 This is really sort of a light ending to the whole course. And it's partly just to start the discussion, but also to see maybe a slightly biased perspective of where I think metabolomics is going. And it might be useful for you for some of you that are just beginning and are interested in, I guess, maybe seeing a bigger picture of what metabolomics might be or where it should be going. So this is some data from a year or two ago, but just the number of publications that have been presented or appearing in PubMed in metabolomics. And basically, it's growing exponentially, which is good. The very first publication was in 1998 where they mentioned the word metabolome or metabolomics. And at the time it was metabolomics and back and forth. Anyways, the question has been is it's growing, but where is it going? And obviously people would like to see it go straight up through the roof. If you were doing the same plot for genomics, it actually is following the green line now in terms of publications. And if you did the same plot for proteomics, it's also following the green line. If you're doing the plot for structural biology, it would be following the red line. So there are changes, obviously, with different fields and different endeavors. And so this is where will metabolomics continue on that swing or level out. One of the issues of where it will go up horizontal or down has to deal with the bottlenecks in metabolomics. If we can't clear those bottlenecks, it will certainly be a passing fad. If we can clear the bottlenecks, then I think it will probably continue to grow. So a lot of people involved in metabolomics or those that have been thinking about it for a while are looking at these areas. One, trying to make metabolomics much more automated. And this is basically trying to copy what's already gone on with DNA sequencing. Next-gen sequences are basically making or made sequencing a commodity item. You can sequence anything, and they literally are sequencing anything. And this is a great way of characterizing stuff because it's so automated. You can buy the mini-IN or minion-oxford sequencer to stick it into your laptop and do sequencing at home. And it's that automated. So automated metabolomics is one trend. Another trend that's likely going to be continuing and it's also urgent is expanding metabolome coverage. We talked about the typical study generally aims for between 50 and 150, maybe 200 metabolites. A few people can get up to 1,000, which is really the exception rather than the rule. Other trends will be trying to make metabolomics portable. And again, look at what they've done with the DNA sequencer technologies, whether it's the MySeq, which is sort of a small refrigerator size to the mini-IN Oxford system. They are now pushing towards sort of toaster-sized mass specs. So making metabolomics portable will also make it much more accessible and cheaper. When we started off yesterday, I was giving you the whole list of equipment that a typical metabolomics lab has, which basically is a Christmas wish list for all analytical instruments. And that makes it hard to access. And many of you who are doing metabolomics largely work through core facilities rather than trying to run your own metabolomics lab. But what if I told you you could actually do metabolomics in your kitchen and for a few hundred dollars? So that's one thing that I think is going to be happening. Quantification, I've talked about that yesterday. I've talked about it today. I'll talk about it again. This is really, really important. And this has been a fundamental barrier that has prevented other omics technologies from being widely adopted into standard out-of-lab practice. So one of the examples of moving metabolomics out of the lab is the concept of moving it into the clinic or into the field or to the bedside or into the home. The original motivation for metabolomics, actually, if you go back far enough in the literature, was actually towards drug discovery and drug development. It was intended to be a field that would prevent or avoid having to do necropsies on rats and mice and to have to do studies on adverse drug reactions or toxicology. It was taught up until about 2002 and 2003 and then the pharma industry just completely abandoned metabolomics because it wasn't giving them the data that they wanted, largely because it wasn't very quantitative and the coverage wasn't very extensive. So I'm going to talk about each of these things a little bit and elaborate on them. Again, I'll invite people to maybe make some comments because I say, this isn't, I'm not teaching anything, I'm sort of discussing where things are going. So in terms of automated metabolomics, I've mentioned this already. There's the wine screener and juice screener that Brooker has developed. So if you have a million dollars, you can screen your juice and study your wine. But it is a really automated system and the equivalent of having a basil wrapped in with an automatic sample changer and an NMR instrument. I've also mentioned the biocrates system and this was the first kit, still one of a few kits available. It's a company in Austria. They sell them for about $4,000. And if you have the right instrument, you can analyze about 80 or 90 samples at a go and it's fully quantitative. About 186 compounds can be measured. And so the numbers work out to be roughly $50 a sample. So it's not very expensive and it's fully quantitative and very automatic or semi-automatic. We've talked about