 So today's webinar is titled, Linking Metabolomics to Diseases Using Human Genetics, a CLSA study. This webinar will be presented by Yuen Heng Chen, PhD candidate in the Department of Human Genetics at McMagel University, who is under the supervision of professors, Brett Richards and Celia Greenwood. His research focuses on understanding the interplay between metabolites and diseases using genomics approaches. He is also interested in the impact of sex on the genetic variant metabolite associations. Okay, so now I'll turn it over to you. And yeah, and thanks everyone for your presentation of my webinar. And thanks for Celia Chen for giving me this opportunity to present my work on the metabolomics and genetics. So today's talk will be centered around this study entitled Genomic Atlas of the Plastametablon prioritizes metabolites implicated in human diseases. So specifically I will talk about the general background relate to metabolites and diseases and briefly describe the method that were used in this study and the dataset that were used. And then I will talk about the details for this research study, including the research aims, designs, results and conclusion. Let's start with the background formation. So let's start with the main focus of the study, metabolites. So what are metabolites? So metabolites are small molecules that either the intermediate or the end product of in the metabolic reactions. The definition shows, depending on their chemicals features or about function, they can be usually grouped into categories such as lipids, amino acid, carbohydrates, vitamins, energy substrates, nucleotides, xenobiotics and more. They are important because they are involved in the essential biological process in the organisms and contribute to the development of numerous phenotypes. Some well-known metabolites include like glucose and fructose from candy, like caffeine from coffee and vitamin C from the fruits. So manufacturers can influence metabolites. Genetics is one of the major endogenous contributors. So genetic variants, they can influence the transformer levels and also protein levels. And those proteins, especially those enzyme transporters, directly metabolites and metabolites play a key role that regulate the levels of metabolites, either in the circulating system or in specific tissues. Exogenous factors like food, drugs or environmental factors like even side exposure, they can also influence metabolite levels. Lastly, the gut bacteria have been found to, they can further metabolize the food or those rich the gut and the produce metabolites, which can be re-observed by human body and influence metabolite levels. So then the next question is how a metabolite measured? So one technique that widely used is called high-performance liquid chromatography mass spectrometry, HPLC, LC, MS. This is a technique that can measure hundreds of thousands of metabolites at the same time and is also the one chosen by CELSA to measure the metabolites. So in brief, so the blood samples or plasma samples first go through this HPLC step for chemical separation and then the separated chemicals will be ionized and detected by the MS part. So then the mass spectrometry or the spectrum generated by mass bag can be used to identify or and quantify those metabolites, such as cholesterol and glycine. So as I mentioned, this technique can measure over 1,000 metabolite simultaneously. So to the next question is, why do we need to study metabolites? Why do we need to spend so much effort to collect samples and develop those two O's to measure metabolite levels? I already partially mentioned that one of the reasons is that those metabolites, they are essential for the metabolic process in the human body and they also contribute to the development of the phenotypes. Additionally, studies have found the association between those metabolites and many diseases, such as branched-chain amino acids with obesity, saturated fatty acids with liver disease, some bio-acid with COVID-19 infection in lung, some bio-acid were explored as the treatment. And lastly, cholesterol with coronary heart disease. So those associations has been found in the animal models and also in cohort studies from human. But this still, those association fund is still many steps away from the real clinical practice and it's partially because of two major challenges. One is called reverse causation and the second one is called confounding and it will explain those two concept in detail here. So in, for example, if we find a metabolite that are associated with increased risk for obesity, there is also a possibility that this association observed that those higher levels of the metabolites in obese individuals is could it because of obesity itself led to the changes of metabolites instead of this metabolite increase occurred before the obesity because obesity as a physiological change, it come with a lot of, for example, more adipose tissue, more adiposeides and those changes can lead to in alteration in metabolism too, which can lead to changes in metabolite levels. So that's the challenge for reverse causation. Another challenge confounding is in this example, we find metabolites to be associated with cardiovascular disease. We also know that obesity is also important factor that could influence both cardiovascular