 Okay, well thanks so much for inviting me and it's great to be here in beautiful Paris So I'm gonna talk about this genetic landscape and before I get going I just want to say that it's it's all a major collaboration with Brenda Andrews who's the director of our Institute here at the Donnelly Center and I'm gonna try and explain Why this graph is a model of the cell that's basically what I'm gonna try and tell you guys and Yeah, it could be anything right and But you know here's the model that we have so far and When you think about it you know, we know a lot about the parts of the cell, but we really don't have a good model for how they're integrated together and Omics functional genomics systems biology Tries to do that in a quantitative way and it's it's It's not that we're very good at it at this stage But it's sort of the beginnings the beginnings of it and our project has kind of contributed to this So I'm gonna talk about the budding yeast which a Bunch of us work on here Georgina and David are leaders in the field and it's a it's a eukaryotic cell So it's a billion years further evolved beyond prokaryotes and it's got a Nucleus and you know a mitochondria with its own DNA. It's got all the goodies that we find in human cells and so It's been a major model organism and in fact There's a thousand yeast labs around the world and we all feed data into one big common database that Is put together at Stanford by SGD so yeast has 6,000 genes and 16 pairs of chromosomes and you compare that to a human has 20,000 genes 23 pairs of chromosomes so it's it's a pretty sweet little cell and The yeast community this international community. There's labs all over Europe major labs in Japan and of course in North America and it's expanding to China and India in Rapidly as we speak Not in Russia. There's nothing in Russia Don't you think that's weird? but But anyway the goal is to figure out and so we've got a graduate student There's six graduate students each lab. We've got a graduate student for every single yeast gene, right? And you'd think that we could figure out how this thing work this cell works and we could model it but We're so far from it because it's because we have to figure out how they're all connected to one another and how they're talking to each other and the dynamics of it all and That is way beyond the capability of the yeast community, but any community studying any form of life, but just to give you a View of that here's 16th the 16 chromosomes in each line represents a gene so there's these genes are just lined up one after another on on each chromosome and One thing you can do with these because it's so easy to manipulate This organism genetically Partly through homologous recombination is we can go into a diploid so it has two copies of every chromosome a diploid organism and delete one copy of each gene and Then the beauty of yeast is you can take it and convert it from a diploid to a haploid Where it only has one copy of each chromosome and you can under those conditions ask whether that gene is required for life and So the yeast community did this Deleted each of the 6,000 genes one by one in a diploid Converted it to a haploid and asked for the gene was required for life and 1000 of the 6,000 genes when deleted individually is required for life 5000 you delete them in the cell basically barely knows it's happened Okay, and this is true not just in yeast. It's in true true in all forms of life basically any sophisticated form So any coli In yeast in worms and flies and mice and in human cells Only a fraction of the genes are required for the basic Yeah cell division And replication of chromosomes most of the genes are Appeared to be functionally redundant Okay, so that you can delete them and and they must be doing something because they're highly conserved, but There it looks like there's a backup plan for whatever the role of this is so a nice example is DNA repair So there's many pathways that control DNA repair different pathways and they're they're required under particular conditions But in a laboratory setting you can remove non homologous end joining and the cells perfectly fine It can repair its DNA through homologous recombination for example But if you remove both those pathways Homologous recombination on hummus and enjoying them that the cell has a problem so we think a lot of the re many of the reasons these cells aren't required for viability is because this the cell is wired with backup systems and so that the analogy is it's like a Finally tune and jet engine where if something goes wrong, we've got another way to keep keep the plane up in the air so Why maybe it's a few years off because in some way, but why wouldn't the cell work on these 1,000 remaining essential genes and have backup That's it. That's a good question. I I've never I can't figure that out except to say that You know it wants to streamline everything as as much as possible and then but still then why wouldn't there be a backup pathway for that thing? Yeah, and there's really only one proteasome It doesn't make another copy of all those genes and a and another I mean, there's other ways autophagy and things to degrade proteins, but You know the major Essential complexes are things like the ribosome the proteasome and whatever Yeah, yeah, no, but but that's not like so Craig Venter tried to make the minimal genome right and basically if you read their paper They tried to take these thousand the equivalent of these thousand genes and see if it would work and then oh, they were really surprised when it didn't work You know so yeah, you won't even give me any feedback He didn't bacteria but they they basically their idea was this and then if you read their paper They're really surprised to rediscover synthetic lethality Is not the right way to think yeah, I don't think it's the best way to think about it I mean and it's way more complicated than just one pathway backing another one up like so I'll when I show you So it's just a it's just a very Simple Beginning thought that's not correct. Can I just say something that for example some of the genes become essential once you give them other Other conditions for example stress. So now they become essential. So