 Okay, thanks, Anne. Yeah, this is normally supposed to be something we give at the end, but just the timing today looks like I might not be able to stay for the very, very end. This also gives Jeff a chance to rest because he's going to be talking for the next four hours as you guys go through my table analyst. So this is a more speculative presentation, a little bit about the future of metabolomics. A lot of you are attracted to metabolomics in part because it is a growing area. This is just a plot of the number of publications over the last 18 or 19 years. The very first mention of the word metabolomics I think was about 1999, metabolomics, 1998 or 1999. And so for a number of years, there's just one or two or three papers at most, mostly coming from Jeremy Nicholson's lab in Imperial College, but then others started joining in. So growth is almost exponential, which is nice to see. So you're choosing a good area to be involved in. But the question is, will it continue going, rising in the next years? Will it flatten out, which has already happened in terms of if you look at the same graphs or plots for the world of genomics where it's largely flattened out? Or will it plummet down like in the red curve, which is actually what's happened in the world of proteomics and structural biology. So ultimately, where that error goes over the next few years has to do with what we can do as either a community or you as scientists to get through some of the bottlenecks. So there's about five or six bottlenecks that I've identified. There are probably others that people might comment on or suggest with respect to metabolomics. So one is a lack of automation. If you look at the field of genomics, just about everything is automated. Sequencers, sample prep can all be roboticized. The throughput that DNA sequencing labs is able to achieve is astonishing. There's obviously lots of automation in many other fields of science. Robots are pretty much everywhere. But those of you who do metabolomics understand it's pretty manually intensive. And there's lots of work, both in terms of sample prep, data analysis, and then the data interpretation. And there's lots of variability. And this course is trying to help people standardize some of that, but that's an essential problem in metabolomics. Another one is the incomplete coverage of the metabolome. So I've highlighted that a bit yesterday, but often people are only able to get about 100 compounds, maybe 200 compounds identified. Yet we know there's more than 100,000 compounds in most metabolomes. So we're just getting a tiny fraction. And compared to genomics or even proteomics, we're doing quite poorly. Another challenge in metabolomics is a very expensive and large, physically large equipment. So someone to play in the sandbox of metabolomics, your advisors or supervisors have had to have a raise somewhere on the order of $2 to $3 million to get the equipment. Core labs, $20 million. Most other scientific endeavors, people can get underway for a lab at maybe $100,000. So the entry fee is generally very, very high. Another point, especially as you try and move metabolomics from discovery into laboratory analysis, clinical lab analysis, veterinary analysis, agricultural analysis, you have to quantify. And currently, there's very little quantification going on in the field. That lack of quantification means that we haven't been able to translate many of the findings in metabolomics to the clinical lab or to veterinary labs. And it also has meant that we haven't been able to make metabolomics matter very much to drug companies. And drug companies have hundreds of billions of dollars in revenue, but are also willing to spend billions in R&D. And they haven't, particularly in the area of metabolomics. So unless we can address those problems, metabolomics will take the red curve rather than the blue curve. So I'm going to talk about some of the things that are going on, some of work that our group, Jeff's group, working on, other activities that are going around in core labs around the world. So talk about automated metabolomics. I'll talk about expanding coverage. I'll talk about making metabolomics more portable, efforts for quantification, and then also efforts to move metabolomics from the lab to the clinic and into the drug development world. So while most people in most labs try and do metabolomics in a very manual way, there is already and are already tools and systems to do automated metabolomics. And it surprised me, I think, just to find a number of people who didn't know about some of these systems. So one route that makes metabolomics very automated is to work with kits. So Biocrates, which is an Austrian company, has made kits available for a number of years. They're very easy to use. They work on a number of platforms. They give quantitation, absolute quantitation. They're very reproducible. There are other groups. Shimatsu is making kits. The metabolomics Innovation Center is also making kits. And over the next year or two, I think we'll see a large number of kits rolling out for metabolomics. And it's not totally automated. You know, throw the kit into your machine and it automatically opens itself. And no, you have to do a little bit of work. But what used to take literally months can generally be done in a day or two. There's Kinomics, which some of you guys have heard about or learned about. That is commercial software and it's now semi automated or in some parts almost fully automated to do spectral analysis. Brooker has produced NMR instruments that are able to do automatic analysis of juice, of