborn errors of metabolism or IEMs. So imagine if you had your spectrum, it could be NMR, GCMS, LCMS, but in some case it generates a list of metabolites. What were the ones that could have been a bunch of peaks? You didn't know what the peaks were. The peaks then tell you what the metabolites are. And then you want to find out what do these metabolites do and what are they relevant for. And so we used the example of, here's another list of mass spec feeding it in, you get your list, and what does it tell you? So what you guys would have gone through in that tutorial was from these lists of compounds that were up or down regulated, you started saying that there was obviously some effects in whether it was diseases or by metabolic pathways that these are associated with phenylalanine metabolism and that they're also associated with a disease, PKU. So if you hadn't known anything about PKU and it started measuring some of this stuff, this would have led you to it. There are other applications beyond just sort of using the databases we've talked about. These are some interesting papers, which I'm going to show, where people have used metabolomics to identify in some cases the causes, in some cases probable biomarkers that help understand the etiology of certain diseases. This one, and actually it's not type two, it's type one diabetes, but this is the diabetes, this is juvenile diabetes is what used to be called. And this is a disease that strikes children and essentially wipes out their pancreas. Pancreas doesn't function and they're on diet and on insulin for the rest of their lives. And what's been known is that most type one diabetes cases, I mean there's many causes for it, but the ones that seem to be most fundamental are antibodies to this enzyme. And I'm trying to remember was this, it's GAD, it's a dehydrogenase, but it's sort of a glutamate dehydrogenase. And they wondered why. I mean, what's important about this GAD enzyme and why are there antibodies to the GAD enzyme? It has nothing to do with insulin and it has really nothing to do with sugar metabolism and so what's the deal? Anyways, in Finland they started doing a study where they were taking and monitoring kids who seemed to be at higher risk where parents had diabetes, where grandparents had diabetes, it was okay, what's going on? So this is a young girl where they were tracking her over every three months or nine years collecting blood samples and measuring some of the compounds that were in her blood. And one of the things that they found, this is the case of enzymes, this is the case of some of the amino acids. So one of the things they found was this child was born with a 13-fold higher level of glutamic acid than normal children in their blood. And as the glutamic acid started trending towards normal, another trend happened by a year or two where a metabolite of glutamic acid, gamma, gamma-aminobutyric acid also spiked and then gradually normalized. As these compounds are sure way high, really high, started to normalize, they started to see some of these enzymes occurring, trying to deal with this high dose, these high levels of glutamine. And it wasn't too long after these enzymes started peaking to deal with this huge level of glutamine that diabetes started appearing. So it argues that in fact, at least in this case and perhaps many other cases of tycoin diabetes, that it's, yes, it's an autoimmune response, but it's triggered by high levels of amino acids, a dysfunction amino acid production. The result is the body tries to normalize the amino acids, but to do that, it produces an enzyme. That enzyme becomes too high and then the body tries to normalize the enzyme by creating an antibody to reduce the enzyme action. That antibody unfortunately starts targeting the source of the enzyme, which is the pancreas. So it's one leads to another, leads to another, but the perspective has said, oh, it's, you know, it's some kind of autoimmune disease. No, it's a metabolic disorder that the body's trying to normalize and it has unintended consequences. So that's one example and this is where you can start looking up on some of these compounds and some of the interactions again with the HMDB where there's a fairly lengthy connection of high and low levels for certain metabolites and what they mean and how they can be interpreted. Another example was how to identify pathogens by metabolomics. And this is one of the very early applications. These are from 1995, 2001. Tabloids didn't even have a name. It was only on two papers and tabloids played this time. This is some other ones, but these are examples where people are trying to say, can you use what typically bacteria will produce or kick out as they grow or in culture to identify them? Because most bacteria under microscope look pretty much the same. So if you wanted to use metabolomics to identify urinary tract infections, which is one of the classic examples of a bacterial infection, and there are different sources of bacteria. Some are E. coli ones, some are Femona-based, some are this Urabalus one, Eruginosa. Each of them can cause UTIs or urinary tract infections, but you can actually look at the substrates that these bacteria may be producing and some cases by adding appropriate substrates. So some, so this is to urine, but you can add a substrate. And depending on how the case of lactose metabolism is very specific metabolites can be produced. Glycerol, it's very specific to pneumonia. Nicotinic acid is very specific to pseudomonas. And then methionine is very specific to the Morabalus species. So you can probe by adding substrates and identify specific bacteria that may be in urine that are causing or resulting in a specific UTI infection. When you look at biomarkers, and these weren't quite exactly biomarker studies, but there are certain classes of biomarkers that we're interested in medicine. UTI one, we're looking at diagnostic. Someone's complaining of difficulty with urinating. So you go in and you say, well, what is it? And you try and identify the source, you figured it out. So it's whether you have a disease or condition. They're prognostic biomarkers which are basically designed to say, if you have the disease, how will you cope with it? How long will you live with it? How what's your likely resulting condition? The one that I think most people are excited about are, can you get a predictive biomarker? So we've read out your blood, we've read out your urine, and 10 years from now, we predict you're going to develop, you know, prostate cancer or you're going to develop diabetes. And if you know that, you can potentially do something about it. And in the case of genetic biomarkers, the BRCA1 gene is an example. If someone has these mutations, people get prophylactic mastectomies. If you go to 23andMe for $200, you can take a, you know, a genetic test and it'll give some snips and they'll give you some readouts about what your potential disease risks are. So those are the things about getting odds of a given disease. And then there's other ones, markers of response, markers of exposure. So how can we tell if you've been working in a, you know, chloroform formaldehyde plant? Well, we can look for certain compounds that might be in there. How can we tell you if you've been eating bad food or fatty food? Well, we can look at, you know, cholesterol levels and that might be an indication of what's in your body and what you've been eating too much of. We do this for drugs. And this is why metabolomics is of such interest to the drug companies because they are interested in drug toxicity and adverse drug reactions. So those ones, the last two are particularly interest to the pharma world, to epidemiology and environmental science. The first three are particularly interested to physicians. One of the things about biomarkers is that there are different technologies