25-26

-

we went through the basic concepts of data, the frequency and pop data. There was sort of the first story from, sorry, . And there was actually covered

This is all about the data model.

one way to

So the fourth way is

We start from the phenotype as we discuss

sort of the way.

is to change.

-

forward and reverse genetics. So this is all the things about very classic methods, linkage or the genetics about mouse and how you manipulate forward engineering,

of the way people in

on that.

the way people are using it.

the video. So this is the entire history of looking

from families, mostly disgusting families, and they found the microsatellite markers and they found some genes with very

So I mean, if you're doing a society

mostly a problem

if we're not using that thing in the company

So people using my career using very small numbers, not really millions of people at the time used 100 or 1,000, not enough.

If we increase the sample size, now people are looking at around million-scale GWAS. So many companies, many biotech, I studied at Kaiser, Pfizer, Roche, Genentech, Amgen, all these companies that use GWAS data, that's the thing. So if you go back to the general biology textbook, or if you go back to the high school textbook, No, no textbook discuss about yours. That's the problem.

So reading the papers, that's another standard. Another important thing we keep discussing these things over the semester is that we learn about genome sequencing. We learn about who genome sequencing is. Why? Because this is cutting edge technology. Which means that if you go to a company after two years, three years, you're going to use genome sequencing. None of you are going to use linkage as you learned in our textbook.

None of you are going to use your speakers because it's always tablet speakers. So many of you in this room might use, not might use, must use Geno sequencing data. So this is the point people are moving forward. And all the companies are also equipped with this work. And also this is another interesting point for Korean students. Many of the students are actually aiming some companies like Samsung or

or whatever else they buy, right? And really surprising things is that these companies are really good at profit. They're really good. They want to award

is that you probably remember COVID-19 vaccine. One of the companies developed that is Moderna, right? Moderna based in the UK. And AstraZeneca. AstraZeneca's method is developed by one professor's group. So this is actually developed by UK people. But they don't have any factory. So manufacturing is actually done in Andong City, in Gyeongsang-do. That's where SK Bio So, Bob, Katie, right?

They are actually producing the vaccine products because we have so many factory workers. This is the same situation actually designed by California but made in Huxcombe in California.

So the question is that during the COVID time, the Korean companies

move forward, whether it can move forward advanced technology, whether they can

So that's the situation. And Korea BioTech companies is a little bit interesting position. So in terms of drug biology, most of the companies in the States, like the international companies, they are developing the drugs from various starting points to the ending point. So from trial one, they are trying to do the screening, then they found the lead product, and they optimize it.

and they do the trial one, two, three. All of these processes are quite long. It's, you know, at least it takes 20 years. But like Genentex or Amgen, AstraZeneca, they would literally do that, because they have money, they have less of investment. But it's a little bit different situation, with a little bit different, you know, the status about the, you know, whatever, labor force and everything, right? to people strategically focusing on manufacturing, right?

But the question is that in 10 years time, is the situation consistent? That's an interesting point. Because now they have some money, they have some capacity to do something. So which things they can move forward? One way in the future might be screening. So they are looking at some patients.

and they found some factors we can optimize for Korean patients. Because we have different ancestry background. And we can use the medical information we draw from the hospital and we apply that things into the drug development or treatments. So the things can be happened. But who knows? And one interesting story, I actually mentioned these things. Samsung Biologists or Samsung X is one of them.

They actually bought a company called GRAIL. GRAIL is the company based in Silicon Valley, San Francisco. And they're the company finding some biomarkers in Kansas, you know. And they developed with one of the company-- I remember it's-- I think it's-- you probably know the story about GRAIL and things. and grant, fail, define,

is collected.

So at some point, leadership or statistical points, people are trying to find some way of touching-edge technology. So people are not really relying on the manufacturing things, people are moving toward with more advanced technology. Sequencing would be one of them, AI would be one of them, and whatever technology we are going to mention in the reverse genetics would be one of them. So when I mention these things, maybe a few of them,

I want to go to a or PC in the companies, that's still a good job. But see the timing, 10 years or 10 years, because you are in the position of you can lead some things. So study about cutting edge technology.

