Zymergen: Unlocking the Power of Biology through Technology - Zach Serber at impact.tech at GoogleSF

You've covered, and I think you all probably know that microbes occupy every ecological niche in the environment. We have microbes everywhere, all around us. As a way of calibrating myself for this talk, I'd really appreciate if people who were involved in some way or trained in the life sciences would raise their hand. Okay. And how about engineering and computer scientists? Okay, terrific. All right, that helps, that helps orient me. So I want this talk to appeal to everyone watching at home or in the audience, and I'm going to begin by explaining why my passion for replacing petroleum has anything to do with microbes in the first place.

We heard that they themselves and occasionally produce other products as well. It's not necessarily commonly known, but there's already a huge sector of the economy based upon this feature of microbes. One you know well: many of you are drinking it now. It's ethanol produced by fermentation, and ethanol is definitely the most popular and well-known fermentation product made via microbes. What you may not know is it's just scratching the surface. There are literally hundreds of additional products made via fermentation in volumes that are staggering, down to tiny, tiny volumes depending upon the niche that that product is filling in the marketplace.

Almost all the amino acids produced by the microbes that are made industrially, used industrially. This is an important distinction between the past and the future. So in the past, we have built an 80 billion dollar industry around fermentation products, almost entirely based on what microbes natively produced. The way this would work is someone would find a microbe in nature that natively, just naturally, happened to secrete a little bit of some amino acid, for example, let's say glutamate. And they would find, 'Oh, that's curious. Glutamate has a market.' In fact, monosodium glutamate, MSG, is a huge product. It's a 4 billion dollar a year product. Most of it's consumed in Asia, and in fact, one of the first really big initiatives around this was to use mutagens, carcinogens, things that naturally and in ordinary circumstances caustically modify DNA randomly. So they would expose the microbe to something that caused in us would cause cancer, but in them would just cause some mutations and sometimes kill them. And every once in a rare while, they would find a descendant that would produce more glutamate. It would be the sort of equivalent in humans of exposing us to toxic waste, and most of us would get leukemia, but occasionally Wolverine would emerge, some amazing mutant with superpowers. Very rare event, but if you know how to find it, you can capitalize upon it. And then you repeat the process, and this would go on for years.

However, the people who did this don't really know what they did. They know they've incurred some mutations in DNA, and DNA sequencing has come along so far so fast that it's also possible to look at the genomes of these two microbes, the one you began with and the one you have now 40 years later, and see what the differences are. And there are thousands of differences, and ascribing cause and effect, which of those changes are actually responsible for the gains you have, is an unsolved problem. That is how things have been done in the past, and it's a good preview because it highlights how little we actually know about biology, but yet we've still been able to build an economy based upon it. Going towards the future, and even today, there are a number of products that have been made through this approach.

There's some common examples of this. In the Bay Area, there in fact was a huge initiative that began around 10 years ago to make biofuels. If you recall, in the summer of 2008, a barrel of oil was $150, and people were talking about $300 a barrel oil. The price of the feedstocks that commonly used in these processes were also at all-time lows, and the economics under those circumstances worked out that if you could have a microbe that for example produced a hydrocarbon, you could make transportation fuels. And a number of companies were founded around this premise, including one I used to work for called Amyris. Unfortunately for those companies, the economics, the macroeconomics of the situation have flipped. A barrel of oil is now closer to $100, and the feedstocks have gone up in price.

A difference moved into the field with the intention of leveraging their skills and knowledge for societal good, for turning these chemical factories into things that could replace petroleum. And I was one. I maybe I wasn't talented or any of those other things, but I was certainly motivated to go participate in this initiative. And my motivations are no different today than they were back then, but the means must change. Transportation fuel seemed like the low-hanging fruit back then. If you look at a barrel of oil, roughly 85% of it goes into energy, the remaining 15% goes into materials. The carpet on the floor, the ceiling tiles in this room are all petroleum derived. The plastic on the chairs we're sitting on. Our modern world is based upon petroleum products.

About the same, which means there's a disproportionate impact economically of the materials component of that barrel. These materials tend to be higher value products. Oil made into energy sources, heating fuel or transportation fuel, you're burning it. I mean, this really has to be the cheapest of the cheap stuff. But you can, the economics of the whole value proposition of pumping oil out of the ground, if you have some impact on the material science aspects, can we make new materials from biology rather than from petroleum that come over time to displace the petroleum source? Even this is hard. I don't mean to diminish the challenge. One of the major challenges is the oil industry has a 100-year head start. They have $3 trillion dollars of depreciated assets.

Our angles on this problem is to pursue targets, pursue materials that are advantaged over the status quo in some way. A lot of people talk about drop-in replacements. Here's a concrete example: baby diapers. The absorbent material in that is polyacrylic acid. Polyacrylic acid is made from petroleum. Polyacrylic acid is an 11 billion dollar a year industry, 70% of which goes into the absorbent material in baby diapers. Acrylic acid is also a biological product. It should be possible to make a microbe that makes acrylic acid at sufficient quantities to rival that of the petroleum source. And indeed, companies have been founded on this very principle.