that produce different numbers of biomarkers. Using high throughput DNA sequencing, it's now possible to get up to 2,000 different genetic tests. You can go to a variety of labs around North America and they can do very specific genetic tests because they know certain mutations. So the BRCA1, BRCA2 genetic tests, but they're genetic tests for down syndrome, they're genetic tests for cannabis disease and lesion of hands and all kinds of ones and because they know the mutations. Some of them are like, you know, 600 different tests for cystic fibrosis. So that kind of adds it up. But, you know, some are unique. The problem partly with this is that these are lots of, this is an example of how we're translating genomics into, you know, real tests. But the tests only address about 5% of the disease, of the disease burden because BRCA1 type breast cancer is rare. Cystic fibrosis is rare. A lot of these monogenetic diseases are very, very rare, but we can sure find them. Now, protea or transcriptomics is another one that's been around for 20 years. We've been doing gene chips for a long time. We can use gene chips and there's thousands of publications that come out from gene chips every year. But as yet, there's been very few that have actually translated into tests that are used clinically. There's actually five that have been approved. Maybe two or three that are still available. One that's sort of famous is called the MAMA print, which is a gene chip-based test for predicting or essentially characterizing breast cancer. But there hasn't been a lot of uptake. Proteomics, again, has been around as long as transcriptomics. And there's proteomics labs all around Toronto and everywhere. Despite all of the work they have done and all the facilities and equipment they have, as yet there is no approved proteomic test for any medical disease. Now, there are protein tests, but the protein tests are all done using ELISIS. They don't use proteomics technology. They don't use mass spec. They don't use any of the things that you've learned about proteomics. So there's a big funnel. Genomics, lots of tests, but not many diseases. Transcriptomics for some important diseases like breast cancer, but not many. And then proteomics, zero. Metabolomics, although it's not been around as long, technically has probably produced more approved tests in part because of the fact that newborn screening and inborn areas of metabolism are technically metabolomic studies. They use mass spec to screen dozens of compounds at a time. And they will take blood spots from newborns and look through those and look at about 25 to 30 different compounds and make characterization assessments of whether they are going to get or already have a metabolic disorder. And for research labs in some of the bigger centers, especially in the U.S., they will have some of these tests, a new test added every couple of months as they find out a new compound. And we'll translate that into chemistry. So these are some numbers that I think are surprising to many people, particularly when we think about the huge investments that are going into this. And frankly, many funders, including the NIH, are giving up, feeling that these approaches are not leading to useful medical translations. They're just not. They tell us a lot about biology. They're really cool tools, but they're not translating into human health. What's that? The transcriptomics and proteomics. So here's some more examples of biomarkers where people have been looking at metabolomics. One was the discovery of the University of Alberta with some of our collaborators that you could use urine to diagnose pneumonia. So pneumonia is a lung disease, and everyone was rather surprised that urine can actually tell you something. But there's a close connection between what goes on in the lung and how the body fights infection and what actually happens in the gut microflora and the gut microflora intimately connected to what's in the urine. So they were able to identify a number of markers in urine, and the advantage of working with urine is it's fast, and the analysis is cheap, and it's easy to get urine. Whereas with a normal course, if someone presents with coughing, persistent coughing, they might say, well, come back in a week's time. If you're still coughing on your deathbed, okay, maybe we'll take an X-ray, and then it'll take a while for the X-ray to develop, and then they're in the wrong position. Okay, let's take another X-ray. And then, by the way, we're going to do a sputum culture, and then they have to wait and wait. And about three days later, you might see bacteria, but in about half the cases, nothing grows because the pneumonia isn't bacterial, it's viral. And again, they might still take an X-ray. And by then, because they can't do anything, you either have to get better or you're dead. So that's a pretty lousy protocol for diagnosing one of the more common diseases. And it's one of the reasons why pneumonia is essentially the number one killer of very elderly people. And I mean, arguably, when people die, if they die from cancer, it's actually pneumonia. If they die from Alzheimer's, it's actually pneumonia. If they die from almost anything, broken hip, it's pneumonia. So it's the ultimate cause of death for most chronic diseases. So how do you diagnose it? In this case, they were trying to look at healthy people. It was pretty easy to separate from someone with pneumonia. But TB presents with many of the same symptoms as pneumonia. And so they were looking at patients, and they also looked at people with viral pneumonia and bacterial pneumonia. And then that result was that they were able to distinguish quite consistently between those with pneumonia and TB, those with viral pneumonia and those with bacterial pneumonia, and of course, those with pneumonia and those who are healthy. And this is just illustrating, this is an NMR spectrum, but it could have been a G-C-M-S, L-C-M-S. And you can kind of look at between the TB patient, the pneumonia patient, and the healthy person. There's a big difference right around here. There are other amino acids, other organic acids that are quite substantially different. And so this differences are statistically significant. So these suggest that these could be very useful biomarkers for distinguishing and diagnosing pneumonia in a non-invasive way. Yeah, quite a few actually. So another idea of diagnosis is a case of organ transplant monitoring. Kidney transplants are the most common transplant operation. Thousands are done each year. But whenever a person gets an organ transplant, typically they have to go to have a needle biopsy. This means that they stick a giant needle into your back and take a tissue sample out of your kidney. And they'll do this every few months for as long as you live. And it's not fun, it's invasive, it's unpleasant, and it's also not very accurate. Because what they're trying to do is monitor a situation of are, is the organ rejecting or not? And if it's not rejecting, you're fine. If it's rejecting, they have to change your doses of medicine. In some cases, they have to take the organ out. So sticking needles into your back to take out tissue, costly, invasive, and accurate, is there a simple way of doing it? So people have tried to develop blood tests, they've been trying to develop microwave tests, but the microwave tests still require taking the tissue out. So it's invasive. Ideally, you'd like to have a blood or urine test. And if you think about it, the kidneys produce urine. So we've been looking at this one. This is identifying patients who were found by biopsy to be rejecting their kidneys. And we looked at the compounds in urine that were substantially changed. And there are quite a few. So this is called the VIP plot that you guys learned about. And this is a list of metabolites, carnitines, carnitines, or