Healthcare’s Next Frontier: The Power to Extend Life-Full Transcript
Solving the riddle of the genetic code
Sue Siegel, CEO of GE Ventures, sits down with J. Craig Venter, PhD, one of the leading scientists of the 21st century, at the StartUp Health Festival to chat about the power of our “software of life,” the human genetic code.
Key takeaways from this episode of StartUp Health NOW can be found here.
[00:00] Intro Music
[00:43] Steven Krein: Please join me in welcoming Sue Siegel to the stage, who will then welcome Craig Venter. Sue?
[00:54] Sue Siegel: Thank you, Steve. And, I am going to introduce a man that needs no introduction, I suspect most of you know that. If you think back to ESTs. How many of you know what an EST is, actually? Different audience, Aha. So, Craig, way back when, when he was at the NIH, essentially found ESTs. Which are expressive sequence tags.
[1:18] Sue: And a lot of people said, “You cannot sequence the human genome using ESTs.” This was the first, sort of, incarnation of doing sequencing. And as you will find, a pattern in what Craig has done, is essentially proved pretty much everybody wrong.
[1:35] Sue: Ok? So, that was one of his first things. He then went off to found HGS. Human Genome Sciences. He was a cofounder there. He founded TIGR. Which, for those of you who don’t know, TIGR, different audience, this was the Institute for Genomic Research. And this was the first sequence of a prokaryote, so bugs, essentially was done by Craig.
[1:59] Sue: K? A lot of folks not know that. And then we went off and he said, Now, I’m gonna start Celera. He had this vision that in fact if you took a bunch of machines. You put them all together. And you put the power of company and resources behind it, you could sequence the human genome faster than anybody else could. And in fact, what did he do? You know that the human genome was supposed to be done by the year, sort of, 2002–2003? When was it done? 2000. K. So, he did it. And then people said, well, what’s, sort of, what is he going to do next? And remember, during that time it was also amassing all of the information you could possibly imagine. And created Celera and this model of informatics and data is going to be the currency of the future. After that, if that wasn’t enough, I mean, sequencing the human genome, I mean, come on. Right? Ok, so we do it now everyday for a little bit of time and it’s a lot of less money. Back then, man? It was really hard. So first one.
[3:01] Sue: And then he said, now I’m actually going to go and I am going to figure out, sort of the next thing. He created the JCVI Institute. And he said he was going to go comb the oceans. And see, and sample the oceans. To see what in fact might be found, cuz we really didn’t understand that unchartered territory. So, human genome, uncharted territory. Ocean, uncharted territory, in many ways. And he went and sampled, and he came back, and he sequenced millions and millions of organisms of which he discovered a number of them. When you start to think about that. And so, he did it again. And during that time he also created synthetic genomics. Are guys tired yet? Because this is like this all happened in like 20 years. He created synthetic genomics. And what did he do? He synthesized the first organism. K? And we’re going to explore that a little bit. So, again, Craig not having done enough, said well, you know, now, I’m actually going to bring together all the things I’ve been doing before, but I want to do something about translation of it too. And I’m going to create Human Longevity Institute. And that’s what he set out to do. And again, we’re gonna talk a little bit about that today. So with that, I’m going to welcome Craig on stage. So that we can actually talk a little bit about what he does in this spare time. Which I’m about to tell you. What he tells in his spare time, which he wouldn’t tell me, but I got it from my litle sources out there. His favorite TV show. “The World of Dumbest Criminals.” [laughter].
[4:50] Craig Venter: That’s not my favorite.
[4:51] Sue: Ok, his favorite is “Jay Leno’s Garage.” And, “Chasing Classic Cars,” and “American Pickers.” There’s a theme. You guys, right? As you might imagine, he’s got some nice cars. Alright. Craig. Here we go.
[5:06] Craig: I’m happy to sit over here and let you keep talking.
[5:09] Sue: [Laughs] No, no, enough of that. Now we’re going to get to some of the hard stuff. So, first of all, tell us about Human Longevity Institute. What are you doing, what are you thinking, where are you going?