they're not essential But they are essential. So they are more essential genes for other things. Yes So it's the number is a little higher, but still why did why they don't have a backup. Yeah, there are some backups Yeah, I mean It's it's complicated, right? Yeah Okay, so Anyway So here's what we're gonna do. We're gonna explore gene function through this genetic interactions and this is a This isn't a term that does not mean of any it does not mean interaction physically Okay, it just means like you know Some kind of more like a mental interaction You know sex no, no, I don't know what it means and But here's here's a definition that a genetic interaction occurs when mutations in to we'll just focus on to right now But it could be more two different genes combined to generate an unexpected double mutant phenotype Okay, so if I have mutant a and I know what it does to the cell and I have mutant B And I know what it does to the cell when I put those two together I expect to see both things happening, right? That's the additive effect of the double mutant But we call a genetic interaction when I take this and this and put together then something crazy happens, right? something unexpected and that and So the guy that we all know and love that articulated this to the East community was this guy David Botstein and He really did articulate this when you go back and look at the literature and Which is kind of surprising because we we think that we we the East community thinks it's been looking at genetic interactions for years and years and years But it wasn't actually, you know, I'm a grad. I'm almost going to graduate from my PhD at this point and this paper from the famous Peter Novik and Botstein comes out and they're looking for Suppressors, this is a genetic interaction of yeast actin mutations and so and this is This is a yeast cell stain for filamentous actin with rotamine Filoidin Yeah, maybe turn that yeah, cuz that's a good that's a good picture. Yeah. Yeah, I know I know Yeah Yeah, so Of course David is the yeast accident expert But look at that. Look at that the mother cell the daughter cell and we've got these patches here that are David's gonna talk about that are all involved in endocytosis and then these cables that are tracks for mice and motors that Build the bud by blowing it up and yeast only has one actin gene Okay, so this makes it and it's essential if you delete it and you don't have any of these guys the cell can't It can't form any shape properly and it can't divide but And if you go to a worm anemoto, they have lots of different actin genes and so you can't really do this type of experiment with a worm so what what Novik and Botstein did was they mutagenized the actin gene and This dash one means it's a particular allele. So some particular mutation and they isolated a mutant that was alive at 23 degrees and dies at 37. Okay, so this is a temperature sensitive mutant and Yeast and and fly geneticists Use this to study essential genes in a conditional way So we can look we can grow them at a permissive temperature and then shift to the restrictive temperature and See what happens reveal the the essential phenotype and One way to think about it But there's lots of different things that happen the protein might not be made at the restrictive temperature might be degraded But the classic idea is that you have a mutation in the polypeptide chain that Allows it to fold normally at the permissive temperature But when you put energy into it you miss the protein misfolds and you reveal the mutant phenotype and there are examples of that so So what they did was they looked for suppressors of actin mutants and so what happens here is if you combine act one dash one and you you basically Select for a random mutation the genome that suppresses this temperature sensitivity and you you identify a new gene that way and So act one dash one plus this mutation the cells are alive at the restrictive temperature And we'll call this a this is an example of genetic suppression And we'll call it a genetic interaction because it is an unusual phenotype you go from dead to alive, right? You know, that's pretty powerful and and since it's a Gain of a phenotype. We'll call it a positive genetic interaction Yes, so when they did this it was completely unknown they wouldn't even have known what this gene is It turns out that it's involved in phosphatidyl and hostile signaling Which is a signaling pathway that can regulate actin dynamics through other proteins, but we probably still don't know the mechanism But there's some major functional connection between these two genes Okay, and it this is not that case, but I like we've been studying some of this, but I'll just show you a phenotype So here's neo one and it's a temperature sensitive mutant. So the mutant will grow at 22 degrees It's it's totally dead at 38 and now if I take that TS mutation and Delete another non-essential gene YMR 010W perfectly alive at the restrictive temperature You know, it's these are powerful phenotypes and in this case I'll talk about layer neo ones of flip A's that flips phosphatidylsirin ethanolamine from the outer leaflet to the inner leaflet and this thing's probably a scramble is that antagonizes it and if you just get rid of the antagonizing enzyme then There's a there's another backup flip A's in there that can that can take over Anyway, that's the kind of phenotype so So we have this genetic interaction and then here's the here's the fun part Here's another allele of actin so a different mutation in the actin gene and When you take that mutant actin and combine it with sac 1-6 the cells are dead even at the permissive temperature Okay, and Again, I was kind of trying to look up the mutations. I don't think we understand any of this, right? We should be able to we should look it up though and try and figure it out Anyway, so this is an example of the So this is a cool. This is a rescue positive genetic interaction. This is synthetic lethality Which is a negative genetic interaction? so And this is a very famous one that I'm going to talk a lot about But this is unusual like Mutation this is viable at 23 degrees. This one's perfectly viable at 23 degrees, but the double