basil already, you guys have given it a chance. So this is an example of non-commercial efforts to produce automatic metabolomics. So this is an example of automated NMR. It's a lot of work as we found and it requires, I think, for automations, people follow specific standards. And so what we've determined based on what's happened with using or releasing basil is that about 50% of the people do it wrong. So we give them instructions, exact instructions, and they try and use it and they get it completely messed up. So we're doing essentially what Biocrities is doing. The best thing to do is put a kit together so they can't do it wrong. Unfortunately, of course, that means it does cost money. But at some point, eventually people will figure it out and then we won't have to do the kit anymore. The GC AutoFit, a little more elaborate because they're having to collect different samples. And again, you have to follow a process. But if you follow the process, then I think you guys have seen that it's pretty automated. And that if you've ever done GCMS and you've ever tried to do it manually, you would have taken, well, how long would it take you? Anyways, to do the quantitation and everything else, it takes quite a bit of time. So automation helps. And I think this is something that's going to be much more common. And these aren't going to be the only things around. They just happen to be web servers that make the class easier to teach. Okay, expanding the metabolism coverage. This is arguably the biggest bottleneck in metabolomics. It's the one that obsesses most people in the field. And this goes back to this issue of sensitivity. In the world of NMR, we're completely happy. We know all 220 compounds that can ever be found or measurable in mammals. And in NMR, all we have to do is just figure out what's up and what's down. That's because it's not very sensitive. From a publication perspective, NMR metabolomics is really easy because you're always characterizing known things. You're always getting quantitative results. The only limitation is that sometimes you don't find the really interesting compounds because the coverage is small. GCMS is a little better than NMR, but most of you are obviously in the LCMS world. And the reason why you're there is because of the sensitivity and because of the large number of compounds you can potentially detect. The problem with this, and we brought this up before, is that you're mostly looking at unknowns. You're only identifying or able to identify about 200 features out of the 20,000 that you're seeing. So that's a pretty poor rate. And so the question is, what are those other 18, 19,000 features? So the current thinking is that those features, those unknowns, are metabolites of metabolites. That is, they are transformed products coming from Phase 1, Phase 2, microbial or promiscuous enzyme reactions. So Phase 1 and Phase 2 are the enzyme processes that the liver largely has to deal with the xenobiotics. That can be the food coloring, food additives, drugs, food supplements. But they also work on endogenous metabolites, and they'll process them. Your microbiome does a lot of processing as well, and it transforms things in many exotic ways. And then your own enzymes will process things. They're not perfect, they are fallible, so they will take some substrates that look sort of like what they're supposed to work on, and they'll work on them, and they'll make some unusual or unexpected metabolites. So in the case of food metabolome, we eat lots of things, we're exposed to lots of things, and this just illustrates the fact that they are all transformed. And so we have a good idea about what's going into us. We've talked about some of those databases like Drug Bank and Food DB. We have very little idea of what's coming out of us, and the transformations. And the terminology that they're using here, which I'm sort of on the fence about, but they call the transformed food the food metabolome, the transformed drugs, the drug metabolome, the transformed pollutants, the pollutant metabolome. And then there's the endogenous metabolome, which are sets of metabolites. But it's the stuff at the bottom that is largely unknown. So how can we solve this problem of the unknown? So one of the ideas was to try and do systematic spectral collection. So if we took all of the compounds, synthesized all of them, then we could actually run them through every NMR instrument, every mass spectrometer, collect for everything, and deposit that into public databases. So that would essentially be the equivalent of the human genome project in terms of an effort and cost. I put in down 100,000. I think it's closer to 2 million compounds. And so if you said that the average cost for a person to synthesize a compound is between $1,000 and $2,000, which is very low, 2 million times 1,000. It's a $2 billion effort. And it would take probably 20, 30 years because it's not something you can systematize. It's people doing work. And then you'd have to collect the spectra for these. And you'd have to collect spectra for many different instruments under many different conditions. And again, that would be billions of dollars. That's not going to happen. I wish it could, but it just, it isn't. So this goes back to can we do systematic structure generation? Can we do systematic spectral prediction? And this is in fact a trend that's picking up tremendously. So there are efforts now to calculate transformation products. A tool called