wine, and of lipoproteins using NMR. And again, it's pretty automated in the sense you just have to get the samples and press go and the rest is done for you. Syax is made, what's called a lipidizer. It's a mass spec designed to be something like a black box. It can measure over 1100 lipids quite accurately, quite consistently. Again, it's not perfectly automated, but it's a whole lot easier than what most people typically do. So again, I'll just ask how many people have ever used or heard of a lipidizer? One. Okay. How many, at least prior to yesterday, had ever heard of biocrities? One, two, three, four, five. How many people had prior to yesterday heard of Kinomics? Two, three. And then Brooker and its activities and automated metabolomics. Just one. Anyways, now you guys have heard about all of them, and I think if you want to talk to your supervisors or work in, eventually, establish your own labs, these are things you should really consider. It makes your life a lot easier. We've also introduced to you some of the non-commercial automated tools, Bazel. It's one that you guys ran yesterday. It's not really suited to have 15 people hitting it at exactly the same time, but there are efforts to make it even faster. Manoche is working with Mark to make a new version of Bazel called MagMint, which should be about two to three times faster and generally more accurate. I've talked about other tools. I mentioned them, things like R-Dolphin Batman. These are other NMR automation tools that are making metabolism metabolomics by NMR very fast, very automated. We tried GC-AutoFit yesterday. There are efforts I think in other labs to try and automate GC-MS to make it more quantitative. I think almost all of you were able to get results, and all of you were able to see the number of compounds that are typically identified with an automatic GC analysis. So I think the answer is, at least in terms of the trends and the bottlenecks. Automation is here. It's getting better every year. If you're still stuck doing largely manual work, you are in the stone age. Metabolomics can and should be fully automated. Another issue is the expanding metabolome coverage. As I mentioned, untarget metabolomics typically identifies less than 2% of the detected peaks. Most published studies report fewer than 100 identified compounds. We know that there's more than 100,000 in most mammalian systems. Likewise, most published metabolomic studies do not achieve a metabolomic standard initiative, level one, which is the highest level verification by an authentic standard. Most actually don't even achieve level two, which is matching to multiple spectral inputs. Terms of coverage relative to genomics and proteomics were about 100th of what is achievable with the genome studies. The result is whether it's clinicians, veterinarians, drug companies, even funding agencies don't trust omics methods that don't provide comprehensive coverage. Now, we know that there are different tools that can give more or less coverage, the most sensitive being LCMS, DIMS, the least sensitive being NMR. People still use NMR because what you can find is clearly identifiable and you can report almost 100% coverage of what is detectable. Of course, it's still a small number out of the total known, but it still allows you to report on specific compounds. Now, in the field of metabolomics, whether it's some of these automated tools or whether it's service providers, the coverage is getting better. So the biocrities tests that I mentioned typically can get about 160 compounds in serum. The test is designed for a little over 185. There's a new version of the biocrities test or kit. It measures 408 compounds. About 370 can be typically detected in serum. Metabolon, the numbers keep on increasing, but on average, they can detect about 500 compounds in a serum sample. They describe more, but most of them don't have compound names. So the total number that they will measure in a metabolome study is 1,000 plus. And then I mentioned the lipidizer. If you have access to one, you can generally get up to 1,100 lipid species. So if you're content with just reporting between 50 and 100 compounds, even if it's an untargeted study, again, you're in the stone age. There are both kits or systems that can get up into the hundreds or thousands. And these are fully identified. In many cases, fully quantified. Now, what are the unknown unknowns? These things that make up, we call the metabolomic dark matter, the 98% that you can't identify with your untargeted study. The current thinking is that many of these are metabolites of metabolites. These are compounds that could be food, drug, or endogenous metabolites that are transformed by your liver, by the gut enzymes, by the microflora. This is sort of illustrating the things that happen. What we eat, what we're exposed to, the drugs we take, they're taken in and transformed by our host, metabolism, our microflora, our digestive enzymes. And so those lead to metabolites of the food, metabolites of the drugs, metabolites of the exposures. And then accompanying that is this smaller section of the endogenous metabolome, which in humans might be 60 or 70,000 endogenously composed or known metabolites. Now, one way of trying to get better characterization of these unknowns is to sort of systematically collect all of the metabolites, synthesize them, prepare them, and then pass them through all the different types of spectrometers to get different reference spectrum. And I've mentioned this before that if you look very, very hard and if you have an unlimited budget, you might be able to buy