disease and metabolites, so it is a confounder. So if that's the case, then the association we observed between the metabolite and cardiovascular disease is just occurred just because obesity, not because this metabolites have any direct effect on cardiovascular disease. So that's the situation or challenge called confounding. So then how can we address those challenges that will lead to the message I'm going to talk about. So one or the golden standard to solve or address those challenges is randomized study. So a general design for randomized control trial include when eligible participant got recruited, they will be randomized randomly assigned to arms, intervention arm and the placebo arm or the control arm. So here I'm using vitamin D and multiple sclerosis as an example. So if we want to study whether given people, vitamin D supplements can reduce their risk for multiple sclerosis, we can design these randomized control trial and give the intervention arm, the people in the intervention arm the supplement of vitamin D and which can lead to increased cellular level of vitamin D. And in the control group, we just give them placebo and their vitamin D level will remain the same. So then we after follow up them for many years, we will see whether the incidence of multiple sclerosis is higher or the same between those two arms. The advantage of randomized control trial in brief, of course there's a lot more advantage but specifically for the two challenges I brought up is they can reduce the bias related to the confounding by the randomization step. So if we think obesity could be the confounder by randomization, the intervention group and the control group will have equal number or like highly celebrated number of BMI or obese individuals to make obesity less or unlikely to influence the different incidents we may observe for the multiple sclerosis. Another advantage for randomized control trial is that we are sure that intervention like the supplementation of vitamin D happens before the incidence of multiple sclerosis. So the timeline is very clear which avoid various causation issue. Of course randomized control trial have some limitations such as having costly, it usually require long duration to see enough or to get enough cases for the comparison and also sometimes it's unethical to test the potentially harm or harmful intervention. For example, if you want to know the causal effect of saturated fatty acids, we cannot give people saturated fatty acid directly if we know it's really a harmful thing. So yeah, so then some scientists come up with this new genetic epidemiological approach. What I say new is has been there for at least 10 years but it's become quite popular for the past five years because of availability of the data. So this method is called Mendelian Remediation. So I would call it MR for the rest of my talking. So basically MR is a genetic epidemiology approach where we try to mimic the design of a randomized control trial. So similarly in the MR study, we have the population. We assume that the Leo's oxygen variant were randomly assigned. So that led to two groups. The group for them who have the GLEO which we know it's associated with higher level of vitamin D and another group individual with CLEO where their vitamin D level is lower or which similar to no change. And then by checking the incidence of the disease within those two groups that have different alleles of this gene variants, more specifically we are checking whether the GLEO is also associated with the risk for multiple sclerosis. We can infer the effect of our exposure vitamin D level and the outcome multiple sclerosis risk. So actually this study, the MR study has been done to investigate the vitamin D effect on risk of multiple sclerosis. So this study has been published on plus Madison. So in this forest plot, I'm showing you the effect of single genetic variant. There's four of their effect on them through the exposure, which is vitamin D on the risk for multiple sclerosis as well as also there is met analysis results. So here it shows the per standard deviation decrease of the vitamin D level. We observe the higher risk for multiple sclerosis. So which kind of support the conclusion that the vitamin D is causal for multiple sclerosis risk especially for those individuals with lowers and average vitamin D levels, they have higher risk. So there are three key assumptions for MR study. So the first assumption is that the genetic variant, they are associated with our exposure. In my case, here is metabolites. And the second assumption is that the genetic variants are not associated with the confounder of exposure and outcome. And lastly, the genetic variants are not only associated with the outcome through the exposure. With those three assumptions, we can use the genetic associations between the variant and exposures as instrument variable to infer the causal effect of exposure on outcomes such as metabolites to disease risk. So just to summarize this method. So MR kind of uses measured variation in genes of known phenotypes to examine the causal effect of a modifiable exposure such as vitamin D on disease such as multiple sclerosis. So the advantages of MR, especially to address those two challenges are so they can avoid confounding by simulating the randomization