So, for genetics is all about this. So, you know. Okay, so a few things we can think about in future time is that it actually has lots of GWAS. GWAS is very interesting that the things people, you know, already have so many tests out of these data sets. So we know about that. But the problem is, few applications. So, people find the GWAS. So if you go to some

database like GWAS results. Then we can easily find some database. So everything in the database. So this is the GWAS catalog. This is all about the research. So even you can search.

It's open to everyone. It's open to your grandma too. Right? So go here and you can search them. So grandma can say about my time on

Or you can search anything. You can search Parkinson's, you can search diabetes, you can search whether I'm sensitive to caffeine or not. You can search caffeine. Anything you can see from here. So distance is all open to everyone. This is a little bit slow so I can continue to explain it. But the problem is that we have so many, so many results in this area. We have so many.

You can see here, you can see that there are so many data and let's say

So this is the thing, nicely summarized, how many genetic variants you know, associate with, it's around a million, right? So many, right?

My point is that there are so many, so out of 7,000 papers, we have so many variants out of here. The problem is that we have million numbers, right? These million numbers, do we know the mechanism out of that? Can we test? Let's put in the situation, so these students go to graduate school and study one gym out of a million.

So one gene is studying mouse. Okay? And mouse. They study that mouse for many years because mouse cannot be made in a few weeks. Mouse takes time, right? You have to breathe in. You have to live with them. And you spend at least four or five years in your graduate school time. Then you see the result out of one gene, one brain out of a million. Is that fat? Is that fat? Is that main cannibal?

Okay, we can put it in different situations. So these students graduate without graduate school, just go to a master's at NECA, they study one variant out of media, okay? And try to develop drugs. Is that possible? It can be possible, but it takes a long time. So that's the sort of the issue, right? So people looked at so many different bioline data setters. There's so many different violin papers that people looking at

So many studies, people looking at so many traits, and they found millions of SNS and thousands of genes. The problem is that we don't know the virus. We don't know where to look at it. One hint, people thought that many of these GWAS variants, PITs means that it's variants associated with traits, right? PITs is the target. And many of them are located in a non-coding region, right? Even they're not indulging.

and we end up changing any protein sequence. We can study. That's another challenge. So always innovation, always advance, is coming from when you deal with the challenges. How you can deal with these challenges. So, I mean, I can mention about companies. If you join the company, if you join QA, QC, you work as one of the factory leaders.

You are operating machines, sometimes replaced by robots, somebody, that's my guess. But in the meantime, you are an educated person at a university, so society expects leadership from you. So you are in a position where you can suggest something for future. So if someone in the, let's say, Shang Tsang's company present coming to you, how is this problem?

and you have to suggest some agenda. That's what people want, right? This is the same thing. In an AI era, AI learn anything from the knowledge existing on the internet, but AI cannot create something. So creating something and suggesting something and collecting some humans, collaborators, team workers, why does AI can't do that?

One point is if you see these strategies, you have to question yourself how I can solve this problem. It hasn't to be 100% angry. At least you can answer the question. So this is one challenge, limitation. So instead of memorizing limitation, you have to think about how we can solve this problem to look at biological mechanisms.

How we can converge this issue. Another issue is the linkages at pavilion. So if you find some variance, around the variance is actually correlate to each other because linkage is happening. So we don't know which one is really the major public. Which one is the major public. And also GWAS is mostly looking at the common variance. right? Because this is the thing. Okay, so,

I also mentioned about the rare variant issues as well. So people are very private variants and very rarely present in certain person's genome. So we cannot compare these things at the locus level. So people are looking at the gene level. So that's why we do the little bit different test for the bottom test, right? So this is the thing. And again, this is the same situation. We're not just looking at the one gene. we're not looking at the one gene association, we're actually looking at all the genes

present in our genome. So how many genes? We have 20,000 genes for protein coding. We have 50,000 genes for the genes not coding proteins. So this is also the things covered in here. So the difference is two. Then we can move on to the reverse standard. So this is the beauty of we can think about this. So we can start from