about 2,000 compounds for about $1 million, which would cover or would correspond to metabolites that you find in the body. Again, 2,000 out of 100,000 is still 2%. So how do you get the other 98,000? You would have to pay chemists a lot of money to make them. And one estimate would be that it would cost about $5 billion to have the other 98,000 compounds synthesized and made available. That's not going to happen. Governments aren't going to fund that. So there's been a trend now in metabolomics to go to something called in silico metabolomics. This is to help identify the unknown unknowns. So we aren't going to have $5 billion. We aren't going to ever collect the million or so NMR or MS spectra that are needed. What we can do is use computers. We can use computers to predict these biologically feasible metabolites, the metabolites, the metabolites. And there are tools. There's tools called mines or software like Meteor Nexus. There's also a tool called Biotransformer that's been developed in our lab that will take compounds and rationally predict with good accuracy what they should look like after going through different transformations, knowing what enzymes do and what substrates are, where the active sites or sites of metabolism are. If you can predict most of these structures and some you will overpredict and some you'll underpredict, at least you have a body of chemical information and then you can use in silico prediction of observables. So what we observe in a metabolomics experiment is a retention time. We observe the NMR spectrum. We observe the mass spectra, the NMOPRZAD values. We can look at the collisional cross-section area. These are observables. And so again there are tools that accurately predict those observables. So you've got predicted structures and predicted observables and you can start matching to what you actually see. Now in terms of systematic spectral prediction, I mentioned briefly yesterday this tool called CFMID. So it does both spectral prediction for MSMS as well as for EI or GCMS. It's quite a bit more accurate than other things that have been done. It uses machine learning, probabilistic graphical models or hidden Markov models. But it is being used by many groups to help with this in silico metabolomics. So if it's not clear, the idea is to take the known compounds that are cataloged in various databases. I've given some examples like drug bank and HMDB, use keg or kebi. And then you run them through these tools like meteor nexus or biotransformer. So if there's 100,000 compounds in HMDB, the estimate is that the number of predicted compounds that would come out through a tool like biotransformer is around half a million, perhaps upwards of a million. Taking those predicted metabolites, and in fact there are in the latest version of HMDB about 8,000 predicted metabolites that we have pretty high confidence that they exist. You run them through programs like CFMID, which predicts the GCMS and EIMS spectra. And currently in the HMDB then there are almost 300,000 predicted spectra for EIMS and GCMS and LCMSMS coming from CFMID. You can also use other tools to predict retention time, to predict the collisional cross-section area, the NMR spectra. These are the observables. So you now have a reference database of compounds, compound names, their spectra. And then you can start matching those observed spectra that you're collecting in real time on your real experiments with the predicted ones. And it's not going to be perfect, but it'll certainly narrow your scope down from, oh, it could be anything in PubChem to, yeah, it's got to be one of these three compounds. This is being done on a small scale in a number of labs. And so it certainly suggests that in silica metabolomics is a feasible option. Certainly if we're going to expand coverage it's going to be important to continue to collect and synthesize authentic compounds wherever we can to also prepare isotopic standards for those so that we can quantify them. Some of the metabolite prediction software is very expensive. I think an effort is trying to make these tools freely available. Biotransformer is one that is freely available. In silico observable predictions, these are becoming freely available. Unfortunately, most of the NMR predicting tools are expensive. So again, it would be nice to see freely available NMR tools. I think it's important to prove that the concept of in silica metabolomics works. To say a few labs are showing it can be done, but it needs to be more generally proven. And that if we can expand the coverage and that the current coverage that we are obtaining is actually adequate, then that might convince more and more people to fund metabolomics, but also support it in their applications. As we know, metabolomics instruments are expensive and weigh a lot. I don't know if anyone's tried to lift a mass spec on their own, but they're not exactly portable. I see some of you have things like smartwatches and Fitbits and things like that. These are examples of tools that are very portable that are monitoring maybe not your chemical composition, but a fair bit about your body. There are applications as well to make systems that you can attach to your body that will do things like continuous glucose monitoring or swept evaluation. As I said, currently most metabolomics instruments are pretty unwieldy. This is a picture of an 800 megahertz, maybe it's a 900 megahertz NMR spectrometer. It is enormous. It has its own building. It's about 10 meters tall overall. And it costs about $10 million to put the whole thing together. Minimum costs to do the NMR