process. So according to the Mendo second law, the independent storm and so alleles will be sorted into gametes independently and randomly. So which means the individual who carry the GLEO as I mentioned that increase or led to higher level of AMD, they should be similar in terms of their potential confounders or other features as those carry C alleles. So the confounders should be, you know, evenly distributed in those two subcommittee. So another advantage is that it can avoid reverse correlation is partially because genetic function, they usually happened before the occurrence of the disease. So the generic variants, since we know they're related to metabolites, they are also near the gene that encode an enzyme or a transporter for vitamin D. They are likely to first influence the vitamin D levels than influencing the disease outcomes. So one missing part I didn't mention when I describe MR is how can we find the association between the specific Leo of the variant with either the vitamin D levels and with the risk for multiple sclerosis. So to find those associations, we perform a type of analysis called genome-wide association studies. So for example, for vitamin D, we had good for people, we recruit them, we genotype their genotypes and we also measured the vitamin D levels. Then we can check the association between all the vaginal variants and the specific Leos of those variants and whether the dosage of specific Leo and the social ways may be higher or lower level of vitamin D. And here is just one example of a positive association between the GLEO and vitamin D levels. As for the case control or for diseases, we really have a cohort with both cases and controls. And then we check whether specific Leos and the variants are more frequent in the case group than the control group. Then that observation can lead to the identification of the association between this Leo and this variant and the disease liability. So using GWAS, we can find the association between the genome variants with metabolites and with disease risks. So that's conclude the two methods I used in my study. So lastly, for this introduction, I will talk about the data I used. So I think everyone here knows what is CSA. So basically CSA recued over 50,000 individuals to collect their information for research use. And specifically they have this comprehensive cohort which have about 30,000 individuals. They get more data from those individuals. They also collect the blood samples. They also the urine samples and measured metabolites and genotype data using those biological samples. So one important consideration when we try to use this comprehensive cohort is that this cohort is more educated. They have generally higher household income. They're more Canadian born and really have better general health compared to the general population of Canada. So I've been more detail about this comprehensive cohort. So they're clinical phenotypes such as, you know, body mass intakes, disease risk, or sorry, disease prevalence has been collected. They're socioeconomic status, data are also collected, lifestyle behavior data were also collected. And importantly, they have their genotype data collected and also imputed for use from metabolites about 9,500 individuals a month from this comprehensive cohort, you know, have metabolomics data. And by the HPOC MS technique over 30,000 individuals 1,500 metabolites were measured by this company called Metabolone. A few more words about the genomic data. So being specifically for this comprehensive cohort about 26,000 individuals have genomic data, 50% females, 93% were identified as European SS3 individuals. Genotyping step, they measured about 800,000 variants. And then those genotyped variants were imputed using the top meta reference panel to which can lead to identification of over 300 million genetic variants that can be used for genetic research. From a tablet or from a metabolomics data, so the EDTA plasma samples were sent to the company for them to measure the metabolables. So as I mentioned, the metabolome, they used this HPOC MS technique. So they measured 13, 14 biochemicals include 1,071 compounds with no identity. So they can, based on the mass spectrum, they know what chemical those compounds are. And they also have about 243 compounds that they can separate, but they just cannot map them to any known structures. So those metabolites are unnamed biochemicals and usually characterized as like partially characterized or unknown metabolites in the data provided by metabolome. So for the metabolomics data, metabolome provided the data after different normalization and imputation steps. So the data after batch normalization or QC matrix normalization has been provided. So both approach aims to address the potential batch effect with the measurement metabolables. So then data without or with imputed values, with minimum values detected for the given metabolites were both provided because some metabolites that could have relatively low abundance in the plasma. So it could be below the detection limit. And then the data, imputed values based on the minimum values for the metabolites were provided for those metabolites. So finally, we reach to the part I would talk about the research aims, design and results and collusion for my research project. So for