So this is very classical things here. So one is that we use a cell for the reverse analytics. We use the animals for the reverse analytics. And nowadays people use the computational predictions. We're back to here. So let's focus on the cells and animals first. So if you're looking at the classical perturbation, if you're looking at the classical reverse analytics, People normally call river aesthetics or

the genetics they use the CRISPRs. That's the things in the context of previous textbook. But I want to put a little bit different context. I want to put the words called "perturbation." I think this is the most accurate description these days. What is the perturbation? Perturbation is the thing you edit the DNA, you edit the genome, then you see which one is archery, which one is archery, which one is archery phenotype, which one is archery protein. Ferturbation means that

it has the changes in terms of some entity, right? The entity here can be genes, it can be mRNAs, gene and mRNAs, same good things, or a phenotype, right? So anything we can see the changes, we can say there's a perturbation, okay? So what is the classic way? So the classic way is the knockout or knockout, right? Knockout is completely eliminate the gene, so we totally delete the gene, and see how the gene of high changes in .

And knockdown is a reduced expression with the RNA I, RNA interference. So why? Because RNA is a single strain that interrupts the complement to the DNA. So it interrupts the DNA's expression. So this is the thing that people, when people are trying to reduce the DNA expression, people use the RNA I. so when

Let's say the consequence is the lethal, then people say it's an essential gene. So this is really critical for biological function. Or some specific genocard they can detect, then people say it's a normal function. This is very basic things people normally discuss in all the data, right? Or other things. We can say that the gain of function, which means that another function

added on the biological system. In case people artificially increase expression, we call it over-expression, where people do the expression, which they turn on some expression, then see which phenotype is working, and whether this increase expression is inducing some phenotypes in the experiments, that's how people are looking at the gain of function. Or in these days, people also see some

more elaborated systems, both conditional or inducible systems. So they turn on some genes expression at a certain condition. So here is the tetron tetoff, which is the drug control timing manipulation. Or people use the crissotox, which is very tissue-specific lesion in my mouth. So crissotoxy is the system you put in the system wiping a certain tissue. So let's say I want to control the cortical.

area out of the brain, then people develop some, you know, more than six times using the pre-line, specifically working in a cortical brain region, right? So this is, you know, people trying to avoid and growing this for the constitutive knockout. So this is the very basic thing. Why? Because if you, let's say if you knock out some genes, let's say If you're trying to knock out something

I don't know which one would be, I don't remember. I'm mostly working on the HAPA Institute. So let's put it this way. So let's say gene A is really lethal to early development. So if you want to develop some knockout of gene A, then the mouse will be lethal because it can break down everything during the mouse. But the problem is that people actually want to look at very specific outcome.

So let's say you found some mutation within the gene A, then you put the Cree A on a particular system and see how these things are working in a cortex. In that case people are developing Cree, right? Then you can avoid the embryonic identity because the gene is so critical during the development. But there are a few issues within this method. The first one is:

So back to here. So this case, all the measurements is the things you're looking at the information at the gene level. So you completely eliminate the gene, or you completely interfere in some genes and reduce the expression. So you're basically discarding all the gene expression of the certain gene. So this is the way they would do that.

And this is really good system in early days. Let's say in '80s, '90s was good. But is this still valid in these days? Maybe yes or no. The problem is that, as I mentioned, we have 20,000 genes, right? We have 20,000 protein-coding genes. And most of genes actually study. There's a problem, right? So I can put it in this situation. So one of students, let's say this student, go to company or graduate student.

wants to study the novel genes. And, you know, which genes we can say is novel in these days. Because already, you know, previous researchers, they did all the research, right? So, and no more novel genes in our genome? I would say no, right? Most of the genes actually study, right? So, not really repeating these things. another thing is that

Why we are studying genes? That's the issue. 80s and 90s, people study genes because they don't know anything. And they don't know about the mechanisms of genes. But now people know it, right? So what people want to study these days? People study variants. People want to study real, right? Why? Because we are living in the era of genome sequencing. to the living area of WTIWA.