carnitines, or carnitines, mucos, or carnitines. These are the ones that had changed substantially. So we use PCA, POSDA, everything. But this is the thing that was our list of these in the substantially. Why carnitines? Well, acylcarnitines are produced by white blood cells. And when white blood cells are going through beta-oxidation, and when blood cells are dying, they release the acylcarnitines, red blood cells don't do this. And red blood white blood cells should not be in the kidneys. But if they are, it means they think they're fighting some kind of infection. And they're fighting really hard, especially because they're dying. And so we're seeing the waste products of dying white blood cells. And this is showing up, and seems to show up even before a lot of the symptoms of rejection. And so you can use that set of compounds. This is a relatively small cohort, but it's been done for I think a larger one. These are the normals, people who are not rejecting. And these are the ones that are going through rejection. And there's quite a distinct difference. And you can use a level of significance, and you can measure the sensitivity, specificity from this. This is quite a bit better than what they can do by histopathology. So it's a non-invasive, quick urine test looking for a few compounds that tells you whether there's rejection or not. Last year, there was a couple of papers that came out in nature, nature medicine, about metabolomics. And one that was particularly interesting was metabolomics and cardiovascular disease, particularly atherosclerosis. So all of us, if we're over about 20 or 25, probably have arteries that look like this. These are plaques that are building up, and they build up over a lifetime. Now some people can eat fatty foods all their lives, and will have arteries that look like this. Others can pass by fatty food just smelling it, and their arteries suddenly changes to that. And that's one of the puzzles they've had, is why are people more prone to certain atherosclerotic events. So this one tied things together quite interestingly. So this was published last year. It's done in Cleveland, and they were connecting a fact that if you eat lots of fatty foods, yes, atherosclerosis does go up. But there's also a connection they've noticed between individuals and between gut microflora. And what they found was that choline, bt, and trimethylamine oxide were three compounds that were changed people having, and in rats, with atherosclerosis. Choline is a product of phosphatidylcholine. That's the fat that you get in eggs and fatty foods, less of them. This is cleaved off when the fats come into the large intestine. Bacteria start chewing on this stuff and convert the choline to betaine, and then the betaine trimethylamine oxide. This ends up in the blood. It ends up in the urine. This ends up in the blood. This also ends up in the blood. Now, choline in itself isn't bad, but this one is bad. High levels of trimethylamine oxide actually induce atherosclerosis. They cause plaques to form. So it's a plaque-inducing chemical. Turns out some people have bacteria that produce lots of this, and some people have bacteria in the gut that don't produce a lot of it. So there's some people who could use much fatty food as they want, and they'll never get atherosclerosis because there's not too much TMAO. And other people who have certain types of bacteria produce lots of TMAO, and they're going to be at higher risk. So your gut microflora partly determine whether you're going to get atherosclerosis, but your diet also. So this is an interesting connection, but it's also, I think, an interesting fact about this poison TMAO, which the body actually needs, and it's used to help a number of things with protein folding and stabilization. So it's a vital chemical, but too much of it can lead to plaque formation. And so this is this diagram that they show where fatty food, fatty acids, fatty acids are broken down, choline ends up in the gut microflora, cholines transform to tri-methylamine, and then eventually the tri-methylamine oxide, and TMAO causes atherosclerosis, which leads to heart attack, stroke, and death. Another publication that came out last year, same time, Nature Medicine, a group been in Harvard and at the General Hospital, was looking at the Framingham Heart Study. This is a 50-year-old study where they've been collecting data on patients and their heart health, and they know everything about them medically, and they have huge banks of data and samples, and so they can monitor things about what people do, what they eat, physiological features that put people at risk. And so this is how they've learned that, you know, too much fatty food increases heart risk. And if you go to the Framingham website, they will give you all these predictors where you can type in your heart risk. This has been determined from many, many studies, so it's a hugely important study. It defines what we think about heart disease. But you can also identify people who have diabetes and Parkinson's, and they have all of this because this is a large population. So they looked at the group that was actually developing diabetes, and they took basically healthy people, middle-aged, well, maybe in their 30s, and basically said, um, okay, what's their metabolic profile? Let's take, we'll start with control for people about, you know, height, weight, control for people that are basically healthy. We'll also track a record that they have a family history of diabetes and a few other things. So they got basically two groups that looked to be about the same, but they knew that one group eventually developed diabetes. One didn't. So physiologically they are identical. Um, height, weight, body mass index, age, smoking, drinking preferences, all those things that, you know, think you would put you at risk or reduce the risk, whatever. Um, one court that developed diabetes had a slightly higher risk from, you know, parental or family history of diabetes, but, you know, they weren't that much different. What they did notice is that in many of the people that developed diabetes after 12 years when they presented it as normal, is they had very high levels of branched chain amino acids and very high levels of phenylalanine and tyrosine. So they were puzzled by this, and so they replicated the study using a Swedish cohort. And then there's other work that's been done that's been done on much larger ones, and they may not have found all of these amino acids, but they either find loose seen or they find phenylalanine over and over and over again. As being elevated in people at risk, not just immediate risk, but 10, 15 years from now for diabetes. So they present as healthy, they present as fit. And this has to do with understanding of certainly the branched chain amino acids that typically found in very nutrient rich food, but they also play a role in, in altering the activity of mTOR, sort of a master regulator. And branched chain amino acids are insulin analogs, especially leucine. So this chronic dosing of leucine, or release of leucine, we're not sure what, essentially is like having a, well, it's like someone playing loud music all the time, eventually you get resistant to it. So this leucine creates insulin resistance, which is type two diabetes. So, some of you might be aware of very extensive genetic studies that people have been doing on type two diabetes because people have really thought, still do believe, most people believe that diabetes is a genetic disease, that it comes from, that there is a genetic dependency, people with a family history of diabetes develop diabetes. And they've done millions and millions of dollars, tens of thousands of people with GWAS studies, four diabetes. And the risk increase that they've been able to identify from all the known mutations and polymorphisms can