mutant is dead, okay? and This this would say to us that these two genes are working together to control an essential function in the cell right and presuming that's actin assembly and This thing's kind of cut this mutation compromises the actin gene itself and then this mutation Must you know exacerbate that somehow and have something to do with all the whole actin dynamics Okay, so synthetic lethality is this very simple definition where delete one gene and the cells are viable delete another in their Viable, but the double mutant is lethal and this is a rare a rare event if you test all If you take gene a and test all possible double mutants you'll only find a few cases like this and and and because it's rare geneticists Thought that this must be important and it must be telling us that these two genes are working together somehow in the cell and what what David did was he went into the literature and Figured out that this this combination of two mutations leading to a dead phenotype has a name called synthetic lethality and it was coined by this guy Dubjansky who's a fly geneticist in the 40s and And and so synthetic lethality occurs when the combination of two mutations either by itself lethal causes lethality and The wild thing is that Dubjansky was working with flies from an outbred population So he was going out into the wilds of California isolating flies Crossing them to one another and he could trace defect in development to an allele of one gene from one parent and allele of another gene from another parent and Because he's doing diploid genetics. This could not have been trivial, but he was seeing it and There's a couple implications there one is that it occurs Not just on extreme mutant alleles that we make we make in the lab It's occurs on alleles of genes that are out there in natural populations that carry natural variation and therefore It may underlie a lot of genetics as we know it so genetic interactions might underlie a lot of genetics But there's no proof of that so anyway There's a lot of combinations of genes so there's 6,000 yeast genes so there's 18 million gene pairs and I'm maybe I'll talk about triple mutants But there's 36 billion gene triplets, right? So when you start thinking about Genetic interactions combinations of alleles driving inherited phenotypes. There's huge potential for this to take place in any in any genome So John Pringle was in You know his famous each nest at Stanford now He was in Heartwell's lab where they developed this sectoring assay and and along with Alan Bender They carried out the first Synthetic lethal screen and so this this works you can set up a plasmid with a gene on it and You delete that same gene from the genome and then you mutagenize the cells and you look for Secondary mutations that make that plasmid Essential for viability so the cell can't lose the plasmid and so you know this guy can lose the plasmid this one cannot And so it's probably carrying a synthetic lethal mutation with the query strain that they were looking for so it's So they started off With query mutations in BEM 1 and BEM 2 they're involved in butt emergence And they mutagenize those strains with this plasmid loss thing and they would find synthetic lethal interactions with MSB 1 and and This was the first screen. It took probably a couple years to find that one synthetic lethal interaction Because there's a lot of work and so the beauty of this set of Viable deletion mutants is that it opened the door for systematic genetics and So we could now have a matrix of 5,000 mutants and we could Cross any mutation into that whole set and make 5,000 double mutants Because we can make temperature sense of alleles of the essential genes. We can also Look for synthetic lethal interactions just like David botzine and Peter Novick did by Mutating essential genes getting TS alleles and then doing crosses at a permissive temperature where the cells are still alive But they carry a mutation that compromises the essential gene function and so this is where Brenda and I got together because We started to think we knew that the yeast community was creating this Reagent set and we started to think well the double mutants are going to be the most fun And so we'll try and set up a system to make the devil mutants and and we can do it because yeast genetics is just so easy to do and and so I often wondered why did I spend five years of my life studying yeast mating as a young adult Where and I should have been studying money right and making money like my brothers did and And then but in the end if you study mating then you can figure out how to make lots of double mutants so I can't make millions of dollars, but I can make millions of double mutants and So then you can explore gene function with synthetic lethality, so You can make you can cross this and this and find these things and so the way we think about it and again, this is just a Model is that if you have two pathways like homologous end joining and no Homologous recombination and non homologous end joining and they both and pinch on an essential function You can tolerate a mutation in one because the other is still functioning or this one the others functioning But a double hit will kill the cell okay, so this is a Synthetic lethal interaction if and it often occurs between sets of genes that are encoding molecular machines that work together to Drive essential functions So if I remove one component of this machine and I screen all other genes for synthetic lethal interactions all identify Synthetic lethal interactions with all the components of the backup pathway And that's that's true for this component and this one So then genes that have similar patterns of genetic interactions are in the same pathway or complex and So if you have a protein complex we can often see if these two complexes are working together We'll see coherent sets of synthetic lethal interactions between all the components of the of the complex And they can be strong or they could be weak, but they're all coherent. So we know they're They're true or statistically real so Lee Hartwell won the Nobel Prize for his cell division cycle He he was really interested in synthetic lethality and he articulated these two