Biotransformer is being developed. And then we've highlighted some of the spectral prediction tools. So CFMID now predicts both ESI-MS or Kandem-MS spectra. And it also predicts GCMS or EIMS spectra. There are efforts to improve NMR spectra prediction. There are also tools that predict retention time, retention indices, collisional cross-section area from ion mobility spectra. All of those components and the capacity to predict them if we have the structure should make metabolite identification much easier and essentially cover those unknowns. Example, already we brought up my compound ID where all of the structures aren't generated. The masses are generated for literally millions of feasible compounds running through just some very simple 75 transformations. And identification went up from 5, 10% to almost 50%. So imagine having a little smarter tool that generates the structures where you actually have not just the masses, but also fragmentation pattern predicts also the retention time and retention indices. I think it potentially could lead to a much better, more comprehensive coverage. So that's, I think, another trend that's emerging. Certainly more effort already going into that. Several workshops and conferences have already been called on that. Next one is making metabolomics portable. So I think some people may even have a Fitbit. Does anyone have a Fitbit? Okay. There are and has been a program that was sponsored by, what do I just want to say, is it, I want to say Qualcomm, but it's not. It's called the Tricorder XPRIZE. And I think if you type in Tricorder XPRIZE on Google, they've had their first round of entries and they've awarded people sort of seed funding. But the idea is to be able to do precise monitoring of body electrolytes, physiology, and to do it on a smartphone or something smaller. Probably the best known medical monitoring device is the blood glucose monitor. And they are now portable, they can plug into your laptop. And they're essentially a portable metabolomic device. They measure one compound, but they measure it very well and very accurately. So in the case of metabolomics, trying to make things a little simpler and more portable is a way of democratizing it. As I said, there's only a few places around Canada and North America that actually are able to do full-scale metabolomics, buying some of these instruments that's millions and millions of dollars. Getting trained to use them is decades of work. If you could have one that's handheld, any of you could use it, any of you could probably run it. And this is actually a real instrument. This is, I think it's a little more than $1,000. But it does a number of clinical assays, chemical tests. I mentioned the Qualcomm Tri-Quarter X Prize. That's the one that led to these Zensor things. And in fact, some of these things are already being made. There's a lot of work also going on in microfluidics, nanofluidics, and nanotech. So Kepler-uptropheresis has now been ported to CHIPS. HPLC can now be on CHIPS. You can actually put GC, gas chromatography on CHIPS. There's also efforts to put mass spec systems on CHIPS or cantilevers that can measure down to a few Dalton's. So it's happening in the sense that you could probably put a lot of these things on something the size of this mouse or the size of your smartphone and power them with not much more than a lithium-polymer battery. In terms of measuring volatile compounds, it's already here. So there's electronic noses that are able to identify a number of volatile compounds and the systems are not much bigger than a smartphone or maybe a laptop at the largest. They look for patterns and they are trained on patterns. So they use things like machine learning so they can identify the intensity of things. They function a little bit like our noses and they're receptors and using the equivalent of sort of like not necessarily proteins but at least components that have avidity for certain combinations of molecules. But the idea of using receptors is something that we're actually trying at the U of A and this is to try and make metabolite sensors. And this is essentially moving sort of like the way that the glucose oxidase is an enzyme that's used for glucose sensing. The idea here is to use either antibodies that are specific to small molecules or paraplasmic binding proteins which are naturally occurring metabolite binding proteins that are found in bacteria or aptomers, RNA and DNA aptomers, which are also naturally occurring compounds and can be engineered to bind metabolites. So if you bind the metabolite, these things will clamp on fine. How do you get a signal? So the trick is actually to modify the metabolites so that they have a gold nanoparticle stuck onto them. And you can set up things that are either competitive assays or just simply binding and release assays. And when the compound is bound the gold nanoparticle or is bound a regular metabolite you'll see differential changes. So you can use things like surface plasmid resonance. So you can get little SPR instruments that are about the size of a laptop. You can use enhanced Raman spectroscopy or surface enhanced Raman spectroscopy that serves. Or you can measure equivalent of resistance or impedance, which is the AC equivalent of resistance. And so that's an electronic measure. The electronic one is actually cool because they've already progressed to making a handheld device, converting an impedance meter, and it's about gaining the size of a smartphone. And