this project, I have two aims. The first aim is I try to identify the genetic determinants for circulating metabolites using GWAS. And second aim is using the association I identified between varying the metabolites, I then applied MR to infer the causal effect of those metabolites and 12 treats and diseases. So let's start with the M1, the GWAS part. So for the GWAS, I performed separate GWAS for each metabolites and also some metabolite ratios. I used up to 8,299 unrelated individuals in grouping S history from the San Jose cohort. I surveyed 1,091 metabolites that prison in over 50% of the individuals and also 309 metabolite ratios. So after GWAS, I will have the association between all the genetic variants and the metabolites. Then to identify those conditional independent genome-wide significant associations for the metabolites and metabolite ratios, I performed this conditional and joint analysis to identify the leading variant or the leading association for the variant and the metabolites in each genetic regions. So here I'm showing Manhattan plot. So by performing GWAS for metabolites, I identified 1,702 independent variant metabolite associations from 690 metabolites. And they come from about 248 low side or genetic regions. And that's what's showing in this Manhattan plot. So the X axis for this plot is a chromosome and the different genetic positions and the Y axis just showing the p-value or negative log 10 transformative p-value for the association between the genetic variant and metabolites. So each line of the dots is actually a locus. So all those association happens in this specific or chosen genetic regions. And each dot is a variant metabolite associations. So for example, for this locus, we can see many genetic variant in this locus associated with some amino acids, which indicated by the color of the dot and some lipids. And from this plot, the key message is that we find the association between variants from all the autosomes with many, many different types of metabolites. And those metabolites come from different categories, which kind of one advantage of those metabolomics measurement approach because we can simultaneously check the association with many different type of metabolites. So the second part I'm trying to show here is I explored the novelty of the association I fund as well as the general heritability of the metabolites. So on the left panel, you can see this is the number of no one or novel associations for different metabolites under or belonging to different categories of metabolites as shown in the X-axis and the Y-axis, just the counts for the association for those metabolites. And the dark blue one, I think is no one association. And the light blue one as those novel association identified in the current study. So for example, for lipids, we find about 150 novel association and another 150 no one association that have been reported in the literature. And interestingly, we also find maybe 100 metabolites, they do not have the association, sorry, that some lipids that do not have significant associations identified in this study. And this kind of the light green part, which is for the metabolite without significant associations with generarians seems to be have a higher proportion for iso-xenobiotics, which are the metabolites that usually you obtained from exogenous source, such as food or medicine. So here on the right panel, I'm showing you the heritability estimated for different metabolites. And this is a volume plot on the X-axis, just the categories of the metabolites and Y-axis, just the estimate heritability. So in general, we find around like 20% of the variants of metabolites can be explained by genetics, that's the medium heritability. And for some categories of metabolites like xenobiotics or peptide, their medium heritability is lower compared to others, whereas cofactors and vitamins and nucleotides, they seem to have higher medium heritability. So last feature related to genetic architecture or last two features related to genetic architecture, I want to talk about here is polygenicity and pleiotropy. So polygenicity is about the number of low side associated with each metabolites. So it basically saying, whether this metabolite is only determined by one or two low side, or they are influenced by many, many different genes. And so that's something tried to reflect or tried to be reflected by the polygenicity. And in this panel here, we can see that, so the X-axis is the number of low side associated with, associated pyramid tablet and Y-axis is heritability estimated for that corresponding metabolites. So each dot is a metabolites. So we can see most of the, I wouldn't say most, but a majority of the metabolites, they have fewer than two low side. As we can see, most dots is for here. So which means the polygenicity from metabolites at least based on the evidence from the current study is moderate or is not a super or in other words, metabolites are not very polygenic. And we also see this positive correlation between the number of low side for metabolites and heritability. So which just showing that the more polygenic the metabolites are, usually their heritability is higher. Then for the pleotropy feature. So pleotropy is a measure about how many metabolites are associated with each locus. So on the