concepts in the around 2000 so about 20 years ago one is that if you have a cancer pathway that's in a it's a non essential gene set and You you can then kill cancer cells specifically by targeting synthetic lethal interactions that take advantage of those vulnerabilities in in the cancer cell, okay, and And that works like mad everybody's going nuts with this now because when he talked about this He couldn't do any kind of genetics in human cells, but with CRISPR. It's just so it's so Painfully obvious that this can be done Hartwell was arguing you should do it in yeast and worms and flies and if it holds true Then we'll go and and you know tested in mice and and maybe human cells somehow But now you can do screens like this and find it But then the other idea he talked about which is even wilder is this genotype to phenotype potential and so He was thinking of okay now. We're gonna start sequencing everybody. We're gonna get their whole genomes and then we have to interpret it and We'll have to interpret natural variation and if natural variation can combine either as Digenic interactions or Trigenic interactions to modulate pathways and lead to Inherited phenotypes we have to know the rules of how genetic networks work, right? And and we have to figure out how prevalent it is so And that's that's a wild idea and I've I've drunk the Kool-Aid because We write our grants and we cite Lee Hartwell's paper all the time and we say this is why we you should give us money To do this, but there really is almost no in no evidence of genetic interactions and associated with human genetics And the reason is because it's a it's a statistical game game that we can't win 20,000 genes 200 million possible gene pairs and if you look for a say a thousand people in a GWAS study You just don't have enough statistical power to find a single genetic interaction And even more than that and when you think when you think that if it's if the genetic interactions are mapping Say two different pathways that impinge on you know something that's important for Disease state then there's many different pairs of genes That will lead to the same phenotype So you're trying to find multiple different pairs of genes to associate with a disease state, right? And so that that's another thing and and and Eric Lander So there's this missing heritability for any any human disease and this is where I don't know what I'm talking about very much except for what I'm saying so Bear with me on that, but if you have For any disease you look at it usually we only can explain about 20 to 30 percent of the heritability a lot of it's missing and You know, there's many different thoughts about this That it could be rare mutations blah blah blah, you know, we just don't know what's going on But it's quite possible and it looks from yeast models that Definitely genetic interactions are gonna be some significant component of that That being said there's almost zero evidence for it and Eric Lander walks through this This is why he instead of calling it the missing heritabilities He's calling it the phantom heritability as if the variation is there. We've documented it We just don't know which combinations are leading to the phenotype But he sort of goes through the math and says to find one particular genetic interaction You'd have to have around five hundred thousand individuals So someday we'll have that when our sequences are on our phone and all our records are being monitored by Google or something but But there's there's other ways maybe to get at this and our collaborator Chad Myers has a has a some ideas and all If I get to that We can talk about it But so anyway, we started based on Hartwell's suggestion Um Basically and just the the pure You know fun of it. We decided to try and make all possible double mutants and yeast Yeah, thank you. Thank you I'm definitely gonna skip some things, but yeah, and so but I can go to the highlights Anyway, we we really Amy Tong was a graduate student who started this She's not on this picture, but then Michael Costanzo took it over. We have a team of technicians and they feed These locally designed machines To just make all the double mutants so we can just replica plate and it works the computational analysis is is The tricky part and Chad Myers was in Olga Troy and Sky's lab at Princeton And he has a team of postdocs graduate students Anastasia Was a graduate student in with Brenda and I? And she's now at Calico and these guys really formed a team and it's It's very simple Our model for You know how to score a genetic interaction if you delete a gene and you reduce the growth rate But you don't kill the cell by a defined amount So that's this one reduces it to point seven delete this gene Produce a point five if I delete this gene in this background I expect to see a reduction of fitness by point five. Okay, that's our that's our working model so we the the fitness of the double mutant the combination of the two single mutants we expect to lead to a phenotype of point three five so Synthetic lethality if the double mutant is dead then That's a negative genetic interaction, but if the double immune grows better than We'd expect that we call that a positive genetic interaction so we quantify negative and positive genetic interactions and we we do it just by looking at colony size on a plate and There's all kinds of systematic errors that we we go from noisy data to data where we we normal it where we you know Normalize all this stuff out and get at the true genetic interactions So then Basically what we get is a vector of negative and positive genetic interactions So I delete gene a and if it shows a positive generic reaction with B or with C and then negative for these other guys I we score it and we quantify it and We do that for we cross gene a to all six thousand other mutants and and then Just just score these things and You get basically, you know, here's an example So rad 52 is involved in DNA repair and we get negative interactions with all these other genes that are very have very well Defined roles and DNA synthesis and repair so you you connect in General functionally related genes. I Mean there's noise in the data And if the gene is a signaling molecule