so it's using these gold nanoparticles and these aptomers. And it's not measuring hundreds, but potentially should be able to measure four or five metabolites by the end of the summer. So this is a concept which can be generalizable because there are literally dozens of known small molecule antibodies, dozens of perplasmid binding proteins, dozens of aptomers. And making these gold nanoparticle decorated metabolites is actually quite easy now. So hopefully perhaps in not too distant future it should be possible for everyone in this room, including your friends and neighbors, to actually do metabolomics. The other part I think that's trending is quantification. And this is actually made the cover of a journal called Trends in Biotechnology. So here's a little picture of a ghost, but it has this phrase here, a specter is haunting metabolomics, a specter being ghost, a specter of quantification. And it was really sort of a call to arms saying that it's important to quantify. At the time when this was published, 90% of metabolomic studies were either semi-quantitative or not quantitative at all. Less than 10% actually had absolute quantification. Trend is improving, maybe up to 20% now, but it's still not the level that really needs to be there. So the field really has to be quantitative if you're going to do any type of translation. And these days, given that most of us are taxpayer-funded scientists, the reason why we're getting the tax dollars is because people are hoping that we're going to translate what we find into something practical. And I think this has been a problem in terms of moving things from the lab to the practice. So quantitation, it's already happening. We've highlighted it with the biocrities, we've highlighted it with Rooker, there's Kinomics, a company in Edmonton that develops quantitative, automated, semi-automated NMR. There's quantitative efforts you guys have already seen with Bazel, Batman, and GC AutoFit. And if you look at actually what's happened in terms of quantitation, actually Metabolomics is doing pretty well. So in terms of a number of metabolites ever quantified, fully quantified in serum plasma, Metabolomics record is 288. Proteomics, the number is now up, it's actually about 160 that were fully quantified. However, we looked at the data and I would say it's only semi-quantified. So that's sort of what the qualification, and it's certainly the data would not be transferable to other labs. Cerebral spinal fluid, 172 metabolites identified and quantified. Proteomics is 130 proteins identified and quantified. Urine, 378 metabolites identified and quantified. Proteomics, 63. A real challenge in at least with RNA-seq and micro-rays is that you cannot absolutely quantify. You can get a relative quantification, but you can't move what you measure on an AFI chip to an RNA-seq platform. And between different RNA-seq platforms, you can't necessarily move them. And between different labs, often you can't move them. And this is why there's only been a tiny number of tramskip-based assays that have actually moved out of the lab into general use. On the other hand, there have been dozens of metabolite assays that have moved into clinical use or general use because of the absolute quantitation. So, if you can get through that quantitation threshold and focus on that and make that part of your regular routine, then you can start thinking about moving metabolomics into some of the applied space. Just on that slide, if I can direct, how reproducible is the quantum economics versus the proteomics? Does that make sense? Like, I mean, from lab to lab? So, typically, again, when you're doing absolute quantitation, the target that many people want to have is an assay that has a coefficient of variance or CV of about 10 to 15%. So that means that within your lab, within the instruments, with different people using the instrument, different people using the protocol, you should get a result that's within 10 to 15%. Many people can and do achieve 5% or less. It's a function of abundance, but someone's not careful, someone doesn't follow the protocol, then you'll mess up and that happens. If you talk to people in the world of protein measurements that do elises, they're happy if they get within 100%. There's very little consistency between labs and there's real problems with antibodies and things like that. So, it was kind of shocking for me to find out just how variable and inconsistent a lot of these things are. But in the realm of analytical chemistry, we're building on 100 years of people really focused on you have to measure this to precision of 1% or something. And that's something that you really should strive for and it's certainly possible to do that. Not for every compound, maybe not for every assay, but if you do your best. So, the well-controlled biocrity system, they have a QC system, we see that 5%, 10% CV all the time. So, in terms of moving ideas from the lab to the clinic, scientists don't do so well. One example would be biomarkers. So, if you type the word biomarker in PubMed, you'll get about 700,000 hits. So, lots of people looking for biomarkers and only 200 or less than 250 have been approved for clinical use in the last 45 years. Proteomics has not yet produced a biomarker that's made it into clinic. And proteomics has been around since, you know, the 1980s. There's some good proteins that have been identified by proteomics, but they immediately are converted to ELISA assays. In part