[5:19] Craig: It’s Human Longevity, Inc. People confuse these things, cuz the Venter Institute’s a not for profit institute. Human Longevity is definitely not for profit. It’s not yet profitable, but it’s not intended to be a not for profit. So, the goal was when we finished the first genome, it was hard to interpret. In fact, that hasn’t changed much in 15 years. And we realize we needed really large numbers of genomes. And we needed phenotype information to go with them. So, we’ve set out to try and do a million genomes by 2020 and collect phenotype information on everybody who’s genome we’re sequencing.
[6:00] Craig: So, it’s a huge informatics challenge. It’s a computational challenge. And, we’re adding into it machine learning to try and put this data together. So, it’s mixing all these together. Just in our startup phase, we passed 20,000 human genomes all with phenotype information and using GE and some other equipment, we’re making some combination findings with the genome that I think are gonna blow people away.
[6:28] Sue: That’s awesome. And by the way, you can’t do it unless you use GE instruments. Just FYI. [laughs] Just kidding. In reality though, you talk about, and again, this is a different audience than what you typically speak to.
[6:41] Craig: Yep. They didn’t know what an EST was..
[6:43] Sue: Well, and so that’s what I’m going to say, I want to make sure that everybody understands the notion of genotype and phenotype. And explain what you’re trying to do as you bring together many different phenotypes, and not just genotype, and how important that is.
[6:59] Craig: So, the way discoveries have been made before most have you entered science. Was you had to get very large pedigrees of people with common ailments and get a little bit of information to try and map those genes. So, things like the Huntington’s gene took about 25 years and maybe 10 major research groups around the world to find. Now, having a phenotype information and really accurate genome information we can make the same discoveries on a single person. And the correlation between getting the estimate of a risk from the genome is totally different when you have a yes/no answer. Do they have this trait? Do they have this disease?
[7:45] Craig: So what it makes, that data builds on each other so it makes the predictions much more accurate and from mere statistics. For example, brca1 and brca2 aren’t actually clearly associated with the cause of breast and ovarian cancer. They’re markers that get us in the vicinity. And so, it’s only a 50/50 chance of getting breast ovarian cancer if you have changes in both brca1 and brca2. The thing that pushes it into meaningful territory is if you have a family history of breast and ovarian cancer where basically every woman in the family has breast ovarian cancer. That means there’s other genetic elements that are actually causing the cancers that brca1 and brca2 get us in the vicinity of knowing.
[8:34] Craig: So, that gene information coupled with the history is very predictive. The gene information on its own for brca1 and 2 is only 50/50. So, we have to improve the accuracy and get to the things that actually cause the disease verses are just kind of in the neighborhood.
[8:53] Sue: So, contextual information is going to be really important. And how you’re adding that into the equation. So, particularly for the digital health audience, how are you thinking about those kind of technologies coming in to become another phenotype as you gather all this data?
[9:08] Craig: No, in fact the digital world is a challenge and so in 1999 we had to build the third largest computer just for assembling the first genome. And that cost about 50 million dollars and was only one and a half teraflops.
[9:25] Sue: A teraflop is, 10 to the ..
[9:28] Craig: It’s $100.00 today. [laughs] If that puts it in context.
[9:33] Sue: Alright, got it.
[9:34] Craig: So, there’s been a slight change. And with cloud computing, we can use distributed computing and not have to worry about building a machine to do this. Although when you do a brain MRI image it generates about 3 gigs of data and a group of scientists, computational scientists at UCSD led by Anders Dale developed an algorithm that takes that three gigs of data and converts it into a single page, a table of volumes of different brain regions where they can detect the slightest changes and repeated MRI’s over about 3 months that can detect with your developing dementia or not.
[10:15] Craig: So, that information coupled with the genome is very powerful, but it takes 3 gigs of data then converts it into a few KB of data that then is easy to take into a machine learning algorithm with lots of other cohorts of it.
[10:30] Craig: So, it’s not just the sheer volume of data. It’s the intelligent reduction of the complexity so it’s usable for comparing to everything else.
[10:40] Sue: And, tell us a little bit about what you’re doing with regards to Franz and face recognition. Tell the story.