The problem is that this gene is the gene level. It's not the variant level. That's the question. So what people do is people try to move on to CRISPR. Because CRISPR can edit was single base pair. So CRISPR is you can go here. You can basically do the base editing at the single base pair level. So you can change it to T, whatever it is, to A, C, to C. Right?

you freedom so you can edit the base and see the result right so this is sort of the trend people can read it and this is a interesting stuff right so you have the you have the CRISPR system and the CRISPR system can replace loss of content and gain of content systems you did in the model system right so CRISPR i is basically due to the inhibition right so you can reduce the expression so you can basically bring the

and turn down the gene, then it's sort of the knockdown approach. Then CRISPR-A, CRISPR activation, then increase expression by guiding proventures and turning off the gene expression, that was targeted overexpression as well. And all of these things can be done in the single-base pen level and people easily manipulate in their systems, right? And also these days people also do the Christian provide

The question is, you know, discuss about the model system, class Y, and Christop, whatever, A or I. I mean, these things are mostly discussed in the general biology, so not really new things. So the point is that why we discuss these things in genetics class, right? So this is important things. So back to the company. As I mentioned that the companies, the big bio companies like

and these guys, they do lots of things doing high-fielce screening, so simply called HTS. This is the things people are not really doing these days. This is already started from 2017 or '18, so about eight or nine years ago. I don't think HTS is really booming in Korea.

So many, so many, so many, so many RNA from CRISPR into the system. So into this system means that it's normally cells, right? So cells is that you do the culture of some cells, you basically use that sort of RNA into the cells, then you see which RNA is actually working for some overexpression or reducing expression, right? So people are not looking at

one time CRISPR, one time CRISPR-A, one time CRISPR-I, or people not looking at one time knockout lockdown, one time over-expression, complete expression, etc. What they do is they put everything into the one system with a large number of guide RNA. So this is really critical. Why people doing these things? Okay, back to the GWAS. So,

Let's put it in this context. So you're studying diabetes. Okay, so let's put the Wigobi or Manzara. This is good, right? So Wigobi and Manzara is a big market, right? You probably see the news about some young female that had problems about the pankreas by Wigobi. So if they got the Wigobi, and they have some countries issue or kids issue.

that prevalence is a little bit higher than MAI. Why? That's an issue. So let's say people do the GWAS and see the side effect of WIGOB versus NO. People can do that study. Because of WIGOB, so many people already got the WIGOB. So people screening, people collecting, Some groups testing Wigobe versus

UK viral bank general population. Then people compare that because UK viral bank people they haven't got the Wigobe because it's way before them. And they do the GWAS and they found, let's say they found 4,000 SNPs. Okay, 4,000 SNPs. So these students working in Wigobe's company, then out of 4,000 variants, he pick up one variant and making mugs and testing virus chemical.

and how this need is working on the society. It's a valid approach. But the problem is this is too slow. It's going to take three or four years. Probably there's another Wigobi system. Wigobi is more than 10 or something within that year. So the company hired 4,000 people making 4,000 mouse. Is that possible? It's not. the company always trying to reduce money.

So what people do is not making mouse, instead people making 4000 variants into the valid RNA and put it into the set and screening. So instead of making 4000 different mouse, they were making 4000 different variants, put it into the set, do the screening and find one of them. Prioritizing one of them. Prioritizing one of them. Prioritizing, they basically ranking.

So if you do 4000 variants, 4000 testing, then choosing top 10 or top 20, they can move on to mouse X cameras. So this is one example. So another example we can say, let's talk about the depression. So people have done GWAS for depression. I remember 3 million people GWAS. This is the largest one.

and people found 300 different variants. And people can make 300 different mouths, but that takes time, right? So what people do, instead of making mouths, 300 variants, put it into guide RNA, put it into neural assets, and testing which out of 300 variants, which variants are actually working, mostly working

the most notorious way in the neural. People can test it. So choosing top one or two variants, making mouse, developing drugs. Can you see the entire logistics? So starting from many different variants, prioritizing using this system, and narrowing down pick up one or two candidates, making mouse, making drugs. the things people are doing. Then also, you might have some questions.

Depression is not a neuronal problem. It can be an astrocyte problem. What people do is making 300 RNA, put it into astrocytes, do the same thing. Then compare it between neuronal and astrocytes. Then prioritizing it. Isn't that fantastic? It's way better than making 300 mouse. It's taking forever, right? Okay, so put it into another context. I give you the dead questions.