X-axis for this plot here, it's a cumulative percentage acts. So basically we are showing, for example here, about 50% of the locus or low side, they associate with fewer than three metabolites. That's how we read this plot. And interestingly, there are some locus that associate with up to 79 metabolites. So in that case, the gene for that locus, they could be a hop gene for some, for many metabolic process and influence many metabolites. It could be worth further exploration for how that gene involved in the metabolism of those metabolites and how can we use targeting that gene for those process. So that's finished the part related to the metabolite GWAS. Here I'm showing through the results for metabolite ratio's GWAS. So how the metabolite ratio were constructed. So that's this part. We first identified the metabolites that share enzyme or transporter. And then for those metabolite peers, we construct a ratio. And then using the ratio as the treat or as the phenotype for GWAS. So the reason why we are doing or constructing the metabolite ratio, the first is because by constructing the ratio, we can reduce the variance of the measurements and increase the statistical power. And secondly, those material ratio associations may help us capture some metabolic reactions. For example, maybe there are some variant for this enzyme or transporter that really led to this variation of the metabolite ratio in the population. That's something we want to discover. And on the panel B here, I'm showing the association we identified between different metabolite peers. So each band, for the circular plot, each band is one metabolite for the corresponding categories. And each line that link to metabolites is just for the two metabolites for this metabolite ratio. So the darkness of the filling color of this line indicate how strong the association are. So we can see that for lipids and amoon acids, there are a lot of association for metabolite ratios, but they are kind of more, seems to be within their categories. Whereas for energy substrates, we find genetic variants that contribute to the ratio between genetic variants, sorry, energy substrates and many different types of metabolites, which somehow I think is quite met our expectation as many different reactions that involving different metabolites usually need this energy substrates. So by running the GWAS for metabolite ratios, we identified 16 additional associations that were not captured by single metabolite GWAS. Here I'm just showing one example. So we found this RS247227 variant is associated with the ratio for caffeine and paraffin. So, and from the literature, we know that paraffin I think is just one of the derivative of caffeine. And we also find the closest gene for this genevarium is actually the enzyme that catalyze this process. So which kind of supporting, you know, when we are attention to conduct this metamler ratio GWAS or supporting the value of conducting metamler ratio GWAS. Based on those variant metabolite associations or variant metamler ratio associations, we next tried to identify the effector genes that mediated the effect of variant on the metabolites. So specifically, we identified the protein coding genes that near this variant. And then we check whether this variant also influenced the transcript level, which is EQTL or splicing variant of this gene, which is SQLTL first. And then we checked whether this corresponding gene or no one to be involved in the metabolism of the metabolites by checking the HMDB, KAC, and POPCAM databases. So by combining those information, we kind of get or refine a list of potential effector genes that, you know, more likely to be relevant for this variant to metabolite associations. So the expression related relevant genes led to include about 550 genes and the biological relevant check identified about 250 genes. And the overlap part is 94, which is our effector genes. And most of those effector genes enzymes and another 7.4% of them are transporters. And of course, there are some other proteins like some binding proteins and circulating proteins that are part of the, or were thought to be the effector genes for the variant metabolism issues we found. So lastly, to see if this effector genes implicating in any phenotypes and if they can be used as drug targets we explored the linkage of those effector genes with the drugs and the phenotype changes in knockout mice and also human Mendelian traits or disease. So by checking the overlap part, we found 14 effector genes that all have all three sources of information. They are listed here. And these connections may help us identify the metabolites that can be used at the biomarkers for the related disease or maybe we can use the repurposed drug that linked to those effector genes to modulate the level of the corresponding metabolites or associated with traits. So that concludes the results for the M1 part. And here then we conduct analysis to identify the potential cause of metabolites for traits and disease using MR approach. So here is a general design for the MR analysis in this study. So for the exposure, we focused on those metabolite and metabolite ratios that have effector genes identified. Then for the outcomes, we focused on 12 traits from three categories. They are either agent related traits such as bone