spectrum on this are about $200. There is a tool called the iSTAP, which does essentially portable metabolomics. It's been around for about 10 years. You can buy it for about $1,000 or $2,000 and it's about $2 million to test. So that's an example of an already existing metabolomics portable device. Qualcomm has been offering, I think it was a $5 million prize for the tricorder. So if you remember Star Trek, they could use tricorders to sort of scan through you. The idea was to make essentially a portable diagnostic device. And in fact, the prize has already been awarded. It doesn't do a lot of metabolomic measurements. It's more like blood pressure and heart rate and stuff like that. But it does do a little bit of chemistry. If you're into and have been following activities in nanotech and microfluidics and micro technology, you should be aware that there are chips. You can do GC on a chip. You can do electrolyte electrophoresis on a chip. You can do HPLC on a chip. So these are appearing. So that certainly will miniaturize what we do, at least in terms of separations. In terms of volatile compound measurements, I don't know how many people actually measure volatiles here. No one. So again, this is an area of active interest in some of the newer more advanced metabolomic slabs. The reason why volatiles are so interesting is that 90% of what you taste is not from what's in your taste buds, but it's what you smell. So if you have no sense of smell, you can only taste four or five different tastes. You know, sweet, sour, bitter, and so on. But most taste sensations are actually coming through your nose, mostly through volatile compounds. Obviously there's also interest in pollutants and other things. Many of those are also volatile. And so electronic noses are being developed, which can use combination of artificial intelligence and some advanced chemistry to identify specific volatile compounds in relatively short order. We've been working on a different model, which is for water soluble metabolites, but it's trying to make metabolomics doable on paper strips. So paper is a little cheaper than a mass spec. And the idea is that using simply your cell phone, you could take pictures of the paper strips. The phone would have a small applet or an API system that would allow you to read the colors and convert the colors into concentrations. If you've ever seen a set of, there are these urine strip tests, which again have about seven or eight color pads on them. You can dip the paper strip into urine in about two or three minutes to get a bunch of different colors. You can pair the colors to a chart, and that can give you information about glucose levels, bilirutin, biliviridin, ketone bodies. So the concept of a paper strip metabolomic test is not entirely new. What is new is trying to create combinations of metabolites on a single strip that correspond to a diagnostic panel. So this one was measuring carnitine succinate and ascorbate is specific to detection of polyps. So these compounds, if they're at certain levels, allow you to identify individuals who have polyps, colonic polyps, which are precursors to cancer. So this is a cancer test specifically. So it's a paper strip cancer test. There are other approaches, which rather than using enzymes and color dyes, which are typically used in paper tests, is to use things like antibodies or aptomers or metabolite transportation or transport in proteins. So these identify very specific metabolites and will bind to them. If you use gold nanoparticles and label the metabolites, you can actually get these proteins to do some interesting things in terms of detection. When you have metabolite weighted with this gold nanoparticle, you can change things like impedance or the refractive index, which you can measure with surface plasmid residence. You can also have it modify the Raman spectrum. So these techniques essentially allow you to make a very small system that uses largely electronics to measure metabolites. So portable impedance sensors, portable SPR systems are now being made. And the metabolite recognition elements are either antibodies or paraplasmic binding proteins or aptomers. This is depicting an impedance sensor. It's a large one, but in fact the one on the lower right corner is actually one that's already been made. Very portable. It allows you to measure up to eight different metabolites in this little chip that's slipped in, and it produces readings. The concept again designed similar to the tricorder, and these are just showing some examples of the impedance readouts for different metabolites. Impedance is resistance, but it's for AC current as opposed to direct current. So this too is already in existence. It's not yet commercial, but it is being produced by a small company at Edmonton to do portable metabolomic measurements. It also works well with proteins, with DNA, RNA, and even cells. So it's a very general concept, and it's obviously a lot smaller and a lot cheaper than a mass spec or an NMR. Quantification is another thing that has been, I guess, a burden hanging over the necks of many people in metabolomics. More than 90% of the metabolomic studies that are published are either semi-quantitative or non-quantitative. People may quote things like relative peak areas or intensities, but they actually don't quote millimolar or micromolar. Less than 10% of the published metabolomic studies that are out now actually use absolute quantitation. As I say, most of the core labs have now moved to quantitation, so if you are not trying to quantify, yes, you are with the majority, but that's not