increase your risk by about roughly 25%. This one study, which may have cost, I don't know, $100,000 to do, identified these five compounds in terms of risk factors, they have a 100% increased risk. And then they found another compound more recently that doubles this again. So, yes, there's a genetic component to diabetes, but no, there's a metabolic underlying change, whether it's dietary, whether it's gut microflora, whether it's a physiological thing with how your organs connect and function, but it's driven by these particular amino acids that seems to prime it. And again, remember that type one diabetes study, whereas glutamic acid, which eventually led to the autoantibodies that led to progress. So the triggers seem to be metabolites. And this makes them very good predictive biomarkers. So metabolomics offers, I think, a great opportunity to find biomarkers. And these are some examples. Will these things get into the clinic? Probably not, because a lot of people just don't know how to handle or find or produce biomarkers. So I wanted to show you a little bit about how to do that. Something that Jeff and I have been working on is trying to take lists of metabolites or genes or proteins, and something about their concentrations or relative concentrations, and to look at case control or diseased or healthy or whatever you want. And the website is called Marker Maker. Jeff is working another variation called Rocket, which will do fewer things. But anyways, this one is the idea is to take the data that you've got, and maybe you guys have done some of this, and you may have some, and to try and find a collection of compounds that can serve as very specific or accurate biomarkers for, let's say, a disease. It could be predictive or diagnostic or prognostic. When we talk about biomarkers, we don't talk about PCA or PLSDA. We talk about receiver operator characteristic curves, or ROC curves. And what it is, it's a graph or plot of sort of sensitivity versus specificity or true positives versus false positives. And so if you're diagnosing someone with, you know, a PSA test, how many times are you right? How many times are you wrong? And you can plot this out. If you've done a hundred tests, then you can figure out what your performance is. The best test would look like this. Zero false positives, 100% true positives. That most tests kind of look like this. And it's getting to the point that the PSA test is almost like this, which is flipping a coin or better than, I don't know, better than guessing. So some tests work, some don't. But you really want to see a curve that looks like this. And you can calculate it based on the components, the markers that you've worked on. So disease and no disease. So that's a binary or two-state system. So this is something that was developed originally when World War II, when they were trying to figure out whether they were detecting enemy objects and hitting them. And there was a hit or a miss when they were bombing them. And so this is how they would plot their performance. And so these curves evolved from World War II to be essentially used in medicine to figure out whether you've got good or bad biomarkers. So typically, as I said, an RSE curve should have this sort of log characteristic shape. Poor RSE curve is a straight line. And then the area under the curve is a better measure of the quality of the biomarker. So the maximum area for a maximum biomarker is one. A random one is 0.5. And so if you can find an AUC of better than about 0.8, it is a good indication of a good biomarker. Most medical tests are below 0.8. So these are just some examples of AUCs. So this blue one is a terrible one. That's 0.5. Green one is not very good. But once you get up to the red and the turquoise and the purple one, those are very high ROCs. And so if these were diagnostic tests, these would be very helpful. So Marker Maker, we've given you the URL. This is what it looks like. You guys can go there if you want. It's sort of similar in its structure to what you saw with Metaboanalyst. You can take a set of, well, you can either do ROC analysis of single biomarkers, or you can do collection of a whole bunch of biomarkers and see what is available. You can do regression analysis. You can take those models that someone has described or you generated and see how they perform with a new validation set. From there, you can then select a data set the same way similar to what you did with Metaboanalyst. You can browse the data set. You can click on the data set. It'll upload it. Once it's there, you know, we can press upload and move forward. It'll do the same data integrity check that you saw before just to make sure your data sets. These are CSV formatted files just like with Metaboanalyst. And then you can do some data processing. You can remove some low quality, estimate missing values, try and reduce the quantity of data. And then you can now plot out your marker. So this is a set of Metaboanalysts from the data there that we had. And this is for a disease called preeclampsia. So pregnant women will develop high blood pressure and sometimes this can lead to complications. It's actually the leading cause of maternal mortality in the industrial world. So we'd like to find predictive biomarkers to identify people who will eventually develop preeclampsia. These are taken from mothers who had not yet developed preeclampsia but later did. And we were looking at those at another cohort of mothers who didn't develop preeclampsia. And this is the list of biomarkers ranked in terms of their AUC. So we clicked on here with Colleen and this identifies an AUC with this one marker of about 0.75. I think that glycerol and acetate is somewhere up around 0.9. And it produces an example. And this is one of the things I wanted to show is that, yes, they're different, but what was most intriguing is we found about 10 people. This is the control group that had incredibly high levels of Colleen. And you might recall this connection between Colleen-Vetain and TMAO and heart disease and CBD. We don't think these people actually were at risk. They were younger mothers but they were taking Colleen as supplements. And it turns out the Colleen supplements give you huge amounts, 100 times physiological levels, which may actually be harmful in the long run. So this one is an example of not a real biomarker but a bogus biomarker because we were dealing with supplements. And so when you're collecting from patients, you need to know something about what they eat or consume. Otherwise, you end up with sort of bizarre results. So that was an example for one biomarker. You could also then click on biomarker selection for multiple biomarkers. And so you could create a model where you use not just one but two or five or ten biomarkers to come up with a biomarker profile. And these are some examples. So with that preeclampsia, using a slightly different modeling system, there's a list of about 10 or 12 compounds that if we use them and combine them all together, we get this AUC of about 98%. So that's very predictive. And it actually could be very useful for kind of diagnosis. There's early preeclampsia. There's late preeclampsia. This one is not so much of a problem. You usually get it in like your 33rd week. But if you can predict those who might get it, then yeah, it certainly comes to the pregnancy. There's autism. There's a whole bunch actually on this particular page called tricorder.ca. ROC curves that we calculated using metabolic profiles that are from the literature that we've measured. And as I say, anything you can get above 0.8 is generally better than what's currently offered as a medical test. Yes. Yeah. These are biomarker panels. So you have a list of several hundred. So you're not sure which ones are going to come out. So these are the ones that came out to your best biomarkers. Is it