marrow density or metabolism related traits that body mass index and also some immune responsive related traits such as asthma. Then with exposure and outcome we performed a two sample MR analysis and identify the cause of metabolites and ratios for those outcomes. Buffer only corrected P value or thresholds were used to prioritize the associations and it's an active analysis like metabolic pleotropy check, reverse MR as well as MR result prioritization analysis as colonization were performed to prioritize the more likely or more that the cause of metabolites for the disease or traits tested. So here I'm just quickly show you the metabolites that we fund for the agent related traits. It's not for you to see the details. I will talk about two representative examples later but here just to show by MR approach we fund many metabolites that are related to aging traits relate to metabolism related traits and disease and also relate to immune related traits and diseases. So the two example I want to discuss more here. One is the R rotate with estimated bone marrow density. So through MR we find that the increasing level of genetic predicate level of R rotate is associated with lower estimated bone marrow density. And this is interesting because using an independent cohort we find that the higher level of R rotate is associated with higher risk for heat fracture. So this is consistent with our MR results. We're like both of those results suggesting that the increased level of R rotate is somehow involved in the impairment of bone health. Another example is alpha hydroxyl aloe verite. So from the MR we find that the increasing genetic predicate level of this metabolite associated with decreased BMI. Interestingly we find the effector gene for this variant metabolism association is called lactate dehydrogenase LDHA. And this gene if we knock out in the mouse model those heterozygous mass have decreased total fat amount which is kind of consistent with our observation that metabolites could be the mediator of the genetic effect on the obesity or anthropometric traits. Lastly, since we know many diseases and traits they can also be influenced by obesity. Here we try to perform analysis to differentiate the effect that whether our metabolites influence the outcomes such as asthma and BMD through the BMI or they just they can influence them independently. So this is the code approach at GWAS by subtraction which allowed us to differentiate or dissect the genetic genetic effect on the BMI dependent part and BMI independent part on those outcomes on the disease outcomes or trait outcomes. So here MR then with those association we performed MR again and we found for example the orotate for orotate the association with BMD is mostly through the part that independent from BMI. In another example for citrulline we find the relate to the asthma but at least it's partially it's affected partially through influencing the BMI. So there are some limitations of the study for example most available diseases and trait GWAS from individuals in European ancestry and this can limit the general stability of our finding to the non-European population. And secondly the metabolomics data relatively measurements so the finding we have for example we found this higher level of orotate associated with lower ABMD and we don't really know what exact level that higher level is. And it's to really apply those use of metabolite in clinical practice we need to additional experiments to really quantify the absolute measurement of those metabolites in the blood. And lastly MR because MR using the genetic variants as the instrument to check the effect of exposure and outcomes usually the genetic variants can only explain the part of the variants of your exposure that's kind of led to the low statistical power of MR approach. In conclusion we identify genetic terminates for citrulline metabolites and we also infer the causal effect of metabolite levels and ratios on 12 traits and diseases that are predominantly influenced by different mechanisms like aging metabolism and immune response. And I would like to thanks all the lab members from Dr. Brent Retru's group, Dr. Celia's group and also the collaborators that contribute to this research and also the funding agencies and also the CSA for sharing this data with research community. And thank you for listening. Thank you Yi Heng for your excellent presentation. I think you did a great job at taking a very complicated topic and making it very accessible. So I would now like to open it up for questions. So just a reminder, muting will remain on but you can enter your questions into the Q&A box at the bottom of the Zoom screen. I see a few questions have come in. Okay, so I will start with the first question from Yi Heng Chao. Thank you for the great presentation. Did you find the significance of any locus metabolite associations different from that reported in the literature? So for example, the significant association between metabolite A and locus B was reported insignificant in a previous study. Yes, definitely. But the reason why they are not significant it could be due to many different reasons, right? For example, they could be because of sample size. That's kind of one major difference. Also sometimes the technique they used to measure the metabolites