that controls many different pathways Then you connect that that signaling molecule to many different pathways, but if you're a focused Player in the cell then you end up connecting it to other focused players in the cell and So genes genes that are in the same pathway have very similar genetic interaction profiles Okay, so that's simp. It's easy for us to define functional modules because we just cluster their genetic interaction profiles and if we have two pathways that are Backing each other up then we'll see that they share synthetic lethal interactions and that it's a coherent set all the components of each pathway We'll you know, most of them. There's always some false positive and false negatives in our noisy data But we'll see lots of connections more than you expect by chance I've drawn Positive interactions between these components because if it's a non-essential pathway, that's what we see and And that's this is just because of our model Basically if you if you have a non-essential pathway You kill the pathway if you delete a component and then the double mutant looks just like the single mutant Okay, so the double mutant look just like the single mutant not like what we expect And so we score that as a positive genetic interaction just simply part of our model but If it's an essential pathway you can have you have two essential pathways backing each other up if you combine two TS mutants in the same pathway you often the cells often die So you see negative within and negative between So we look for these types of network motifs and I'll come back to that But in order to create that that graph I showed you at the beginning Of the talk the thing that I was referring to as the model of the cell What we do is we just measure the Pearson correlation coefficient of the similarity between two genetic interaction profiles So a and b Pierce they're you know very similar Pearson correlation of point 8 so we connect them by a short a Short-edge so all these guys are connected by short edges These ones are all very similar to one another They're connected by short edges, and there's some similarity here. So we put a longer edge like that Okay, so we just Basically we create a graph where if a gene has a very similar profile to another one It's it's close and if it's you know more distance further away and and because genetic interaction profiles are a quantitative Measure of gene function. We're basically just sorting all the genes in the cell out Based on their functional similarity, which we've measured through this genetic interaction profile and So that's what creates this map in over 15 years. We we put it together. So we started Mapping here and we started quantifying stuff here So that's why we now have positive and negative at first. We were just sort of looking at the cells and saying oh that looks bad and In the end we map half a million negative interactions and slightly less positive interactions when we test a bunch of different Gene pairs many of which we tested twice and then you create these three Maps where we have non-essential network the essential genes are very rich in genetic interactions So temperature sensitive mutants are on the edge of dying and you can give it another hit and it'll you'll push it over the edge the But the and so you see lots of interactions with them And it but again they they sort of are often functionally coherent And so you're you're talking about genes and this they interact with genes in the same sort of Essential essential roles in the cell and we can put it together to create this global network and you can see You can see there's clusters of genes that have similar genetic interaction profiles And these are made there's hundreds of genes in there And so we would think we think these are them Defining bioprocesses the fundamental bioprocesses set of bioprocesses in the cell and Anastasia came up with a nice Way of Looking at this so she'll take a node and then she'll go at a certain distance around the node and she'll Identify a set of genes and then she'll see if they're enriched for annotations this genontology annotations And so we can she can scan 4,300 different genontology Annotations and look for enrichment on our graph all automatically without without any sort of Human bias and that's what that's what puts this together And so these are all annotated Automatically, and so we've got you know DNA since some repair over here Secretion over here. I'm not showing genes in the middle But those are the signaling molecules that control many different functions So they're not enriched for a gene ontology, but there's still you can see there's lots of genes in here and they're they're often regulatory molecules Yeah, yeah, exactly Yeah, yeah, it's completely automatic and and Then we have a website that you can go to The cell map org put together by Matei and you can You know put a gene So this is one of the ones that John Pringle looked for a BEM one if it lands in the polarity part of the manager landscape You can look at all its negative interactions It's positive interactions. You can dial them dial them up on the on the Stringency test here you can annotate it We have a feature there where you can say What are the enrichment of all the genetic interactions for BEM one and there you can see they're all basically hanging out in the in the same Bioprocessing cell polarity Which is what you expect for functionally coherent interactions. There's this tRNA Modification system that is highly linked. We don't know how or why but it's linked to polarity and we often see it Coming up you can get a list or ranked list of genetic interactions From the most extreme negative going down and this is this is kind of wild because here is Here's the one that Alan Bender and Pringle identified it's right in the middle So you can see like when you do a screen, you know, there's hundreds of interactions where you find a few It's just hard to it's hard to find them and so systematic genetics just totally Change the game Okay, so insights to remember here these genes have similar patterns of genetic interactions but we can also look at the genetic interactions themselves and so We view the cell like this where if you