00:06:41 · Mark Jacobstein, Helmy Eltoukhy
[00:06:41] S1: Although also an entrepreneur. [00:06:43] S0: Yeah, so he also left that after a year and started a couple of companies in the valley here. So, I said, Oh yeah, that's a good pathway. Especially I looked at all the professors at Stanford, they had one foot in the door and a bunch of startups. It seemed like a really good place and I looked at it for a while and I just said this is tough. It's like all of the challenges of a startup. You have to raise money, you have to get a bunch of students with the bureaucracy of a 20,000 person organization as well. I said, yes, maybe that's not for me. So I said, why don't I just join a large company? I'll learn kind of the ropes and so on. And then, or I maybe just start a company. And then I thought about it for a while and I sort of had this Pascalian kind of wager, which is like, I said, well, I'll start a company, and even if it fails, I can go join a large company later, but at least I have CEO on my resume, and I'll probably get a higher place in that company than I would have a year earlier. So it was in some ways I think we overestimate risk for some decisions. I mean, I think that's the thing that
00:07:49 · Mark Jacobstein
[00:07:49] S1: I actually think that's very true, especially in the Silicon Valley ecosystem. The way I think about it in terms of doing a startup is you work for one company at Silicon Valley. You just switch divisions occasionally, and sometimes the division gets shut down. But if you're good at your work and you build your reputation in the ecosystem, then that was one failed experiment. So you come out of the postdoc, you go start the company. And one of the things that's it's a very interesting transition. So if Avantome was like a core technology that helped enable next gen sequencing, we should talk about that. But it's very different than Guardant, which Guardant is an application of the core technology that was developed by of Avantome and by Illumina. And one of the things that I think is unusual about your journey is I feel like most entrepreneurs play in one layer. Like there's people who make infrastructure. They usually tend to be so closer to the metal, more science y types, and then there's people who make applications, they tend to be more commercially minded or product people. It's very rare for, I actually don't know many other entrepreneurs who sort of played between the different parts of the stack, and Maybe walk us through that and like what motivated that? We just talked about this a little bit in the other room, but I think it would be great to hear.
00:08:59 · Helmy Eltoukhy, Mark Jacobstein
[00:08:59] S0: Yeah, I think part of it is like what problem are you trying to solve? And I think I was fortunate that the PhD I had was sort of one foot in multiple doors, in two doors, so the one was the circuit design side of things. The other side was information theory signal processing is very sort of underrated because it really helps you break down problems, think about like what's core, what are the core variables, what are the core elements of this problem, distill them into sort of those core fundamentals and try to come up with a solution. And that's essentially what I think you can do with almost any field. There's jargon there, there's things you have to learn, you have to understand like the ins and outs. But once you do, you can sort of break those things down and make a lot of progress there and come up with solutions. So some of it was, you know, if you think about I started in like straight electrical engineering doing like device physics and so on and there was one core set of problems there in terms of like how do you make the frequencies faster or this or that and then translating it into photonics and basically detecting very low amounts of light from DNA sequencing reactions and really understanding the physics of that, what is the problem with that, and then taking that at Illumina and scaling that down. [00:10:28] S1: Was that sort of the core, I know of Antone was like CMOS is applied to very early versions of NGS, but I don't know much more than that. Was that sort of the core thing?
00:10:39 · Helmy Eltoukhy
[00:10:39] S0: Yeah, core thing there was a technology called four fifty four that was the first next gen sequencing technology. That technology used the same camera that's in the Hubble telescope. It was a $100,000 camera. So, the system was a $500,000 machine. We realized that essentially something like 90% of the light was being lost in the optics. When you think about a camera, it has a big lens, the actual source is far away. So the amount of light that's actually coming into the pixels there is very, very small. So we said, you know, if you take a CMOS sensor and you put the light emitting reactions directly on each pixel, you could miniaturize everything, you could actually make the CMOS sensor disposable because you could make them for $10 But the problem was CMOS was not even good enough to be used as webcams back then. It was still a very CCD centric sort of ecosystem. And so we had to sort of improve the sensitivity of CMOS, photodiode and photo pixel probably by about a hundredfold in terms of what was there. And so that was part of the problem. So it was really systems engineering, thinking about the whole system, How do you get those 10x improvements in certain areas so that you can make the whole thing work? And it's the same thing with DNA sequencing with liquid biopsy. There was a 10x problem in terms of a lot of molecules were being lost when you take a tube of blood and you look at cell free DNA. At the time, only about 5% of the DNA was actually being used in the sort of sequencing reaction, and we improved that to about 70% of the DNA being used. So that gives you 10x, 14x improvement in terms of sensitivity. And then there was a noise issue in the back end. The sequencers at the time had about a 1% error rate. We were trying to take things down at the level of one in a thousand, maybe even a little bit lower, so we had to improve the error rate. So it's fundamentally, there's biology there, but these are engineering problems.
00:12:46 · Mark Jacobstein
[00:12:46] S1: Engineering problems, and information processing problems, and early bioinformatics problems. So maybe let's jump ahead then to Guardant and put the company together in 2012. One of the things that I think is really interesting, I was lucky enough to join Guardant when in the early days when really the the only thing that Guardant was selling was an assay for therapy selection to get tumors characterized genomically and get people onto the right therapies, which was the business for a long time and still is. But the North Star was always kind of early detection. Was that true when you and Murali started the firm in 2012? Early detection was the thing?
00:13:23 · Helmy Eltoukhy
[00:13:23] S0: Yeah, it was always the vision. Like, our first pitches were essentially the ultimate goal is a simple blood test, annual physical for early detection of cancer, but we have this, I think, concept of intellectual humility and understanding that, look, people have been thinking about early detection of cancer for at least dozens, hundreds, you know, a hundred years and so on, and so it's probably going to be more difficult than we expect, so let's do it in a way that's iterative. Let's start with solving easier problems first and then scaling essentially the technology that allows us to solve that problem, or the commercial technology that allows us to run that as our essentially compass for developing the next technology and the next one on our march towards early detection. It's similar to like, you think about Tesla starting with the Model S or the Roadster and then moving to the X and then the Model three, there's a technology scaled, you can expand it to wider and wider sort of use cases. So, we really had that sort of I I think that's the benefit of coming from the semiconductor field is that we were ingrained in Moore's Law and we understood that when you have scaling laws and you improve things, you can essentially widen the aperture in terms of how those technologies are used and the use cases that they're applied to. And so, we always said, okay, the first technology will have a sensitivity of X. It can be used to detect stage three, stage four cancer in existing cancer patients, mashing them with better therapy. We improve it another 10X. We can help fifteen million cancer survivors with residual disease detection, we improve it another 10x, then we can do stage one disease and do early detection of cancer. And so that was always part of the business plan. It's pretty