could also limit how sensitive, how accurate you measure the metabolites. Okay, great. So the next question from Andrew Patterson. Thanks for a great presentation. In your Manhattan plot, I didn't see any results for the X or Y chromosomes. Yes, that's a good point. So first, I think for a CSA way, we do have X-chromolone, but at the time, we did not specifically perform X-chromolone analysis because X-chromolone is a very complex topic because the, sorry, I think there's the silencing because the silencing sometimes incomplete. So how to model the variant association with the metabolites sometimes is tricky. So at the time, we want to focus on autosomes first, but definitely X-chromolone should be studied for, to check their association with metabolites, especially should be investigating, you know, six specific design. Great. So another question from Andrew Patterson. Statistically, is there a difference between using the metabolite ratio as opposed to co-varying one metabolite for the other one? That's a good question. I don't really understand what you mean by co-varying. You mean creating the new phenotype using, instead of using a ratio, but using some other statistical way to aggregate their levels? Yeah, I haven't explored that. So unfortunately, I don't have the answer for that. That's okay. We'll see if he can comment on the question if he wants to clarify. Yeah. Okay, so Leela McKinnon writes, thanks for this great presentation. I thought your descriptions of the MR and GWAS methods were very clear and helped me understand them better. I have a very broad general question, which is what would you say the applications of this research are? What are some of the ways these findings could be used in a clinical setting? Yeah, that's a very good question. So as I mentioned, with the relative measurement from metabolomics, we don't really know, for example, which threshold we should set for or rotate should be labeled as high risk or to help us differentiate the high risk group for people with osteoporosis. But at least we kind of prioritize some of those metabolite targets that, you know, and their association with diseases. And then maybe next study could be, you know, get a smaller cohort and then we can more accurately measure those metabolites and then, you know, have a, then create a dosage dependent association between the absolute level of the metabolites and disease risk. And then in that case, maybe we can generalize those threshold for disease risk screening, et cetera, or maybe the orientate actually or the religion can be used as a drug target. So yeah, those kind of potential application for the findings. Okay, so I've got another question from Andrew Patterson. Did you analyze the unnamed metabolites? Wouldn't finding GWAS signals for them provide insight into the potential compounds slash roles? Yes, thanks for your question. Yeah, I did. So basically for my filtering step for to include metabolites, I didn't filter them based on whether they have no identity or not. I do include all of them when GWAS for all of them. However, the tricky part for analyzing unknown or for the mentality without identity is that you can find a strong association between this variant and the metabolites, but it's kind of tricky to know whether that metabolites like are they, you know, that associations because of what, right? So I mean, in my analysis, by integrating the biological evidence, we kind of try to focus on those variant and metabolization kind of supported by, you know, expression or metabolic evidence. But for those unknown metabolites, that part is kind of hard to obtain. However, I do know that if you find, you know, this denavirant kind of near a gene that off your interest, you can always go to the company and, you know, go to the and collaborate with them to further identify those chemicals and there's potential there. Okay, perfect. I think we have time for one last question and we've got one last question in the Q&A. Hi, Yi-Heng, thanks for your work and presentation. This is from Stephanie Chevalier. Can you provide some examples of energy metabolites which are not as obvious as others? Yeah, that sounds like a good question. I don't, I think the only one I remember, I think it's like phosphates. So somehow, they are grouped as energy substrates. Yeah. Okay, perfect. Thank you very much again. That was a wonderful presentation with lots of positive feedback. So before we wrap up, I would just like to remind everyone that the next deadline for the data access applications is April 10th. So please visit the website under data access to review the available data and we will also put a link in the chat. I'd also like everyone to remind everyone to complete their anonymous survey upon exiting the Zoom session. So if, yep, there comes a slide. So our next webinar will be factors associated with developing high nutrition risk, data from the CLSA. This will be presented on Tuesday, March 19th at noon by Dr. Christine Mills, the University of Waterloo. Registration details will be available on our website and we'll also post the link in the chat box. And finally, remember that the CLSA promotes this webinar using the hashtag CLSA webinar. We invite you to follow us on Twitter at CLSA underscore ELCP. Great. Thanks again for joining us today.