just linear combinations of these things? No, it's somewhat more complicated. So these ones are, this one is a SVM, a singular value, or not singular value, but a support vector machine analysis using a Gaussian kernel where it's gone through a whole bunch of the components and tried to identify those. A marker maker itself uses a linear kernel for the SVMs, but it will also use random forests. You've got, what's that? PLSDA. You'll use PLSDA. You also use decision trees. And the classic way that people generally think of markers is decision trees. If colline is above 0.2 and betaine is below 0.6 and if glucose is above seven, then this. And if it's in this category and then phenylalanine is below 0.2, then it's in this category. And those are things that a doctor can understand, something we can understand. So one of the ideas was to try and come up with algorithms or linear combinations where you could actually write an equation or an algorithm that converts the readings to a diagnosis. So that's done for a couple of them. Can't do it for an SVM, but it does produce the algorithm for it. So you can actually download the algorithm and you can use that as your model. So it parameterizes it. And that's been a problem is that people publish markers and they'll say, yeah, I got this beautiful curve and you can't, there's no description of how they got the curve. There's no formula. So you could go off and measure the same or another cohort and they'll tell you all these markers. But now you've got to recalculate it. And now you're using a different model because they didn't publish their model. So you don't know whether their model will work. And that's the fundamental reason why you have, you know, markers at zero and four for transcriptomics and then proteomics. And if we're ever going to translate omics science into medical applications, we have to make the models available, the formulas available, and we have to make the measurements quantitative. So are these routinely tested now? What's that? Are these routinely tested? No. When for autism is in the states, there's a couple labs that are doing it. But yeah, I mean, this literature is filled with papers of great biomarkers and that's as far as they go. So it's just sort of like publishing to yourself. Okay. So that's applications in clinical medicine, pharmaceutical research, I think I'd mentioned before. Drug discovery. Drug industry has been very interested in it. Drug discovery is a huge gamble. It's expensive. It takes 12 years to develop a drug now. To do clinical trials, it's up to 6,000 patient volunteers. Only one in a thousand compounds that the labs discover actually make it to phase one. Only one in five of the phase one drugs actually make it to FDA approval. And only one in two drugs that are making it past phase three actually are on the market after a few years. The costs, as I say, are close to a billion. Time is lengthy. And so different phases, they'll use different techniques. Chemistry's important. Genomics can be important. Proteomics can be important. But metabolomics actually can be used all the way through. And that's one of the reasons why the drug industry is so keen on it. And metabolomics can be used to help in some aspects of discovery. Talk screening, trying to look for efficacy markers, trying to do safety biomarkers in the early phases. And then clinical safety biomarkers and clinical efficacy markers in phase two and phase three. So every point in the drug development pipeline can be used. And metabolomics can be used. So one of the examples was how metabolomics could be used in drug discovery. And this was a paper that came out a couple years ago about metabolic profiles in sarcosine for cancer progression. And then another one with two hydroxyglutarate for another type of cancer. For glioma in particular. In the case of sarcosine, this is an amino acid, methylated form of glycine. But what they found, and this is metabolon, measured tissues and identified people with metastatic prostate cancer. And high levels of sarcosine are seen to be indicative for metastatic prostate cancer. And that was a complete surprise because no one had ever seen or measured part of sarcosine and didn't expect it. But they went on and they did lots of samples, lots of tissues. They looked in blood and urine as well. And they found it significantly higher in the metastatic cancers and in the early prostate cancers. And the cancer-free ones had no detectable level. The other thing is that when they started doing cool, does it mean anything? So this is when they started doing knockdowns. And then when they did a knockdown of the enzyme glycine and methyl transferase, they found that it attenuated prostate cancer invasion. And then when they did a knockdown of sarcosine dehydrogenase, it induced invasive cancer. So the connection between the metabolite and the protein and the gene was reiterated, in fact, to suggest that sarcosine is probably a pretty strong indicator of metastatic prostate cancer. There's been some debates. Some people have had challenges in reproducing this data. But part of it, I think, has to do with the methods that other people have been using. You can also think about metabolomics in terms of drug discovery. And there's an approach that people have heard of genome-wide association studies. But you can also do metabolome-wide association studies. So in genome-wide association studies, like for diabetes, they look at huge cohorts, people that develop disease, they sequence them thoroughly, and they take normals and also sequence them. And they've started looking at very complex diseases, things that are polygenic that have high heritability, Crohn's disease, Alzheimer's disease Parkinson's, adverse drug reactions, autism. Once they identify those mutations or SNPs, then they can try and see how much these genetic mutations affect disease response. This approach is very valuable. It has allowed us to identify hundreds of genes that are associated with dozens of diseases. But it hasn't actually led to any new drugs. And the reason is sort of the following. You'll do a study. From the study, you'll identify your genes. And from those, you have to find identifiable drug-able genes. Then from the genes, you have to clone a target. From the target protein, then you have to test a whole bunch of drugs. Large chemical library. With that chemical library, if you find a hit, then you have to start going into the preclinical and clinical and phase one, two, and three trials. And then from there, you have to go to the drug marketing phase. So that's the model that every drug company uses right now for drug discovery. So you can do some calculations in terms of the cost and the time and the probability of success. So not every GWAS study actually gives you an answer. Only about 20-25% identify some very clear, useful trend. Something that's statistically significant. Many GWAS studies contradict each other. That's been a real problem. So let's say you found your genes. Then you're pretty confident it's been confirmed by two other GWAS studies. Now you've got to clone your genes. Even if it is a gene and not all of these GWAS studies identify genes, you can't always clone them. Some are unclonable, they're too big, too difficult. And so from the cloning, then you have to try and produce them again. Not every gene that's clonable can be returned into a protein for testing or an assay for testing. Once you've got this, then the hard stuff begins. Because then you've got to do your screening. And this is where libraries and compounds, teams of medicinal chemists screen and test and screen, and they refine the molecules and try again and test. And as they do finally find a couple hits, remember you only have a one in five chance of actually going from this to this, then from this one, from the preclinical set to the