Think about this. In a basic science, people study C. elegans or Drosfilla. Is this still valid for human disease? This is really their question because all professors don't like it because Drosfilla, C. elegans, and other friends of our model system. But my point is that this is not really valid for human disease. First question,

They don't have neurons. They have junction. They have homology to human, but not exactly the same as human. Why not using the same system like in here? Why do we have to use George Pilar? So when you learn in high school, when you learn in general biology, textbooks probably say to you that we use the model system like George Pilar, C. elegans or gibberish because they're fast. But these are not faster than the

So that's why big pharma companies are not using Drosfila. But this is really their system, especially in Korean society. All professors have hierarchy. If I say it publicly, Drosfila is not the system we study for them, people are not lying. But you have to choose, you have to think about leveling. If you want to study disease, if you want to study mechanisms, you have to study

method. This is the big thing in these days. Have you heard about the new alternative method called MAN? This is the state's NIAs, Korean NIAs, everyone is talking about that. People are not really doing animal research anymore. For the ethics, for the validity, validity is important things. We do the mouse experiment for depression, we do the mouse experiment for autism, we do

Do they really have the phenotype as a valid issue? Right? So, I mean, why I mention these things? Because in a very short time, you probably choose your career for the companies or graduate schools. Many of the academic situations are using very outdated systems. C.L.A. Gospital mouse, your visual. But if you really want to study patient context, human context, disease context, think think about the fidelity.

This is important things, right? The trend is moving really fast. Okay, so this is another way. Verity is really fun things. I like this guy. This is not the same thing, but for your future. Volting ball is a...

So he wrote about the validity

So the point is that people do the reverse genetics like on animal systems. We have to think about the validity in terms of the three domains. So one is the base validity, second is the post-tractive validity, and the other is I think theoretical validity or something.

So, the base validity and construct validity is that we can compare the hormones between two different organisms. So, let's say we compare the mouse to human. The mouse has really conserved the hormones in terms of the nervous system. So, when you study neurons, when you study brain, the mouse is okay. In terms of gene homology, they're quite similar. They really share the conservation.

But the problem is that they don't really share the developmental timing because mouse only takes 14 days for brain development. But human takes around 300 days. So we totally take different time, right? So when we talk about the base validity, this situation is not really accurate. And also, we can also think about the constant barrier, which is the situation you compare to phenotype.

So let's say you compare diabetes. Can you say you're making diabetes mouse versus human diabetes? It's really the same? Something same because they're fat. Human diabetes is fat. You know, usually metabolism, you know, human mouse same. But mouse not really, you know, eating chips during Netflix time. Right? It's so different. The technology is different, right? I'm lying on the couch and eating some donuts, Netflix time.

that might be my diabetes, most different, right? So this validity is really the problem, right? So if you work in a company someday, if you want to develop some drugs, then always you're faced with the validity question, how we can translate into the human context, how we can design better, right? So especially in the genetic studies, how we can deliver this genetic association, genetic findings into

and which model system should be used in the drug development. All these things are the problem you have to think about. So the solution as we discussed here is CRISPR. You can test many variants of interest in a one shot. This is the same thing as how people find critical issues.

people are discussing these days is the "partoxic" right you have to remember this name "partoxic" literally "perturvation sequencing" so what people do is people introducing CRISPR RNA which is as we discussed in here so you put 300 different you know variants into the RNA whatever the numbers that number is is one right so 300 variants into the system then people do RNA sequencing at the single cell level we're not we're not really talking about these things you know

will be in the later section, but briefly mentioned that this is the system you measure gene expression at a very certain type level. So what people do, people let's say back to the depression example, you you design the 300 different variants and you put the body's RNA and you put it into the cortex which is one of the part in the brain right. Then you do the single cell RNA gene expression which is within the cortex you can measure the gene expression of the neuron, you measure the gene expression of the estrocyte, you measure the gene expression of micro RNA.

microglia. At the same time, you see that out of the three months, which one is really affecting on certain cells. So people use this precoce to track down the general variation in cell levels. So the technology has appeared in 2023 in nature, but the prototype is coming at around 2019, right? So this is more mature technology.

And he said in 2026, "Toxic, they're real music." So you probably remember that the, what's his name, Jackson Han, the CEO of NVIDIA, this famous guy these days, right? So that guy mentioned that he wants to study biology if he choose another major and people muck around. But the question is, the interesting thing is, NVIDIA is already doing biology.

So what they do, they apply AI to iFOP, that's one story. And media apply their GPU AI into learning cryptopsy. Why he learned about, you know, why he training, why media or why Google, even Google too, why they training these cryptopsy data into the AI model. This is really critical. You have to think about where you crave. Okay, I'll put it in this way.