[10:46] Craig: So, Franz Och was hired out of Google into HLI. I’m sure you know what he developed. He developed Google Translate. So, if you haven’t used that I’m definitely in the wrong room. And, he used a unique machine learning approach to convert languages into each others. I convinced him the human genome was the ultimate translation problem. Moon shots are too nearsighted.
[11:15] Sue: Oh wow.
[11:17] Craig: You guys need to aim a little bit higher and further. Anybody can get to the moon these days.
[11:24] Sue: So, what’s yours so we can put it in position here?
[11:27] Craig: I think the next decade in planetary space, obviously colonizing Mars, is where people are headed. And, things like being able to send biology through the internet like my Institute in Synthetic Genomics developed, are ways to get things to and from distant planets. [11:46] Craig: But I think that the challenges we’ve been talking about in this space, you have to think well ahead of where the crowd is going.
[11:55] Sue: Facial recognition. Go to face recognition for a second.
[11:58] Craig: So, using machine learning it’s a way of doing what thousands or even tens of thousands of scientists themselves can never get to with looking at a small bit at a time. So the first challenge I gave to Franz and his group was to see if we could predict your photograph straight from your genome. Just A’s, C’s, G’s, and T’s. And we’ll be submitting a paper on this shortly.
[12:26] Sue: I was much prettier by the way. They did it and they came up with a picture much prettier than what I am, seriously.
[12:33] Craig: It predicts your age, just post puberty. You might wonder what your genome code is for, it’s for what you look like post puberty. And the algorithm does a little smoothing, so it makes you, it does make you look good. You’re ok on your own, but we all could use a little smoothing.
[12:50] Sue: Open mouth. Insert thigh. Very good, Craig. That was good. [laughs]
[12:56] Craig: And the algorithm also makes faces totally symmetrical right now which they’re not, but we’re very good at predicting your face. But if we were recording your voice just from these microphones we can accurately predict your sex, your age, and your height. So, there’s information contained in everything that’s human and it’s a matter of using new approaches to pull this out right from your genetic code.
[13:24] Craig: I was absolutely wrong. I said you would get your genome sequence once in life unless you had cancer. It turns out the exact age somebody is when they have a blood sample drawn is essential because we can now predict your age right from your genetic code. It changes throughout life. It becomes of measurement of aging. So using machine learning we compiled everything that was known about human traits and physical predictions and the machine learning out predicted those by a substantial margin.
[14:02] Craig: So, just knowing the components wasn’t sufficient. The other thing machine learning does, is what your body does. It uses information across the genome. It doesn’t just use one gene and one snip and a gene like we’ve been measuring. It uses this information in an integrated fashion and that’s key to understanding different diseases, complex diseases, understanding aging, etcetera.
[14:29] Sue: Got it. So, in 2010 you did an interview with Der Spiegel. And in that interview you talked about the medical value that had been derived from the genome in 2010 was close to 0. So we’re in 2016. Tell us a little about, are you changing your mind on that ? And then tell us about 2025.
[14:56] Craig: So, it’s moved more in the last 2 years then it’s moved in the previous 13 years. [15:02] Sue: Why. Ok, why.
[15:03] Craig: Well, for these same reasons. We had small data sets, nothing to compare to each other. My genome was the first one that was done on the internet, but there’s less than 1,000 actual genomes that have been completely sequenced that are out there.
[15:21] Sue: What about all that are out there? Because, I was going to say, what about the Manchester project in the UK? I would have thought..
[15:27] Craig: Well, so, nomenclature here is important. Genome has become a very loose word. Back when we were first doing genomes, it meant you actually sequenced every base pair and closed every gap. That’s not possible with the human genome. Sequencing covers only 85% of the genome. But there was a paper just published in Nature, said the sequence of 1,000 human genomes. But they were only done at an average of 7–8X coverage. So, that may sound impressive if you know nothing about mathematics of genomics, but means less than a ⅓ of the genome was covered. So, we sequence things to 30–40X at HLI and we have incredible accuracy, and I’ll show some of this tomorrow morning. Where we sequence every base pair on average of at least 10 times. And have a low error rate and a low false positive false negative rate.
[16:23] Craig: But, even with that, only 85% of the genome was covered. So, when somebody is saying they’re sequencing a genome, it’s important to say at what depth and what technology.