full clinical one, one in 500 or one in a thousand. And then once you've actually got your drug, even if it's the best one out there, if there's a competing one, you have maybe one in two chance that the drug will actually work. So multiply your odds, you know, 20% by 50%, by 50%, by 20%, by 0.5%, by 50%, what does it come to? The odds of producing a drug are 0.001%. Total cost, 1 billion. The total number of years right now is 20%. That's why the drug pipeline has dried up. And every drug company will admit to that. The other thing to remember is that not all diseases are genetic. Even the most genetic diseases, like cystic fibrosis, can have a huge phenotypic variability. There are people with cystic fibrosis who run marathons. There are people with cystic fibrosis who die at age 10. What's the difference? They may have had exactly the same mutation. Many major disease-only account for disease genes account for 1%. So the BRCA1 gene, the one that's in the news all the time, it's 1% to 2% that have it. Alzheimer's, the A beta gene, 1% to 2%. The other 98% are spontaneous. ALS, 1% to 2% have a mutation in SOD1. The polygenic diseases, when we've done twin studies, the ones that are highly heritable, like autism, like Crohn's disease, heritability values come in at 50%. For those really well-studied diseases, when we do the GWAS studies and extensive sequencing, we only get a heritability of 20%. So either the twin studies are seriously over-estimating it, or we're seriously misidentifying the genes. Cancer, like the number one genetic disease, somewhere between 20% and 40% of all cancers are actually caused by bacteria or viruses. This is well-known, but not widely discussed. And so when you look at all the numbers, about 80 to 90% of diseases have either an infectious origin or are acquired or wise from a cumulative exposure from the environment. So this has given an rise to this idea of tabillite, today's the drug discovery. So doing metabolic screening to look for these conditions, just like this autism, or preeclampsia, or cancer, or diabetes, or organ rejection, and to look at those compounds that have gone up or down. And those biomarkers we're listing, not all of them were up, not all of them were down, but it's a combination that produces it. But we can go and use the tools you guys just used a few hours ago, look at those pathways and see where these things sit. And in many cases, because we know metabolism so well, you can often find activators or inhibitors for very specific enzymes in these metabolic pathways. In some case, if the thing is down, say if you're short on glycine, and that's causing the, or at least, part of the character of the disease, eat more glycine. I mean, that's a pretty simple solution. So there are some things that you can do either by looking at existing drugs, existing activators inhibitors, or existing metabolites that allow you to actually address this step without having to go through the phase one, two, three billion dollar trials, and actually go to the point where the therapy is being used, and because you can do metabolomics, you can also monitor whether this is actually having an impact. It's also cheap. Metabolomic studies that we've been doing for quite a number of diseases, actually, success rate is about 80%. The analysis, as you guys have seen, is pretty short, even if you're just learning the software. To go from here to here is pretty fast. We haven't had time to show you guys how to go from this to some of these pathways, but it's pretty quick too. It's one of the reasons why we developed Drug Bank, to be able to search with compounds. And if you can identify either that there's a shortage, so if we told you the reason or that you test very high for leucine, isoleucine, phenylalanine, tyrosine in your blood, what does that tell you? You're at high risk for diabetes. What are you going to do? Change your diet. And what's the major source for leucine, isoleucine, phenylalanine, and tyrosine? Well, there are certain foods that are very rich in those. And one thing is typically avoid red meat. Another thing is avoid foods that are rich in phenylalanine and tyrosine. This is what PKU patients have to do, so you follow the PKU diet. And okay, this is a treatment. We didn't have to come up with a drug for you. You could also go to another thing called metformin, which is given to diabetics, but it is something that actually does change the metabolism as well. So that's another drug that you would have found looking from here that alters the metabolism. So you repurpose the drug. So again, no new drug discovery. Just go to your doctor and ask for some metformin. And so now we can monitor. Is your phenylalanine leucine isoleucine level? Is it dropping? Is this regulating? And if it is, okay, what I'm doing is working. If it's not, let's try another therapy or another approach. So this is another root, one that potentially is cheaper, faster and more successful to drugs. Have we not, not with metabolomics? Well, I think there is roots for, I think everyone understands that there is, once someone is diagnosed with Alzheimer's disease or Parkinson's disease, the damage is already too serious to fix. You can't reverse it, but if you can prevent it, that's the way to look at it. And I think there are a number of compounds that they are finding that do seem to reduce the risks. So in the case of Parkinson's, smoking, I think actually reduces it. Coffee drinking reduces it. Chocolate, coffee. With Alzheimer's, they, there's very good evidence now that people who have diets that are rich in curries have substantially lower risks for Alzheimer's. So there are compounds that have been identified and even labs show these effects. And if you go to the sort of the gray literature, find all kinds of people saying, try this, this is my diet and it's helped save my grandmother or whatever. But it's, I think it is arguing. I mean, we're entering an era where drugs aren't being found. It's too costly to develop drugs. And we've gone past that apex of the drug era where new drugs were coming up for everything. So we're going to have to find something else. Sometimes it's useful just to look to what's already there. So this is some examples of what we could and can do and have done. If we lived 200 years ago and we had metabolomics available, we would have been able to find all these sailors coming home with rotting teeth and staggering tissue problems. We would have been able to do a metabolomic study and found that they had very low ascorbic acid. And then we could have said, well, eat some limes. That's why we have limies or sailors. And this is what allowed England to conquer the world. It allowed their sailors to go to sea longer than any other sailors because of this trick of eating vitamin C. More recently, we found that low folate levels in mothers, expected mothers, or mothers planning on having children lead to neural tube defects. So now we have folate supplementation everywhere, almost too much. If you lived in the 1930s and 1940s, these were very common. Many people died or suffered because of with rickets, plegbra, pernicious enumia. It was widespread. It was incredibly widespread. So people didn't die of cancer, heart disease. They died of bees. And correction was to basically they identified low vitamin B, low niacin, low vitamin D. We've supplemented now. And now you do not see these diseases except in some parts of the developing world. Very easy fixes. Another disease that was very common in the 1930s. Now we have iodine salt. The bees again would have been picked up through metabolomics. This is screened now. You can detect this. And now you can use the low phenylalanine diet. And this saves children. Epilepsy. Right now the best treatment for epilepsy actually is a ketogenic diet. It's not the drugs. And essentially what you typically