because mass specs are really expensive and quantification is really tough. In the case of transcriptomics, again, because they don't perform absolute quantitation, there's only been five marker tests that have been approved for use and they're not used very much. They exist, but I don't think they're really making money. On the other hand, just about everyone in this room has actually already had a metabolomic test. It's called newborn screening. And when you're born, they take a blood spot and put that onto a card and then send that into a mass spec. So it's done for every newborn in North America, every newborn in Europe, and many newborns in Asia now. So this is an example of an extremely successful omics test, and most of you've never heard of it or no one talks about it. But this is, I think, one of the ringing successes of mass spectrometry, analytical chemistry, and metabolomics. They are measuring dozens of compounds on these infants. So if you look at the numbers, the number of approved tests that come out of metabolomics slash clinical chemistry is just about 200. In terms of genomics, the number of approved tests is somewhere over 100. Now, it depends how you count it. You can count every single gene variation that they can detect. And so for something like cystic fibrosis, where there's about 600 or 700 mutations, they can multiply that and say, oh, we're doing 600 and 700 tests, but it's basically for a single gene. So if you look at the genes that they analyze, it's about 100, 110. There's about 60 ELISA tests, things like C-reactive protein and some of the interferon interleukin tests. As I said, the transcriptomics is five, and proteomics still is sitting at zero. So even though metabolomics arguably is one of the youngest fields of omics, it's actually been remarkably successful for making that translation. One reason is because there has been a focus on quantification and reproducibility. But it's also because metabolites are the canaries of the genome. So if you look at metabolites in terms of things like disease prediction and diagnosis, and these are just examples for humans, but you can use this for plants. You can use this for animals. You get some pretty remarkable results. You guys have already seen the result we got for preeclampsia when we looked at the rock curve for predicting preeclampsia. So 90-some percent. So there's early preeclampsia, which is really bad, and late preeclampsia, which is more an inconvenient. But both of them seem to be predictable, just using metabolites from the blood. Looking at, again, pregnant mothers early stage, you can identify whether the fetus has a congenital heart defect. So you're not looking at blood in the fetus. You're just like taking mother's blood early stage. But it shows up. Other genetic defects, trisomy-18, trisomy-21, Down syndrome, you can also detect that in the mother's blood in early stage. The idea here was you wouldn't have to do amniocentesis, or at least you could screen, so you wouldn't have to do it for everyone. And then you guys have already seen the catexia work, and this too was something that showed that it was possible to predict cancer catexia just looking at a spot. You're in sample. It's not perfect, but this is actually better than anything else. Which is just indicating that in fact metabolites are these sentinels. They tell you what's going on because typically they're the first responders. They're the firemen of the defense set. So you can predict diseases. You can also diagnose diseases. And there are some diseases that are really hard to diagnose. So in the case of kidney transplants, you know of anyone who's ever had a kidney transplant, typically they have to be monitored quite closely. And the way they monitor is they come in, and they stick a giant needle in your back to take a kidney biopsy. So it's not pleasant. That biopsy is then analyzed by a pathologist, and they say, well, it looks like there's rejection or there isn't. And then they do this sort of on a fairly regular basis. Wouldn't it be nice if you just have a urine test to say, there's a problem? So the idea here was to simply see if you could look at urine, which is what kidneys produce. And lo and behold, yes, you can in fact identify individuals that are rejecting or about to reject their kidneys with a very high percentage performance. You can look at people with heart failure. In this case, we're looking at blood. And distinguish between whether it's systolic or diastolic heart failure. And there's very different treatments. So again, to distinguish between systolic and diastolic heart failure is sort of a two-week battery of tests. This ideally would just be a blood test. You can get an answer very quickly. Some of you may know people who have chronic fatigue syndrome. It's one that's obvious controversial. Is it a disease? Is it not a disease? But looking at individuals who have apparently chronic fatigue syndromes that are very distinct metabolic fingerprints in those who are generally healthy. So it suggests, yes, it's a disease. And yes, it is something that can be diagnosed by looking at something a little more objective than saying, are you feeling tired and worn out? Which I think all of you guys are feeling right now. There are other conditions, other disorders. Eosinophilic esophagitis. This is a disease that will happen with kids, actually, where they have to chew food for about an hour before they can swallow it because they have a very constricted