[16:35] Sue: And, you know, you’re now using this in a program called Nucleus. In your Nucleus health program, right? So, tell us a little bit about what that is.
[16:45] Craig: The Health Nucleus is our ultimate phenotyping center. So we have the latest GE 3T MRI machine and we have protocols that only we and GE research have. We have DEXA scanners, we’re adding CT and a second MRI. We have 4D echocardiograms, we have ways for measuring neurological functions, and phenotyping.
[17:19] Craig: We measure the human genome, the microbiome, we bleed people and take 18 tubes of blood and measure literally thousands of chemicals.
[17:27] Sue: You’re supposed to be selling this, Craig.
[17:31] Craig: [Laughs] We don’t do it from spit. So, it’s doesn’t quite work that way. And, we integrate this information. So, just the imaging is been absolutely stunning in terms of the discoveries.
[17:45] Craig: The 4D echo, we’ve discovered two people with aortic aneurysms that thought they were totally healthy. And we integrate this data. And you find a physical trait, and we can go into the genome and we found a gene duplication event associated with aortic aneurysms. We can find different types of brain tumors, and find the genes associated with them.
[18:07] Craig: So, it’s not hypothetical anymore. We can start with the genome and go the other way. And we predict these people might have these diseases, because they have a higher risk. And we scored. They do or they don’t.
[18:20] Craig: We can detect whether they have cancer. We can detect whether they have enlarged hearts or they have different types of defects in the brain. Whether they have dementia. So, it’s intense phenotyping along with an incredibly accurate genome that allows us to put the data together in a combination that’s gonna be, I think, really stunning when people see all the outcome.
[18:48] Sue: And, who actually meets with the consumer? Because it’s not a patient anymore. It’s really a consumer who’s choosing to do this, right? Who meets and actually discusses what you’re finding? Because, the information complexity is huge.
[19:02] Craig: So, we’ve developed a new approach. We actually make 3D avatars, so with the, I think there’s 400 cameras in the system, once you make an avatar, you guys have all seen the movies, you can digitally make them do whatever you want.
[19:19] Craig: When I’m not there, they make mine do a funny dance of some sort. And your avatar will walk you through all this data. So, we start with your avitar. We start with the actual predictions of what you look like. It walks you through the MRI data all the way down to your skeleton with the DEXA down into your genome and goes through this data. But the data goes through and is presented with physicians and genetic counselors.
[19:47] Craig: It’s too much complex information. And, we only have healthy people come in.
[19:54] Sue: Or so they think.
[19:56] Craig: But they don’t leave healthy. So, either we’re doing something really unique to them while they’re there. But that’s just that the statistics, this is kind of a younger group, but what if you’re between 50 and 75 and you’re a male. You have a 30% chance of dying in that time period. And a third of that is from cancer. And another third is from heart disease. People just don’t know they have heart disease or cancer until they get to the symptomatic stage.
[20:24] Craig: If you’re female, it goes down to a 20% chance of dying. But still a third is from cancer and a third is from heart disease. 1.3 million people get diagnosed with cancer each year. Those cancers just didn’t’ appear a few minutes before they were diagnosed. So, a third of the population’s walking around with cancer, with heart disease, with pre dementia symptoms that they’re unaware of.
[20:50] Craig: The difference is when we find them early, they’re treatable. Or preventable, or switched into a combination of the two. The problem we have with this comprehensive cancer work we’re doing, we get individuals when they’re discovered and diagnosed with cancer at stage four, where the odds of successful treatment are way down, vs what we have. So, I”ll talk tomorrow about one case. A tumor discovered under the breastbone that, you know, showed up with the MRI imaging was removed. It hadn’t penetrated the tissue and the guy went home two days later. Completely healthy, cancer free, and risk free.
[21:36] Craig: Had he not had this test, probably in a few years it would have penetrated the tissue and the survival time goes down to about a year.
[21:45] Sue: Wow. So, medical value of what..
[21:48] Craig: Medical value, but predictive value is what we’re building out of this data base. So, in the future as you’re get your genome sequence, these things will be predictable right from the genetic code, the way the same, that your face and these other physical traits are.