where we should be. It has to be much more quantitative. And there was actually a cover article in Trends in Biotechnology calling it a specter hanging over metabolomics. It's the specter or ghost of quantification basically saying that that is our central problem. So that article talked about some of these issues. We've already talked about some of these automated tools for doing metabolomics, and interestingly to automate metabolomics basically means you have to make it quantitative. So to come hand in hand, so the same tools that are being used for automation are also the same tools that allow you to do quantitation. Same sorts of tools that we ran yesterday, Bazel and GC-Ottafit also do quantitation. And there's, I think, the focus on quantitation is really one of the great strengths of metabolomics, and it's one area that tends to be, I think, underestimated. So I'm comparing metabolomics with proteomics with transcriptomics and essentially the number of things that have been absolutely quantified. Now, is there little dated, but in the case of serum and plasma, there are 288 compounds identified and quantified. As far back as 2011, with the P400 kit, it's now possible to get 370 compounds. So with cerebral spinal fluid in 2012, 172 compounds identified and quantified. In urine, it was about 380 compounds identified and quantified. Now people doing proteomics on the same samples where they've actually officially quantified them, not just simply said they're there, are about one-third to one-half of what's recorded by metabolomics. Now in transcriptomics, RNA-seq or micro-rays is actually impossible to quantify accurately. And so as much as been produced and published, there's essentially no absolute quantification from serum, plasma or CSF or urine. So intrinsically, metabolomics is highly quantitative, and it is intrinsically so far better than proteomics and quantification. Its coverage isn't as extensive, but at least in terms of quantification, it is. And that's a message that's generally lost, and it's lost partly from the cacophony of so many publications in metabolomics not being quantitative. If more and more people could go to automation and quantification, it would certainly give metabolomics an even stronger voice when people try and approach clinicians, vets, and other applications-oriented groups and say, this is reliable. This is quantitative. This can be used in a translational application. So the quantification is this key thing that allows or would allow metabolomics to go from the lab into the clinic or from the lab or into the field or from the lab into a veterinary platform. Now, Jeff has talked a little bit about ROC curves and biomarkers. You guys are going to learn more about that in Metabolanalyst later this afternoon. But over the last 45, 50 years, there's been more than 700,000 biomarker papers published. You can find them in PubMed. I haven't read all of them, but of the biomarker papers, we know that fewer than 250 biomarkers have actually been approved for clinical use in that 45-year period. We also know that in the realm of proteomics, no proteomic test has actually been approved for clinical use. There are a few transcriptomic tests for clinical use, and there are a few, well, about 100 or so genetic tests approved. One thing that most of you guys wouldn't know is most of you had this when you were a newborn. But just about anyone under the age of 25, maybe even anyone under the age of 30, has actually had a metabolomic test. And that's part of newborn screening. So anyone born in a developed country has probably gone through this. This is taking a small blood spot from the heel of your foot and running it through the mass spec to detect anywhere between 10 and 30 different compounds, which is essentially a genetic test to identify people with things like PKU or alkyptinuria. Even the most genetic of genetic tests, cystic fibrosis, is still done, analyzed using a metabolomic test as a gold standard. So it is very widespread, but again, most people don't know this because they were only a couple of hours old when it happened to them. How many people have actually had a genetic test where they've actually sent some DNA to Ancestry or 23andMe or any one of those places? About one, two, three. And so it's typical, about five to 10% of the population has tried some kind of genetic test, but I think we can get everyone here. Probably 90% of you have had a metabolomic test. So anytime if anyone asks, have you ever had a metabolomic test put up your hand because you have? I mentioned these numbers yesterday, but because metabolomics chemical analysis is very quantitative, the fact is at least in Canada, there are far more clinically approved tests for metabolites than there are for genes or proteins or transcriptomics. And in essence, this proteomics test they marked here is now no longer one, it's zero because the test was ported over to an ELISA. So again, even though people may say this and many people believe this, metabolomics is in the clinic, it has been successfully used in the clinic. And there are many other examples where metabolomics can and should be used in the clinic because it is uniquely quantifiable. And if we don't exploit that feature, then we undermine our research. So you can also assess metabolomics for other conditions, not just for newborn screening. And these are some examples that we've done in our lab with different projects. So we've been looking at trying to predict diseases. So we've been looking at mothers or expectant mothers in their first trimester, taking blood samples and use the