All the companies that I'm aware of that have tried this have found it very, very difficult. So what's your angle? Your angle has to be, in my estimation at least, finding opportunities that go after new materials with advantage properties that are going to take market share by virtue not of their cheapness at first, at least, but because they're better in some dimension. Why is this a good idea? Well, in part it's because the materials that you can make from biology are fundamentally distinct from those that are natively made easily from petroleum. So petroleum is really good at certain kinds of chemistries as chemistry inputs, but biology is better at different other sorts. In particular, and this is for the technical among you, but biology is really good at introducing chiral centers, and this matters for a lot of different kinds of materials.

Works which is a joint venture with Cargill and is gaining market share rapidly on the strength of its features, its properties absolutely depend on the chiral nature of the lactic acid that goes into its manufacture. Petroleum can be used as a source of lactic acid, but most easily makes an equal molar ratio of the L and D types, which when combined make for a very inferior polymer. So biology has a clear manufacturing advantage over the status quo in that instance, and there are many other potential examples like this. Biology is also really good at decorating backbones for polymers with different chemical reactive groups of different sorts. You can put an amine and a carboxylic acid with equal ease on the same backbone, whereas if you were to do that with petroleum chemistry, it would be very difficult.

I joked about the DARPA work we're doing, but with partially the support of DARPA, we are prototyping 360 novel products that can be made via biology, with an intent of being disruptive in a variety of material science applications. And DARPA is interested in this because they are interested in new materials as well for a variety of applications. We are not working on fuels, and we are focusing on a variety of different kinds of materials in different classes. It is a four and a half year long play, and we've just begun. In the meantime, we've directed our energies at solving some of the fundamental problems with biology. So this all sounds pretty rosy. Why isn't this future here now?

If you were to visit our facility and that of some other startups in the Bay Area, you'd find them filled with liquid handling robots doing the work that used to be done by humans by hand with pipettes. When I was a graduate student at UCSF down the street from here some time ago, there were no robots. Everyone did everything by hand, and you had all these incredibly well-trained people doing what has now become increasingly roboticized work. And that's all for the good, because people can spend more time thinking about the problems rather than executing. But it turns out that that's just one way in which the modern world can make biology easier. The other problem has been one of hypothesis testing and incremental knowledge improvement. So there's an old joke that a policeman finds a drunk man looking for his keys under a streetlight. The drunk man says, 'This is where the light is.' This is a metaphor for where we find ourselves in biology. The light is shone on certain features of biology. We do know something about some things in biology, but if we only rely on what we know to incrementally expand the pool of light, our opportunities for making use of biology are necessarily limited. It's a very slow, risky, and laborious process to incrementally expand our knowledge base. So at Zymergen, using the robots we have, we combined that approach with a philosophy that treats genomes as optimization problems. The genomes are incredibly complex. These are the collection of genes inside the cell doing all those chemistries, and their interactions, and even the functions of all those genes.

We don't know what all the genes are or where they all are. So if you're faced with this profound ignorance, how can you possibly have an impact on the system? Well, by owning it, by giving yourself up to ignorance. And what would you do if you didn't have a bio background? I'm a physicist turned biologist, and that probably helped me a little bit in this philosophy. But fundamentally, what we do is we perturb the genome broadly and comprehensively, looking for perturbations that affect the performance of the cell in a way that is meaningful to us, in some way that's commercially meaningful. And we may not now or ever understand the mechanism of action by which that perturbation has an effect, but we can exploit it. And because these are targeted engineering steps, we can connect cause and effect from a given perturbation to a given outcome.

The use of our platform for solving this problem, how do we search the genome to optimize it for context? This problem is quite a bit harder than any that I'm aware of in terms of the possibilities that need to be explored. It said that the learning algorithm, the AI Go program that beat the Go masters, was considering 10 to the 320th possibilities. I mean, this is a massive search space. Well, the search space of biology is much, much larger than that. For a single enzyme in biology, it's often assumed that they're about 300 amino acids long, and each amino acid's position can be one of 20, so that's 20 to the 300th just for a single enzyme.

We need the right kinds of perturbations to evoke the changes that you're after. And we're going about this in a data-driven way. We are generating different kinds of perturbation classes and comparing them one to another in an A/B style fashion to see what kinds of perturbations affect the change we care about most readily. And we learn from that and deploy again in the future. And in this fashion, we have built an engine that allows us to optimize biology without necessarily having mechanistic understanding, which has been the crux of progress in our field to date. So how do you make progress in the absence of mechanistic understanding? It's by treating it as a search algorithm. And we have built the data infrastructure to suck up data from thousands of experiments that are performed by liquid handling systems. And we have, in the last year, generated six million data points all ready for analysis.

The consequences of solving this problem are profound. I mentioned before that biology of this sort is an 80 billion dollar a year market and affects our lives in ways big and small every day. Solving this problem around optimizing biology arbitrarily opens up the possibilities tremendously. And biology will become the petroleum of the future. It will become the source of all chemistries, both for medicine and for materials and for agriculture that we rely on. We will have superior materials, superior medicines, superior agricultural outcomes, all without reliance on petroleum. And it's a matter of time, accelerated through this kind of approach. So I thank you for your attention.