know, there's a Machine in the cell we now want to know all its synthetic lethal interactions or cases where you delete something else And there there really isn't The double mutant phenotype that it expects so a positive genetic interactions that could be suppression Or it could just simply be that the double mutant, you know, isn't quite as sick as we'd expect, you know So here's the origin replication complex synthetic lethal with this other genes involved in DNA replication But positive interactions with the Torah Signaling pathway, right and we would argue that this is the kind of wiring diagram that we want for All human cells, right so that when we go into a tissue We know what the wiring diagram is and then if we have someone's personal Genome sequence we could start to interpret variation that when combined might lead to Collapse of a pathway in a disease state or maybe it's not that big an issue because we've got another mutation over here That might lead to suppression or something and so That gets us back to this Hartwell idea of genotype to phenotype relationships and This is what Chad came up with so Jads Chad's developed this thing called bridge. It's there are some papers published on it the major paper is still out for review but Because we have these modules in the cell where one pathway or one molecular machine backs up another Chad decided that maybe what we should do is instead of looking for a particular genetic interaction that we could gain Statistical power if we monitor a variation that Co occurs but links these two pathways more than you'd expect by chance. Okay, so by taking all the genes in the cell and and dividing them up into functional modules, right and then looking for connections more than you expect by chance between these subsets of genes these modules you gain statistical power, so we look for pathway pathway connections and Only look for pairing with the module. Yeah, just the two modules. We say how often are the genes? Yeah So here's an example. Here's some, you know, here's some pathway in a human cell a bunch of different genes involved in it and Here's another signaling pathway in a human cell again a bunch of different genes. Okay, and then we look for connections that co-occur that are associated either with the Control cohort or the disease cohort more than you'd expect by chance and in this particular case we're linking You know this secretion pathway, which is perfectly reasonable for something like Parkinson's and neurodegenerative disease We only expect 1500 interactions, but we see on the on on the order of 2200 2300 interactions So that's the data that We're starting again and It's not, you know, you have to validate it. No one believes it It's it's basically a nightmare, but from our from our fantasy world, we're uncovering the genetic wiring diagram of humans using human human you know genome sequences and And so, you know, maybe this is true, right? But we have to we have to and but no one will give us money to do it So we could be on the edge of uncovering something awesome about human genetics that nobody knows and Nobody will give us any money to do that The idea is that okay say you have Parkinson's yeah, I mean I hope I can Know what am I doing here? Oh? No, I'm going for it. Okay sick. Yeah so So this is based on sequences. Yeah, this well, this is GWAS data, so it's it's genotyping data It's not but it could be whole genome sequence. You know all you have to do is map variation. Yeah, the GWAS is a little bit trickier because you don't know if you're right on the gene but there's there's Variation here and variation here, and so that's an interaction, right? and And you have to have null hypothesis models and Yeah, yeah, and then you look for co-variation that is associated with either the healthy Because Yeah, I mean we're connecting these pathways more expect by chance, and you try and go into two different independent disease cohorts, and if you see it replicate Maybe that's that's true Do you know that there are three studies on synthetic interaction and you using CRISPR in human cells Yeah, there's yeah. Yeah, so you could use them Yeah, I mean I don't know what what human cell is the best model for Parkinson's, but yes That's what we need to do we need to get into a system where we can find it in in Human genetic data and then validate it in some kind of model That's what we need to do to convince someone to give us money to do this, but Anyway, I want to get back to this modeling the human cell stuff, and so I've probably got ten minutes But you can take this and take the best genes and make a hierarchical Cluster and then we can look at the clusters at various levels in the hierarchy and If you go down to the the most refined sets of clusters, you can barely see them here They're so refined. These are all these complexes and pathways that are identified using genetics and protein-protein interactions if you go Up to here to these larger clusters Now these are these bioprocesses in the cell and then We'd always look at this and realize that you know all the secretion guys are Together and all the stuff in the nucleus together these larger clusters that you see in the hierarchy They're the compartments in the cell so You know the wild thing here is we've just measured single mutant fitness and double mutant fitness And then we draw this graph and it basically reveals complexes and pathways bioprocesses and and compartments Simply from its its layout So then once you have that model we like we like to view the cell or the the graph like this where you have these big Bioprocesses so we can say where do genetic interactions occur do they occur within a a complex or pathway between Complex or pathway in the same bioprocess Between those in in different bioprocesses, but within the same compartment or are they are they distant and We get different answers for negative and positive genetic interactions, so the negative interactions are often associated with Functional specificity and that they occur, you know in the same bioprocess or between process in the same compartment the positive interactions are often distant and In our SGA the positive interactions we we measure are not super strong. They're