throat. And it's a swelling. It seems to be somewhat related to Crohn's-like conditions. How it happens, no one knows. Why it happens, no one knows. But it's hard actually to diagnose because sometimes they think it's just because the kid is wanting to make trouble. And so it takes sometimes weeks or even months. But ideally if you have a blood test, rather than having to often do tissue biopsies, which they do when the tissue biopsies stick something, a needle down your throat and take it out, that would be obviously unpleasant for a child. So using urine or blood is a lot easier. The most interesting and arguably the most successful one that we've been involved with is this detecting colon cancer. So colon cancer progresses, sort of what's shown here, from polyps. And so typically if you get older, they will have you doing colonoscopies. Typically people over 50 are advised to do that. If there's a family history, you may even start that at age 30 or 40. Colonoscopies are expensive and not actually pleasant. There are other tests. There's one called the fecal occult blood test, which basically looks for blood and poop. There's a less than 5% compliance rate when a doctor sends you home with, here's a vial and poop in it and send it back to me. So most people don't do that. And even if they do, the test is only about 20% efficient or correct. So it's better than, worse than tossing darts. So if you could do a urine test, which is a little easier, could you detect polyps? And that was the question we asked and worked with colleagues in the medical school with Richard Fedoric. And the answer is yes. You can use urine and you can detect polyps. So this is before the cancer. And if you detect polyps, then you can start sending people to say, well, let's do a follow-up test. And there's some more precise tests, obviously colonoscopy, but you don't want to do colonoscopy on every single person. There aren't enough gastroenterologists to do it, and it's very costly. So this test was first confirmed using NMR, and then we modified it so it could be done with mass spec in a high throughput fashion. And as of last week, I guess, it was now moved to a couple of testing clinics in the East Coast, in the U.S., and it's going to be offered in many other places. So this is an example of translating an idea from metabolomics into predictive precision medicine, because there are things you can do about early-stage polyps. But if it's a late-stage colon cancer, often it's too late. So the last part I'll talk about is just the idea of moving metabolomics back into drug development and discovery. So this is going back to its origins, which was the intent of metabolomics. It's not the only one. I've given you some examples of, I think, where it could be applied. And most of the things I've given you are examples of human medicine, which I think everyone is human here, so it's probably relevant to you. But it does apply also to animals, and it applies to plant analysis. And many of those things, same sorts of things, same idea of quantification, portable field devices, moving it into environmental testing as well, getting it out from the lab so it actually is used by the public or by physicians will make a difference. But the one that I think still attracts a lot of attention, and generally it's the motivation that everyone says, the reason why I'm doing genomics is for drug discovery. The reason why I'm doing it is for drug discovery. Well, also you can say the reason I'm doing metabolomics is for drug discovery too. So this is a picture of the drug development pipeline. Average cost for a drug is, well, I actually approached almost one and a half billion dollars now, but some of them can get away for only 800 million dollars to get a drug. It takes on the order of 15 years to get a drug from the point of discovery to where it's used. There's a tremendous attrition rate from drugs that make it just through discovery phase. There's about a one in 5,000 chance that they'll actually get FDA approved. So most people who work in drug discovery will for their entire lives never have a drug that's associated with their career. In fact, the most successful person who actually got several drugs through the pipeline ended up winning a Nobel Prize because she was so unusually successful. Now, in terms of how drug development happens, it's a lot of chemistry in the first phase, but there's a lot of omics that's actually used still to help with both discovery phase one and phase two. The interesting thing is, both in terms of the vision and the reality, is that metabolomics is and can be used through all phases of drug discovery, phase one, phase two, phase three, and FDA approval. People discover drugs and drug targets through metabolomics. That's happened. There's some really excellent examples already. There's testing phases, which are both preclinical and clinical, where they're looking at drug toxicity, both in rats and also in humans. Tabalomics is ideally suited for that. It looks for physiological changes very quickly. Phase two, efficacy. Phase three, large-scale efficacy. Again, in many cases they want to target individuals. It's getting to the point where a third of all the drugs that are being produced now are being targeted to specific individuals or certain classes of patients. And so one of the best ways of characterizing or making sure that people are getting the right drug for the right condition is again through metabolomics. And then FDA