blood to see if we can identify which of those mothers or expectant mothers will develop preeclampsia and whether they'll develop early or late. So these are rock curves which Jeff mentioned. And generally a good rock curve, many diagnostic tests have a rock curve area of about .75. So these ones have areas around .96, .99. You can also look at congenital heart defects again in the first trimester. And there seems to be a metabolic fingerprint that you can detect in the mother. So this is not amniocentesis, it's just a simple blood test typically done when the mother comes in for her first checkup with or expectant mother comes in for her first checkup. You can also even do genetic tests, looking for trisomy 18 and 21 using blood. Now these are probably better done through amniocentesis because those have a higher rate of detection, but you can't do amniocentesis in the third month. So this potentially does provide some screening. So for predicting things with expectant mothers, metabolism seems to be remarkably good. It also allows you to predict which individuals will develop a condition known as cancer catechia. So catechia is a muscle wasting condition that happens particularly with things like lung cancer, colon cancer. It's actually the major killer in cancer. Some people just call it dying from tumor burden, but it is the thing where your body metabolism shifts over profoundly. People who develop catechia, about half cancer patients to do, die much, much sooner. Whereas those that don't develop catechia can live for many, many years with cancer and essentially it's sort of like living with diabetes. So you'd like to be able to predict which people long before the cancer has gotten full blown will develop or could develop catechia because then potentially you can have interventions changing their diet or introducing exercise routines that would help keep their muscle mass. So from a urine test it is possible to reasonably accurately predict which individuals will develop catechia and which ones won't. You can also use metabolomics to diagnose diseases and things like in this case rejection, transplant rejection. You can look in a urine sample to identify which patients are going to reject and which ones aren't. Essentially everyone rejects an organ and so all you are doing is just trying to mitigate rejection with various anti-rejection drugs, but sometimes people act too late and then eventually the organ dies and they have to find another replacement or the person dies. Current way for doing rejection testing with kidneys is to stick a giant needle in your back and they take a segment of the tissue from the kidney out and then do a biopsy analysis on the histology. And even when they do that they are only about 80% right. If you could just do a simple urine test clearly you'd be able to detect those who are rejecting with much greater accuracy. There are other conditions, heart failure. There are two types of heart failure, systolic and diastolic heart failure. They are hard to distinguish. It can take up to a week to figure out what is going on, but if you could do a simple blood test you could figure that out quite accurately within a few hours. There are other conditions, chronic fatigue syndrome. I think there is someone here also working in that area, esophagitis. We did a sample analysis of that yesterday. Again these can be very accurately predict or diagnosed using metabolomics. And these are examples of conditions that are hard to diagnose using standard phenotypic assessments where metabolomics gives you some clear answers. I'd mentioned colonic polyps. This has the worst rock curve, but even at 0.78 it's considered good enough to actually be used in the clinic and it is used in the clinic in the US. And it's a simple urine test. Currently if you want to look for someone to see if they have polyps you actually have to do a colonoscopy if anyone's ever had one. Those are one day exercises that take a couple days in advance and a couple days after to recover from. They're not exactly pleasant and they also cost about $1,000 to $1,500 per test. So if you could just do a simple urine test, non-invasive, a few minutes, even if it's not perfect it still allows you to identify people who should or should not have a colonoscopy. So metabolomics at least from these rock curves tells you that it does extraordinarily well in both predicting and diagnosing diseases. These are quantitative assays that were developed. And so they could easily be ported into metabolomic tests. So this is already happening. Metabolon has introduced three types of metabolomic tests. Stamina is another company that has an autism-based metabolomics test. And then MTI and Edmonton based company also has this polyp test. So these are combinations of specific metabolites that are disease specific which can be delivered or measured through mass spec quantitative systems. So I think these are useful examples of how a gain metabolomics can and is moving into the clinic. It's not doing it fast enough as far as I'm concerned. And a lot of that tardiness has to do with the fact that people are not fully exploding like the quantitative nature of chemical measurements. Most people here haven't been around metabolomics long enough but back in the late 1990s and early 2000s most metabolomics was actually done in drug companies and not in universities. And there's great interest in using metabolomics by drug companies as a measure of toxicology. But they kind of lost interest and they lost interest largely because the metabolomics at that time was not quantitative. The coverage