not Suppression in general. They're mostly regulatory things So a lot of them have to do with cell cycle regulation and proteostasis And the reason we're not looking at suppression is because we're just starting with loss of function mutations We're not looking for suppress special suppressor alleles like David Botstein and Peter Novink you do all these under one condition all the yeah Yeah, so we because it was like You know just one pass that took us 15 years. We only did one condition And so but dad now we're sampling on different conditions to try and see what the condition effect Will be and So so it's a good point But there's all these other types so everybody goes they now what are you gonna do now that you've done that and But there's a million things we can do Conditions as you just pointed out we can really attack suppression This is something that David's interested in and he did the first complex haploins sufficiency screen and We are doing that and it's a total nightmare. I'll have to tell you about it. Yeah, and so But and then there's this higher order interaction stuff and So I'm just gonna quickly go just so you can see it Yolanda Publishes paper with Fritz and I and Brenda where she did a literature curation of all the yeast suppression interactions and then we also mapped Several hundred new suppression interactions that were just made When we were when we're you know doing all this other stuff and From that she can put together a network and show you how powerful it is because now you have direction like mutation here is suppressed by a mutation over there and She just got a job in Los Annes. We're very proud of her and Suppression guys though are real severe versions of the positives that we score by SGA and The point is that You know, you've got a sick individual But if you some people walk around and they carry severe disease mutations, but they're perfectly healthy Okay, and so we want everybody wants to figure this out And and the hope is that when you figure out the genetic mechanism that you can develop a drug and There's this cool paper by see friend where they actually looked at this This situation they can find all kinds of people walking around that carry severe genetic disorders Okay, so This all makes sense suppression Suppression interactions are more functionally related than the stuff we score by SGA There's no point in going through all this stuff Okay, here's the one we should we should go through Okay, so once we have We have we have a system now where we can clone temperature sense of alleles of all the essential genes onto a plasmid We can put them into a diploid that's deleted for the gene that's on the plasmid We can convert that to a haploid where the the essential genes deleted But it's the cells are alive because they have the TS allele on a plasmid We can shift this strain to the restrictive temperature Find an extra genic suppressor and then we can ask can we kick out the plasmid? so can the suppressor actually bypass the deletion allele and how many essential Yeast genes and yeast can we delete and and actually bypass with a suppressor and So this I like this one because I Could bet My colleague Fritz Roth Yeah, head-to-head teta teta mwah versus Dr. Fritz Ross, right? Yeah, head-to-head So what if I bet fritz and fritz is one of these guys that if you just lay a number out there he'll go I don't think so, right? You know, he's a contrarian So I laid out a number 15% and he goes no way no way. All right. I Think I bet him a bottle of wine. Yeah. Yeah, it wasn't Because you know when someone does that you start to second-guess yourself And So Yolanda did this and as a collaborative project with Fritz so she We've got a you know a thousand TS wheels covering, you know a lot of the essential genes She gets bunch of suppressors. So and we're sequencing all these things now and in the end She could bypass a hundred and twenty four genes and And so guess who wins the bet? Yeah, so we're talking right right and so that was fun and the reason I made that bet was because the essential gene set is the same in Servicier as as it is in Pombi almost identical even though they are hundreds of millions of years divergent and Except for 15% so there's 15% of the essential genes are different and And we can see that that in fact It's it's that set that is non-essential in in Pombi that Can be suppressed largely in in service. Yeah, so So it kind of all fits All right, so the last thing I'm just going to say is Elena has been looking at trigenic interactions And it's the same game But we took quantitative features of the Digenic network to pick Double mutant queries, so they either had high high degree or low degree They either had very similar profiles or weak profiles, which is a measure of functional similarity They either the queries either showed a negative genetic interaction a low one or a high one again the genes in the same pathway If it's an essential pathway often have a strong genetic interaction Anyway, so she can map trigenic interactions and they have the same properties as Digenic interactions they occur amongst genes in the same bioprocess and I'll show you an example of one So here's There's a trigenic interaction all three genes are mutant in the other cases here It's either a single or a double mutant and So and there's a little bit of action going on but the the trigenic is the one that falls apart and When we look at these interactions there They so the two query genes what we mutated are involved in secretion and here's the double mutant query It shows a bunch of secretion interactions or these black and blue guys, but it branches out into DNA Synthesis and repair and this is often what we see is that many of the interactions are connected to the function of the query genes but the trigenics are a little broader in in specificity and So the the bottom line is that the the ones that trigenics are enriched for Genes that have these particular properties on the Digenic network So they occur often at the level of complexes or pathways or within the same bioprocess They often Modify a previous Digenic interaction. So the trigenic network is built up upon the Digenic one But the the amazing thing is that