approval and after, which they call sometimes phase four, is monitoring effects. Are people getting adverse effects for drugs? And in fact, game metabolomics is frequently used. So this is just illustrating how metabolomics is used. Discovery, talk screen, efficacy, prioritization, safety, biomarkers, clinical safety biomarkers, clinical efficacy biomarkers. So here's an example where you might be able to look at someone who is being monitored and the instruction is do not drink alcohol with this medication. So you're tracking and tracking and now you can see this. But this is something that you can't do outside of metabolomics. Typically it's still done as a questionnaire. Have you been drinking alcohol at all? Did you follow instructions? And I think most cases they're finding people don't. And some of the adverse reactions that they see can kill an entire trial, which can mean a billion dollars lost for a company. Likewise you can look at people metabolizing drugs. So there are fast metabolizers and there are slow metabolizers. An example of someone might be a slow metabolizer. How many people who they drink coffee or tea just before bed cannot sleep? So you are probably a slow metabolizer. I drink coffee or tea just before going to bed and I sleep very soundly. So I'm a fast metabolizer. So again, these are things that are important for understanding people's reactions to certain drugs. And this will indicate whether the cytochrome P450s you have are going to change them. And so looking at proxy drugs actually is a really useful way of understanding a person's drug phenotype. The traditional way of doing drug discovery has evolved over the last 30 years. But this is the general pattern. For the last maybe 10 years people have been doing GWAS studies, and for that they do large scale genetic studies. And they would look for people who had certain genes which indicated that they may have a proclivity for a certain disease. And that was the point of GWAS. Only about one in five GWAS studies actually identify SNPs or mutations that are sufficiently useful or robust to suggest that they should go on to drug efforts. Once they found those mutants or SNPs only about half of them proved to be useful. Some of them are essentially SNPs that are found only in non-coding regions or targets that are too obscured to understand. From there you might try and clone the gene, produce it, and get enough material so you can start doing drug screening. But not every gene is cloneable, especially membrane proteins. They are very hard. So that often prevents the screening system to develop. So whether it's cloning or just developing an appropriate cell system, it's hard. So one in five, one in two, one in two, and then you can start doing the large high throughput screen. And that's where they use libraries of a million compounds called the ones you find in PubChem to see if anything binds and inhibits. And a lot of these large scale screening trials don't work again about 20%. That's the end of the discovery phase in the beginning of the clinical or pre-clinical phase. And that pre-clinical phase takes billion dollars, 15 years. And success rate is about one in 500. Of the drugs that are approved, anywhere from 30 to 50% actually have to be taken off the market because of subsequent adverse drug reactions or issues with respect to, well, who knows what, marketing approval whole range. So if you multiply all these probabilities from the point where you begin with a GWAS study, the point where you actually have a drug in hand that's FDA approved, it's a 0.001%. And it takes at least 20 years and would cost, well, more than a billion dollars. What about metabolite-based drug discovery? This is something that not too many people think of, not too many people have tried, but in this case it's trying to look for people, cohorts, so you'll do a metabolome-wide association study, MWAS, and like to see what comes out. So we showed you examples of people with preeclampsia and not preeclampsia, colon cancer versus no colon cancer. Those studies are all done not with tens of thousands of patients that you have to do with GWAS. They were done with about 60 to 100 people. Total cost was about $200,000 or less. Less to the analysis, as you guys learned today, can be done in a couple of hours, maybe a couple of days. We've done the pathway analysis, and we can figure out some of the perturbations and the biological basis for some of these perturbations, also fairly quickly. And success for these are pretty much guaranteed. In many cases you can actually find things that suggest simple corrections. In some cases it could be things like supplements. In some cases there are clearly drugs that are known to alter or affect things. And in these cases it is possible to come up with some very actionable or repurposed drugs or advice. It also is possible to try and come up with other things like enzymes and antibodies, and in fact that's how some inborn areas of metabolism are now treated. And then once you've actually maybe found something that works, you can go back to the metabolomics route and say, did I make an effect? Is there a positive change? And that's not very expensive. So what's an example of this sort of thing? This is Stan Hazen's work. How many people have heard of Stan Lee Hazen? One or two? Anyways, he did something that was really quite remarkable. So everyone is told that, you know, avoid cholesterol. You'll get a heart attack if you take it. And