So the first one is, as I mentioned, this is the high throughput screening. So let's say you put 300 different genes, 300 different genetic variants into CRISPR and put it into the cortex. Okay? So that 300 is selected from the GWAS people doing depression GWAS with, let's say, 3 million people. Okay? You follow that?

You get that data out of the cell level. That's fine. Three million people with depression. People found 300 genes. What about you looking at one billion people for depression GWAS? You're going to find maybe 1,000 genes or 2,000 genes, not 3,000 genes. That's for sure. But doing Jiwa for a billion people

possible can be done. So what people do? People training data out of 300 genes. Then they estimate another billion people. But first, it's an AI. Because AI takes the data for the training for certain phenomena and predict something. That's how Pertoxy used in many different AI companies' job development.

that will become much more powerful because everyone rushing into these rates. They generate more toxic data. They train more AI models with the system. So that's really trending in these days. And people say this is the AI based virtual set. Okay, virtual set, right? So in a virtual situation, They train this data

they predict something. It's really important one way. And this is one dimension. And second dimension you have to think about it. It's the same story as I discussed in early days. So in an AI era, people use this Pertoxic. That's what we discussed. And the question is why you cannot train data set from mouse.

You have to question to yourself why? Because? Okay, let's assume this. As I mentioned, the mouse takes time for generating data. So these students taking one gen out of 300, then spend five or six years generating data from mouse. These students take another five years. five years. This is taken only five years. Okay? And they are in the different lengths.

That's normally happening. Then after 5 or 6 years, these students get generated some data, put it on the database, and some AI company collecting their data, collecting this data, this data, this data. The question is that whether they are same, whether they are proceeding in the same procedure, that's a major issue. If you are looking at some mouse papers, even they are published in Nature or Science or Cell,

If you track down that mouse study to the previous reference or after reference cited that study, there is no way we can see the regularized method. That's a problem. That's a problem. Because the mouse studies or models, they say they're so subjective that based on the observation, then I really standardized into the data generation. At this point, you have to bring some data.

So, one example here, you go to some company, you go to some graduate school and you study mouse and you do some testing and you measure that mouse this behavior and mouse this behavior sometimes. And you record that thing by your hand or computer by your manual observation. Then at that point you have to bring some questions. Is this really accurate? Is this really accurate?

reproduced by others, AI can learn this. That's a problem. So, one example is if you want to study some mouse, then the mouse has to come from the same single house, and it has to be measured by some video camera, and the same procedures, same data type, and these things can be used in AI training. But--

current situation, current data set is not. That's why big pharma companies are moving on to this technology because it's really regularized. So you have to think about how it is used. And also, in the same line, nothing for the technology called MPRA, methodically per photo essay. This is really a classical method if you read this.

reporter assay. This thing is even in the general biology, but that's only one single X payment. It's not really single reporter X payment. This is the massively parallel reporter assay. How? Reporter assay is very simple things. You basically create the backup, then you put the gene, then you check the expression by cell transfection. That's basically the reporter assay is working. The method in parallel to TESA is hard working.

You put many barcodes into the system, then you can get the message.

Okay, so this is everything and I briefly mentioned about the things in Korean and in Minnichan. If you remember this, you will need to explain what you are going to do. You will need to understand what you are going to do. Because you can understand what you are going to do.

I've heard about the mouse and reverse that I've been talking about. That's true. I'm still talking about the right and the right. But I've been talking about the career.

What is the design of the data? The data and MPI are very much more than the data.

If you go to a university or a university, you can use a mouse to do a machine.

I'm going to test the three-champer test. I'm going to try to see the three-champer test. I'm going to try to try to see the three-champer test. I'm going to try to see the three-champer test.

- Yeah, yeah.

But the robot is a lot of the problems. It's a lot of the problems.

I've been using the G-watch and the Cypancing.

It's a good thing.

It's a point to the work.

So this is everything about the reverse genetics. As I notified, this Wednesday is no class. And next week, I noticed you that next week I'm on the business travel, but I found that on Monday I had lecture. So next Wednesday, I'm sorry, next Monday we're going to have lecture. And Wednesday this week, Wednesday next week,

one way on Slack and see you next Monday.