Well, there's half a million Digenic interactions. There's a hundred million Trigenic interactions in a yeast cell that we can predict so So that when you get in the bioprocess there if they don't show a Digenic interaction You can start combining two or three things and you you cause enough havoc in that bioprocess that you you create this Unexpected phenotype and so we think that you know understanding the bioprocesses of human cells and then looking at variation You know in genes in the same bioprocess might allow us to start looking at inherited phenotypes and so with that I Will just stop and then maybe ask have one of one question or something Thank you How often is aneuploidia problem? I mean you see aneuploidia coming up as a way Yeah, yes a lot you see aneuploidia a lot in in suppression and There's of course, there's a fitness cost with aneuploidy cells don't don't like that But when you go from a really sick cell you can you can suppress it with aneuploidy How stable is the end? That's it. I don't think we've looked at that but by the time we get it and sequence it, you know, it should be pretty Pretty stable, but we've never tried to evolve beyond that But yeah, you see it definitely comes up all the time and we'll see we even have this one case where the genome Well, it's with the signal recognition particle So you can take all the components of signal recognition particle and you can delete them and bypass it So there's some other way to live without it and the way the cells do that is they duplicate the genome and Then they lose a chromosome one chromosome Yeah So is there a way to estimate how often you would miss? interactions Yeah, so They're like definitely with suppression because we look for suppressors more than once you can draw a graph and Figure out whether you're saturated or not We do have an estimate of the false positive false negative rate which is Surprisingly high and we bury that in the supplementary data, but it's around 30 percent for both of them 30 or 40 percent And I'm sure we could do a better job at it now if we looked at it more carefully But that's about how often we miss interactions So botstein likes to say that Leo specific suppressors would tend to be the specific ones. I'm wondering do you think in general the bypass Suppressors tend to be less specific for example I know in bacteria that if you have a secretory defect a mutation that will slow the growth of the cell will suppress That and you and you wouldn't you tend to argue that that's really Right that you've got a specific pathway There no there. I mean in most of the bypass guys are pretty specific and partly because I don't think this We're picking up weak suppressors. We've selected for pretty strong suppressors But yeah, I don't I'll dig it out. But yeah, Yolanda has done all that analysis and in general They're really they are very specific Yeah, I know it's just like that other thing we were talking about today on the train with the dots He doesn't believe me does he? So One side is that if your analysis is the growth rate and Let's go out of mode but outside to to put them in a pathway or in a compartment. It depends on the GO Yes, on the genetic Oh So the okay So when you when they they cluster together because they have a similar pattern of genetic interactions Then they're placed side-by-side on the on the map and then we can go all the yeast genes are annotated by the curators at SGD and So that's an independent Functional annotation and we can scan their annotations and see which ones are enriched in the clusters Yeah, and if they had used our data to create those functional annotations, it would be circular reasoning But they don't use our data So You said 15 years of experiments was one condition Yeah, you started to evaluate how things would regroup and the wire if you change conditions Then can you comment on it? so What we've done is we've taken You know, we have that those thousand genes that are in that that hierarchy that behave really well in our assay We now put those on an array and then we we cross in a bunch of You know genes from every bioprocess, so we make double mutants and then we change we do that under a bunch of different conditions salt You know different carbon sources, you know, all you know drugs in the plate and There's a few people who've done this out there Trey Idyker with DNA damage conditions and they've really emphasized that the network is is Changes quite a bit that it's they call it a flux network but The what we find is That the core network doesn't change at all if you're a a really solid Genetic interaction under standard laboratory conditions You know, you're a solid genetic interaction even if there's DNA damaging Agents in there. There's gonna be some genes that maybe aren't even active under under laboratory conditions that need need to be induced or something if you add a DNA damage damaging agent and Those probably will show different different genetic interactions, but they're not the majority So the core network just like the essential gene set is highly conserved from servici pombi to servici We think the core network is gonna be Solid under different conditions and then there'll be a few genes that are that are changing Negotiated one last quick question very short question What you showed is big biology big data Did you did you try to identify one for instance one single and to validate it because East is very interesting, but maybe it's you validate there are many Segment construction public know the name of thoughts in drosophila a mouse so then you could Check directly and validate up the yeah evolutionary. Oh I see. Yeah. I mean there's your speaking of human disease. Yeah. No, I mean Yeah, I know but yeah, you know in that database. There's all kinds of autophagy pathways signaling pathways Amino acid transport pathways. It's it's all in there But and we put that stuff in our papers because we have people that won't believe that we're biologists unless we Show them some real biology and then and they always try and tell us to take it out And we go no, we've already done two years of work on that. We're not taking it. Yeah, but yeah So it's it's definitely in the in the data For sure, okay