[22:03] Sue: But aren’t you a little worried about what that’s going to do between CRISPR/Cas9 and what you’re able to do in terms of what we might be able to plan as it relates to the populations we want? How do you think about that?
[22:18] Craig: I think right now, and I have written about this, I think that editing the human genome is inevitable, but I think we should wait.
[22:28] Sue: Say that again?
[22:29] Craig: It’s inevitable that we will do it, but I think we need to wait till our knowledge base is a lot better. And I liken back to discoveries that were made early on in Drosophila. There was a early on, a developmental gene that helped form the limbs and tissues, that in the late stage, that same protein became a structural protein in the wings of the fruit fly. We think we know a function for a gene and if you go modifying that you don’t know what functions you’re changing. So, we can’t afford to do it with human experimentation without having clear cut knowledge. That being said there are some diseases, maybe like sickle cell anemia, or ataxia-telangiectasia, that are reasonable and ethical to change.
[23:16] Sue: And you’d say, go ahead.
[23:18] Craig: I would to it if it was my family.
[23:20] Sue: So, it’s single cell, excuse me, single gene, well understood.
[23:23] Craig: And devastating diseases, where the outcome is clearly understood. The challenge for everybody, as we start to predict traits, and right now with circulating fetal cells in the maternal bloodstream, we can sequence the genome and we can give the family’s a picture of their hypothetical child at age 18.
[23:51] Sue: And they might not like what they look like? [laughs]
[23:52] Craig: They might not like what they look like, they might find it was the wrong father, you know, there’s a lot of things people can find. [laughs]
[24:01] Craig: You know, there’s surprises that come up later, you know, when you see them in advance. it might, might cause some increased legal action.
[24:11] Craig: But it’s a challenge as we learn how to read the genetic code. It is our software of life. We’re learning how to read the software. At the Venter Institute, Synthetic Genomics, they’re learning how to write the software with CRISPRs. Everybody can edit the software but were very early on in our knowledge curve. You know, 10 years ago we knew virtually nothing, we’re just starting to get there. We’re at the earliest stages with cancer, but we’re learning rapidly at least some ways to alter the outcomes.
[24:48] Sue: So, you synthesized a genome. Excuse me, you synthesized an organism. And this was, two years ago? Three years ago?
[24:54] Craig: Five years ago.
[24:55] Sue: Five years ago.
[24:57] Craig: Time flies.
[24:58] Sue: So what. Now what’s happened. Ok? What, what have you done since then, and what should we be thinking about as you’ve continued to progress that.
[25:05] Craig: So, what the Venter Institute, the not for profit institute and another for profit company, the Synthetic Genomics that’s presenting at this conference as well, over the last five years.. So, the first one we did was sort of basically copying most of a known species genome. And seeing if we could actually reproduce it from four bottles of chemicals. And we were able to, with some key alterations. Then we set out to design a new species for the first time. And we thought it would be a lot, pretty straightforward to do this on first principles. And first principles of biology dont get you to a living cell. So the reason it’s taken us five years to do is there are a lot of iterations of things, testing them to see if they worked, and didn’t work, and sort of a story that goes along with this, when I was on a book tour of my book about synthetic life, life at the speed of light, I was in Seattle and my late uncle was one of the lead designers for the Boeing 757. And, and I said, imagine building a new aircraft without knowing what all the parts do. And he shouted from the back of the room, “What makes you think we knew?” [laughs]
[26:20] Craig: So, designing life is kinda like building aircraft. If they fall out of the sky they find out what’s wrong and they fix it on the next generation. We kept testing until we found what was needed and so that paper hopefully will be out soon describing the first species designed through this combination of actual design in iterative testing to get something that would work.
[26:45] Sue: Wow. And when you say that would work. So it has functional life?
[26:50] Craig: Well, it lives and self replicates. It’s not extremely sophisticated but..
[26:54] Sue: It lives and it has sex. [crosstalk] That’s kinda cool, right? Just over and over again. That’s kinda cool.
[26:59] Craig: Bacteria don’t have sex.
[27:01] Sue: Oh, bacteria don’t have sex sorry. Ok. [laughs]. Alright. So, in all frankness, I mean, how do you measure that in fact it’s functional. What’s, what was proving that in fact, it’s real.