find are at very low levels of ketone bodies. There are other genetic diseases and again there's some very simple treatments that can be done to handle them. And they are metabolic disorders you can read. And some of them manifest not at birth but sometimes later in life. There are other very common diseases from goat, fatty liver disease, hypertension. These are the chronic ones that many of us will have to deal with as we get older. And in these cases there are very specific compounds and metabolites that are very characteristic of these diseases. And there are some very simple treatments to either prevent or help with the treatment of those diseases. And it's again you're just looking at what the metabolites are. Are they high or low? Can you modify them? The ones that are low bring them up. The ones that are high bring them down using mostly off-the-shelf stuff. Some of them are off-the-shelf pharmaceuticals. Some of them are off-the-shelf supplements or foods. Yeah. I think we're just we're not too far from finishing up here. So there's other applications. People do not only in terms of drug discovery but also measures of absorption, distribution, metabolism and excretion. This is how we test drugs. We look at how they're absorbed and modeled. We can measure them through different fluid compartments. What's nice about this approach is you don't have to kill animals. You can actually disample them and you can reuse the animals. You can watch them temporally. And this is where Jeremy Nicholson got started in metabolomics. And they looked at different liver toxins and kidney toxins and they were able to identify very specific markers for toxicity. And this is a graph you guys have seen but this was actually collected on on animals that were treated with drugs or drug like compounds that caused kidney and liver disease. So you can look for very specific markers and not only can you say it's a kidney problem but you can tell exactly which part of the kidney is being damaged based on those biomarkers. So again you don't have to open the animal up. You don't have to do a biopsy or necropsy. This is something that you can get readouts. So these are some of the markers that you'll typically see with whether it's lactate or phenylalanine or various amino acids that lead to damage. There's the TMAO and kineremic acid and xantheremic acid also associated with damage to tubules. Markers for liver toxicity. Again these have been worked out for animals and you can help. With drug usage when they're doing phase two and phase three clinical monitoring people are supposed to take the drug at the same time every day if they don't sometimes the clinical trial fails. So they like to make sure that the people are following what they said and so in this case they can say here's our drug here's the drug oh you didn't take it here's the drug here's the drug and then go back and say you know maybe we should exclude this person from the trial or maybe it explains their unusual behavior. Likewise when you're on drug trials you're supposed to go alcohol free not supposed to take other drugs again you can see this individual but then you can see on day three they drank a lot and it's showing up and so in fact that may affect the drug efficacy. These are things that people don't tell you but the metabolone does. Their applications in in food and nutrient analysis people are looking at foods and trying to decipher what's in them. Of course we have you know the cereal box lists and other things but that only gives you information on macronutrients there's some very specific micronutrients that are important. So there are some simple foods like Gatorade in the case of beef there's about 6,800 compounds in a slice of steak or hamburger. Meat is different than fruit on very very different sets of kinds of compounds and so understanding what's in these foods is part of what's in food DB. Some of the food you get hasn't been treated correctly. There are cases where people will modify grape juice apple juice they can adulterate with corn syrup they can adulterate with sucrose and you can actually detect these things and there are a number of cases where people have identified foods that have been a poisoned or be changed and sold at very high market value when in fact they were basically fake. You can see if there's beet sugar or corn syrup again you can distinguish between the two different sweeteners and there's some strict regulations about what can be added to food but no way of easily testing them but Metabolomics gives you that possibility. So not only can you figure out what's in food or what shouldn't be in food or should be in food but you can also see how food changes our bodies and the types of foods that we eat and what it might do to phenotype. So fruit vegetables the vegans are thin and skinny people eating meat are supposed to be lean and muscular and people eating fried food are supposed to be fat but in fact you can sometimes get completely the opposite. Why? Well there's some issues we'll explain but there are certain markers so normally we try and take diet questionnaires and say what did you eat today and one of the things they found is that extremely obese people under report what they eat by a factor of two and this is something that completely puzzled and completely messed up food epidemiologists for 20 years. Likewise let me ask what do you remember eating and what do you remember eating for the last week most people can't. So if you could actually get a readout of what people are eating by looking for markers in blood or urine then you can actually get a pretty good idea a very precise idea of what they've been exposed to what they're eating how much they're eating and also how their body's processing it. So some people can have just a you know half a cup of coffee and they're wired for the entire day others can drink six cups of coffee at midnight and fall asleep at 12 30. That's part of your metabolism that's also how your body interacts with food and that's not something that you can read from a diet questionnaire. What we eat also influences what is in our gut and that chronic exposure can obviously change the types of things that are in the gut and will change particularly what we'll see in urine and so again one of the big worries is obesity and diabetes and since 2006 and there's considerably more data now but there's association with the gut microbial and the types of bacteria that you have and whether that will lead to a life of obesity or a life of being largely lean and that set of bacteria actually is determined in the first year of life and it sticks with you largely. It can be changed sometimes dramatically where people have had very intensive doses of antibiotics but apparently breast milk is designed primarily to has all kinds of sugars that your body can't consume but they are sugars that certain types of bacteria consume so they're intended to colonize certain classes of bacteria so people when they've noticed this that were raised on formula have a generally higher incidence of obesity than those who were on breast milk and that's because the formula lacks those special sugars which lead to the good bacteria from inhabiting the gut so they've identified certain nutritional phenotypes lane homes and where they've distinguished phenotypes in China and Japan and England and in the U.S. you can see that England is basically the same as the U.S. but there's very distinct separation between Japan and China these correspond to the foods that they eat and the types of compounds you would see in their urine this is another one that was done in Italy where they looked at a whole bunch of people and found that yes there's considerable variability in across the spectrum of people but individually our metabolotype doesn't change a whole lot so this is collected I think over six