was relatively insufficient and the instruments were too expensive. So obviously these things are critical. But if you look at this and this is something that they recognized very early on in the field of drug discovery. This is a picture of a typical drug development pipeline. It takes about three and a half years to discover a drug leads. It takes about another five or six years to move the drugs through phase one to phase three and it takes about two years to get a drug approved. So on average it takes 10 to 15 years and around 800 million dollars to get a single drug approved. The first phase is focused usually on chemistry. Next phases require some elements of genomics of proteomics. But as I said what was recognized as early as the 1990s is that you could use metabolomics on every phase of drug development, discovery and approval. That said it's shown here is that you can use metabolomics to discover drugs. I'll give you some examples. You can and historically people use metabolomics and still do rapid toxicity screening and absorption distribution metabolism analysis. You can then use metabolomics to assess preclinical efficacy. So you have some markers are the markers for the disease changing when you give the drug. You can also use metabolomics for in vitro and later in vivo studies. Once you move it into humans again you can start using metabolomics for clinical safety biomarkers, clinical efficacy markers, toxicology markers, also compliance testing as well as consumption. So this is an example where you would look at someone with a metabolomic analysis at urine. In most cases they advise you not to drink alcohol when you are on a drug trial and so this is someone who day one, day two was just fine but day three they went on a binge, day four and day five they're just fine. So you can use metabolomics to identify people who are non-compliant. You can identify people who are fast metabolizers and slow metabolizers of the drug whether it's in blood or urine. And many drug trials and many drugs have essentially been terminated because they didn't know whether certain individuals were fast or slow metabolizers. They just gave everyone the same dose. In some cases those doses were toxic. So being able to do essentially pharmacometabolomics or to monitor people's metabolic rates via metabolomics is a very powerful way of assessing things. You can also use metabolomics and drug discovery. Now this is a traditional approach for drug discovery so typically most drug discovery exercises start with a GWAS study. So they look for genes and targeted genes that say okay this group of people has this gene modified and this is where we have literally spent hundreds of millions of dollars over the last decade doing GWAS studies to find genes that appear to be associated with anything from Alzheimer's to obesity to diabetes. Any given GWAS study costs about five to ten million dollars, takes two to four years and of the genes that they typically identify only about one in five are sufficiently important to make them into a drug target. Once those genes are identified some of them are not drug-able. They might be certain things that are you know vital tissues or important for development or something like that. So not every gene even after a GWAS success is usable as a gene target. Once the gene target is passes through then you actually have to clone it and produce an assay that can allow you to do high throughput screening and not every gene is cloneable and not everyone is susceptible for high throughput screening. Some of them are very large, don't fold, are made of multiple components so a screen just isn't possible. If you have gotten through all of those hurdles and I'm giving you the odds of success in these one to two ratios then you can do drug screening and this is where you use libraries of half a million or a million compounds hoping that one or some will actually be a hit. Not everyone, every library is a hit, not every assay works, so again you end up with some losses. So you now have your lead. If you have a lead and you think really good about it, you know this is checked through, this is wonderful, that lead only has a one in five hundred chance of actually becoming a drug and that's because most leads fail at levels of toxicology, cost, upscaling, admit a whole bunch of other things and this is a well-known number so one in five hundred and the cost as they say is about a billion dollars and then even if a drug is approved there's about a 50% chance that it will be successful. Some cases drugs are taken off the market because they just don't sell, they're too expensive, there's also a proportion of drugs that are removed because they turn out to be toxic that the testing done even by the FDA and others was not good enough. So when you add all these things up one in five, one in two, one in two, one in five, one in five hundred the odds of having anything from the point you start to discover or think you're going to find a drug target is point zero zero one percent and from the point you were looking for a gene to the point where you actually are selling a drug it actually is 20 years and well over a billion dollars which is the reason why the drug pipelines are drying up it's reason why there's consolidation in the drug industry and it's the reason why it's very painful to try and develop anything new. If you use metabolite-based drug discovery as opposed to gene-centered drug discovery the odds are much much higher. Metabolomic studies are cheaper a couple hundred thousand the data analysis as you guys will learn today can be done very