[27:14] Craig: It can make billions of copies of itself.
[27:18] Sue: And they’re all identical?
[27:19] Craig: If you could do that people would be very impressed. [laughs]
[27:21] Sue: Ok. So, when is this coming out?
[27:25] Craig: Hopefully in a month or so.
[27:28] Sue: And, how
[27:29] Craig: It’s just under the final review stages in a scientific journal.
[27:32] Sue: And, how complex is this organism.
[27:35] Craig: It is, it has a smallest genome of any self replicating organism ever known. We tried to get to what a minimal set would be. So, yet, it’s software’s still immensely complex. It has all these regulatory functions. It has what people consider epigenetics. And the point is, all of that is coded for, in the linear code of the DNA.
[28:05] Craig: And I think that’s a surprise to a lot of people. Because you can get a separate class on epigenetics, you think it’s a separate phenomenon. All that is coded for, in that linear basic code that we all have as our software. So, it doesn’t matter where this minimal cell, or our human cells, that’s my understanding of the software is so critical.
[28:26] Sue: So, if that took five years. And you think about exponential technologies and the speed of all the convergence of the technologies are coming together. What’s your martian shot against that?
[28:42] Craig: I guess the right thing would be creating life in 7 days. [Laughs].
[28:51] Sue: I did ask for that, didn’t I? I mean, seriously. Cuz it’s just mind boggling, when you think about what you actually can do.
[28:57] Craig: Or 6 days to improve upon history. [laughs]
[29:01] Sue: So there’s no rest. There’s absolutely no rest. So, all right. We’re gonna wind down now. And I’m going to shift gears and just really ask you this. You got a room full of entrepreneurs and folks that want to create the new. And as you think about all the things that everybody said you can’t do, he’ll never do it, and then you just did.
[29:21] Sue: What advice would you give to some of the entrepreneur’s in here? What would you say in terms of how they go about thinking of their own futures and what they should go do?
[29:31] Craig: I think it’s essential to have a belief in what you’re doing. And it can’t be a delusionary belief.
[29:40] Sue: But people thought you were delusionary.
[29:42] Craig: Well, and that’s when..
[29:44] Sue: I mean that’s reality. It’s like, what? He’s gonna gonna create synthetic, I mean, what?
[29:47] Craig: And had we failed, that’s how it would have gone down in history. So, that’s how you build confidence in your own predictions, in your own intuition, by actually doing things. If you’re right..
[30:01] Craig: It’s like making movies. You make a blockbuster movie, somebody will give you money to make the next movie. After you’ve made 10 blockbuster movies you can have a few failures and somebody will still give you money to make another one. But if you never have a success you can’t really be a successful entrepreneur.
[30:22] Sue: But some of the entrepreneurs are starting from scratch. This is their first time.
[30:26] Craig: You have to pick your battles. You have to really be passionate about it. You have to believe in what you’re trying to do. And you have to pick the right thing.
[30:37] Craig: It’s, very easy in science to pick questions to work on. Picking the right questions that can be answered now with a certain technology is the magic of making success in science. It’s that easy. You could pick thousands and thousands of questions..
[30:55] Sue: Somebody write that down. Did somebody write that down? Because that was good. What did you say? Say it again? [Laughs]
 Craig: You have to pick the right question to answer. And if your timing is off, you can be too early or too late. Right now I think our timing was perfect on starting Human Longevity. 5 or 10 years from now it’d definitely be too late. And if we did it five years earlier it would have been way too early.
[31:22] Craig: So, I set a very clear criteria of what kind of computational power there had to be in the world to do what we’re doing. And what kind of sequencing technology. And the Luminex 10 just got to the threshold.
[31:36] Craig: We can improve on it tremendously, but it set a threshold of what it would take to get high throughput, highly accurate sequence, but the compute side had to be there as well and new approaches like machine learning had to be there. None of those existed in a realistic way five years ago. So it was a narrow threshold where this could just start to happen.
[32:00] Sue: With that Craig, thank you for all of your pioneering work for advancing the sciences like you’ve done. Thanks so much.