cholesterol is bad. I think whether it's LDL or HDL and everything else, I think most people have been indoctrinated about how bad cholesterol is. What they were doing was looking at people who devout atherosclerosis, which is one of the consequences of cholesterol. And other cardiovascular disease symptoms from myocardial infarction. And they found a compound that was very high in the blood of people with atherosclerosis and propensity for heart disease. And that was trimethylamine oxide. And I said, well, that's cool. You know, here's a biomarker. But let's go back a little further. Why is it high in these people? And then they started doing some, I think, some really original, useful biochemistry. And what they determined was, in fact, that the TMAO was actually coming from fatty foods. Okay, we all know that eating fatty foods isn't good, but it wasn't the cholesterol actually that was coming from the phosphatidylcholine stuff that you've got to get in the eggs or butter or milk. But that wasn't the only story, because in fact it's been well known that there are people who can eat all the butter and milk and eggs they want. In fact, the French do this all the time. They have the highest consumption of fatty foods in the world in some respects, and they have some of the lowest incidence of atherosclerosis. What's the trick? Well, this is where Hayeson discovered, in fact, that it has a lot to do with the microflora that live in your gut. If you have the wrong bacteria, then the fatty foods will produce this atherotoxin called TMAO. If you have the right bacteria, then you can eat all the fatty food you want. And this is the process. So, phosphatidylcholine is broken down into choline, gets into the system. Microbes will convert the choline into trimethylamine. I think you guys saw when we were looking at the rumen of cows. We saw dimethylamine, and if you look further down maybe trimethylamine. So this is a product that is a biogenic amine. It's something normally you don't want to see, but when it passes into the liver, it's converted to trimethylamine oxide. And normally TMAO is secreted in the body. In fact, you'll find it in fish. It's a salty seawater fish from cold oceans. Typically you can't have TMAO. So someone eats a lot of fish. They will find TMAO in the urine, but it doesn't really end up long in the blood. But when it persists in the blood and is produced by the liver, it seems to stay in the blood. It seems to activate the foam cells, which leads to the development of atherosclerosis. So the question is what should you use as a drug target for these things? Should you try and prevent people from eating fatty foods? That's one thing, but lots of people like lobster and cheese and eggs. Can you try and target something in the liver? And they try. They looked at some possible drug leads, but that actually caused damage to the liver. Could you target the bacteria? And in fact, they found something that actually works incredibly well. It's a compound that resembles I think it's dimethyl butanol, I think is the thing. Trimethyl butanol. Which looks a lot like choline. So this particular compound is found in high abundance of extra virgin olive oil. And that seems to explain this unusual phenomenon called the Mediterranean diet. How many people have heard of that? A few of you? So the whole idea is eat all those, drink lots of or take in lots of olive oil on your salad and then eat as much butter and cheese and lobster as you want. And that's the essence of the Mediterranean diet. And evidently this high consumption of olive oil is essentially removing or killing off a bacteria that would normally cause this production of trimethylamine. So there you have it. If you want to cure heart disease and reduce atherosclerosis, just keep on pouring on that solid dressing. But this I think again is a wonderful example of metabolomics going from a biomarker study, which says TMA is bad or it seems to be a marker, tracing back through the biochemistry and then highlighting something that no one really expected and which seems to explain a bunch of things in terms of cardiovascular disease and suggests some interesting and very inexpensive and clearly rational therapies for treating a very widespread and important disease. So to wrap up I think metabolomics is trending and I hope it will be trending upwards. And I think you guys have had a chance to look at some of the examples of how automation is helping, how quantification is helping. Maybe next year we'll be able to show you some portable metabolomics devices you can try and sample each other with. I think there's also clearly a trend towards expanded coverage. I think in order to make metabolomics relevant we have to move it away from the lab and into the hands of do you want the public to users. So we can say putting in a clinic, putting a lab, putting in a field putting it into the house democratizing it. I think that's an important drive. That's something that sort of already happened with genomics partly by accident but I think if we think about it rationally this is something we need to do with metabolomics. And I think there's some really exciting examples of where it's already happening where metabolomics is helping with rediscovering or improving our understanding about diseases that we thought we all understood, heart disease and opening up new vistas for both the treatment and the rationalization for how some things seem to work. So with that I guess I'll wrap up.