quickly the interpretation of these metabolites in terms of their pathways the enzymes that are associated is very fast and simple in many cases because we know so much about metabolism we also know about all kinds of inhibitors or activators and many of those actually in fact are natural products because that's the inherent nature of metabolism and if they're not natural products there's often an easy way to develop enzymes or antibodies that can address those so there have been a few examples I won't go into some of them but where people have either repurposed an existing drug or have identified a novel drug using metabolomics with very short time frames one to two years with relatively little cost the other thing that's nice about metabolomics is that you can use it to monitor the success or efficacy of the drug that's been developed. I'll give you one example and this is from a study that was reported a few years ago in the Cleveland Clinic about cardiovascular disease and TMAO so I think most of us think of heart disease as it's mostly a diet related condition so if you eat lots of fatty foods showing cheese which we just had lobster steak poutine some of you may have had last night all of those things contribute to build up of fat deposits in the heart and in the arteries and lead to atherosclerosis but there are some people living in some parts of the world particularly in Italy and in Spain who eat cheese every day and eggs every day and whole milk every day and high fat meat and they have the lowest incidence of cardiovascular disease in the world that's called the Mediterranean diet so clearly it's not purely diet and people have been puzzled about this what is it so this is where the Cleveland Clinic started doing some simple metabolomic studies you know why are some people susceptible to heart disease with high fat diets and others essentially immune to heart disease with high fat diets and what they found was a compound called trimethylamine oxide as being a strong marker of individuals who had cardiovascular disease and atherosclerosis and what happens is that when people eat fatty foods and have phosphatidylcholine the phosphatidylcholine or fat is converted to choline. Choline then goes into the gut and it is transformed by bacteria from choline to trimethylamine. Trimethylamine then goes back into the circulation and it's transformed by the liver into trimethylamine oxide and trimethylamine oxide is an atherotoxin it attacks the foam cells or activates the foam cells which then lead to this build up of plaques or white blood cells and fat in the arteries and they've done the tests they inject TMAO into the arteries of rats and they develop atherosclerosis very quickly. So how do you stop it? Well one you can tell people to stop eating fatty foods but we don't want to. The next thing that they looked at is well maybe we can start hitting things that stop this flaming monoamine oxidase or FMO that's in the livers that does the conversion and they tried a few times but FMOs are so essential that it actually destroys the liver function and so that's not a good drug. So then they decided well maybe we could look at what's going on with the gut and what is it that's in the gut that leads to TMAO and they also noticed that not everyone produces TMAO so clearly there are some people with good bacteria that don't produce any TMAO and there are those unfortunate people who have the wrong gut bacteria that produce TMAO. So about a year after they published this someone solved the structure for this unusual enzyme called the choline TMA liase. So it takes choline and it converts it to trimethylamine and it's only produced by sulfate producing bacteria that have this enzyme. Not everyone has sulfate reducing bacteria in their gut or of this particular type. So they said cool we've got a we've got a now a target this is a protein that we can try and target and see if we can isolate the protein clone it and then start doing drug screening. So they started screening all kinds of compounds and one of the first compounds that they found that had a fantastic activity was this thing called 3-3-dimethylbutanol and if you compare the structure of 3-3-dimethylbutanol with choline it looks almost identical the only difference is this nitrogen is replaced with a carbon and this stops the TMA liase dead kills it and so no TMA is produced. So where is 3-3-dimethylbutanol found? Well it's in olive oil it's in grapes it's in red wine and if you know something about the mediterranean diet that is one of the essential features of the mediterranean diet. So people who consume lots of grapes red wine olive oil in their salads basically don't have very active um sulfate reducing bacteria and they therefore don't have very high levels of TMAO and therefore they don't have heart disease. So this is an example of essentially a metabolomic based drug discovery and the interesting thing is that the drug is in your food and the drug is something that you can take at lunch today. So just to summarize I think if we're going to try and see metabolomics move forward we have to look to improving some key areas automation is key expanding coverage is key making metabolomics cheaper and more portable is key but most important of all is the quantification and if we don't quantify then we can't move the discoveries that we make in the lab into applications practical applications whether it's in the clinic or whether it's in the field or whether it's in vet labs and those are I think the critical things that we have to do to make metabolomics mean something. So thanks very much.