months of urine and they're comparing them and they were 25 people they all separate it as individuals but so they know none of them overlapped and they were all in terms of the variability within their metabolomes were much smaller than the variation across the metabolomes so it's a fingerprint it sticks with you for life what's in your blood what's in urine what's in saliva stays with you and that is as unique as your fingerprint on your fingers so what are we looking for and ahead in terms of metabolomics this is some speculation but there's some things that are already happening one of the most exciting parts about metabolomics is the idea of metabolic imaging we normally have done microscopy with staining but the idea of essentially doing chemical composition and imaging both at a macroscopic and microscopic view is really exciting so with magnetic resonance imaging and magnetic resonance chemical shift spectroscopy we can wipe out about a dozen metabolites in the brain and certain tissues pet scanning also was a metabolic imaging technique typically with glucose or a few other small molecules you can also take tissue samples and you can do histology using moldy imaging and if you layer tissues out you can do moldy and collect mass spectra over certain regions and specific cells and then read out the mass spectra and identify certain populations of metabolites so instead of coloring tissue by by a stain you're coloring a tissue by its metabolic composition so this is the way that moldy chemical shift imaging is done just to take your tissue sample raster out where your laser is going to hit collect the spectra at each raster point calculate some of those spectra and then map those spectra and the metabolites over the tissue sample and then you might superpose a picture of the tissue sample on top of it and then you can actually see where these metabolites are and now you're generated not just a single picture often it's a dozen or a hundred pictures of different concentrations shifting around in the tissue said metabolomics is often expensive thing we use big gcms lcms nmr instruments that's one of the things that makes it sort of inaccessible to people physicians don't want to figure out how to operate an lcms or an nmr but if you could give them something this size they would it would be something that physicians and clinicians would use and in fact there is a technique a tool called the i-stat which uses microfluidics and tests about 10 or 15 compounds now it's not metabolomics a lot of the stuff is you know blood gases and sodium and a couple other things but it conceptually is is what people would like to see i think people have also heard stories about how you can use breath to sometimes diagnose diseases so these are things called electronic noses these are also handheld devices and they detect volatile compounds they don't identify and quantify yet but they're getting there and this is something that is quite compelling so this is another version of the enos and again they're using sort of headspace chromatography and there's certain patterns that they can detect that seem to be characteristic of acetone or chloroform or benzene so they are used in industrial centers where toxic chemicals may be around and they help monitor whether it's a toxic release but they're not quite at the stage where someone could breathe on it and you would get diagnosis so to wrap up and we're almost on time or two minutes ahead of schedule trying to integrate metabolomics with genomics proteomics bioinformatics is critical i think what we've seen is that metabolomics is also a very good tool very good technique for modeling modeling cells modeling metabolism we've seen it with the flux balance models that poulsen has done they're they're actually world famous i think we're seeing a lot of the applications metabolomics being used in many areas of life science i've been focusing on medical applications but there's obviously many others you guys have seen it where you're looking at cows there's examples of plants there's lots of work with microbes every field of life science can be impacted by metabolomics i think the other thing to remember is good data comes from good experiments garbage in equals garbage out if you are attentive to the quality of the data tended to doing qa and qc if you understand some elements of the statistics and principles which i try to pass on to you then you can do good experiments in terms of the emerging trends the last points we're seeing i think greater levels of of imaging pushes to imaging and greater levels of automation and miniaturization and i think these are important and if we want to be able to sort of match the throughput that you can achieve with proteomics and genomics that's where metabolomics needs to go um a lot would call the volatilome or volatone this is something that's a new area of metabolomics so detecting those volatile compounds that give something it's it's aroma it's smell um the food or perfumes but also the things that are critical sometimes for diagnosing what's going on in the lungs i've pushed this and i'll push it again quantitative metabolomics is something that many of the leaders in the field are are really advocating so i'm not alone in this it's a difference and it's one of the things that could differentiate metabolomics from the other omics which have struggled because they were not quantitative micro rays are not quantitative proteomics is not quantitative except now in victoria but it's the case where because they weren't measuring absolute concentrations people could take it to the clinic because all the experiment was good for was that platform alone and if that platform is not moved to another building or instrument or someone else is using a different platform they can't replicate it the other hand if you can quantify precisely and everyone's agreed that it's micromolar and we all agree out of measure micromolar um anyone can use it any country any lab any facility smaller kits uh devices portable systems that's also emerging i've talked about the biocrities kits i've mentioned the i-stat these are things that are coming online and the last part i think the integration into more diagnostic prognostic and predictive applications given you some examples that have been published some hints and some tantalizing data and so perhaps the task is for you to see if you can take some of those ideas and translate them to practice so metabolomics is the youngest of all the omics sciences and well genomics still continues to go strong it's now a commodity so we don't have to do a whole lot of technology development we just say let's sequence so there's lots of data to work on but really the the challenge in genomics is bioinformatics it's not the genomics proteomics is still developing but we're also getting that you know understood more my prediction is the metabolomics because it's still the younger one there's still lots to do and technology development still lots to do in bioinformatics there's still lots to discover there's 200,000 compounds that we have yet to measure that could have a role to play in all kinds of things in in cells plants insects mammals disease and etiology um in processes so i think the prospects are bright and there's still a lot to be done so if you're just entering into metabolomics metabolomics i think you've got a long and prosperous career ahead of you so with that i want to thank some people Jeff for helping TA and prepare some of the slides and Michelle for setting this up and keeping us on schedule and feeding us and some of the funders like Genome Canada which is paying and subsidizing some of this and you saw some of the other sponsors as well and then some of the work we're doing has been supported by Alberta Innovates including i think Jeff's scholarships historically uh so thank you very much