Bringing Zymergen to Scale: Enabling Engineering Biology with AI & Robotics

We’ve adapted nature’s ingenuity, selecting strains of microbes that turn their food (some organic material, from grape juice to woodpulp) into chemicals of use to humans. Conducted at scale, and in vessels from barrels to giant steel vats, this process is called industrial fermen­ta­tion. Without fermen­ta­tion, we wouldn’t have beer and wine to drink, cheese and chocolate to indulge in, and insulin, vaccines and antibiotics that help keep us healthy. Many of the discoveries were made due to observing nature, which is a relatively slow process of trial and error.

Even in labs, the field of biology has long relied on extremely labor-intensive and time-consuming experiments. Until only very recently, biologists were reduced to cross­breeding yeast, like medieval brewers, to learn more about key genetic traits. However, in the last thirty years, a revolution in the ease, accuracy, and cost of genetic sequencing has produced a wealth of information about microbial genomes. Coupled with a parallel revolution in techniques for precisely altering those genomes, it is now possible to conduct biology experiments with vastly more precise under­standing of initial conditions and expected outcomes. This twin revolution moves us closer to making biology more like engineering.

Engineering biology doesn’t mean simplicity, but it does mean repeata­bility and precision in a data-driven self-optimizing feedback loop. Today, the semi­con­ductor industry’s path from an idea to functional chip requires the following:

• Precise design: incredibly complex and capable chip design software, which considers all the possible permu­ta­tions of a circuit design and whether it will work in a fully embodied chip
• Repeatable and precise manu­fac­turing: trans­forming precisely polished wafers in a clean room with hundreds of steps of etching and layering circuits on the wafer, as well as many more steps in testing and packaging
• Obsessive instru­men­ta­tion: collecting and measuring data from every step of every process, analyzing, detecting errors ideally before they happen, and learning to fix mistakes if they do

The semi­con­ductor industry also benefits from the ability to rapidly test ideas. Many variations of potential chip designs are placed on a single silicon wafer in a so-called ​“shuttle run,” where one or a few wafers are exper­i­mental candidates in an otherwise production manu­fac­turing process. Conversely, proven designs can be run relatively quickly in an unproven manu­fac­turing process to learn what to fix about a new process.

The semi­con­ductor industry learns from its tests, both of individual chip designs and of overall manu­fac­turing process itself, and collects tremendous amounts of fine-grained data. The industry then uses these huge pools of proprietary data to rapidly iterate improve­ments to both individual chip designs and manu­fac­turing process, as well as the design software that drove the chip designs to begin with. This is the data-driven self-optimizing feedback loop that has supported much of the pace of Moore’s Law for almost 50 years.

The companies that mastered this ​“deep stack” (as DCVC calls it) of high-complexity software running on high-performance compute, cutting-edge hardware, and industry specific knowhow in the semi­con­ductor business became multi-decade sustainable giants, like Intel, ARM, Qualcomm, and TSMC.

Zymergen’s Deep Stack: Engineering Biology like Semiconductors

We believe that Zymergen’s break­through is in its own ​“deep stack” that enables engineering biology akin to the semi­con­ductor industry’s process.

First, precise design: Zymergen’s AI-powered genome design software explores a search space orders of magnitude more complex than Google’s DeepMind AlphaGo. This search is to find exactly the genome of exactly the right organism that will produce the desired results. These include the material produced (e.g, ballistic nylon versus hypoal­ler­genic adhesive), the energy efficiency/productivity (which correlates to profit margins), byproducts produced or lack thereof (better), and even subtle tweaks like inter-organism stickiness (less viscosity means less energy to stir a fermentation tank).

Second, repeatable and precise manu­fac­turing: in 2015 alone, Zymergen’s automation systems supported tens of millions of discrete operations in 2015 alone — the equivalent of roughly 10,000 scientists advancing a delicate experiment one step every hour, 24×7, for a year, repro­ducibly and at extremely low error rates. Almost every operation at Zymergen to engineer and test an organism is automated and de-risked from human error: robots pipetting, macro and micro-fluidic flows under precise computer control, flasks heated and cooled and shaken with tireless machine fidelity.

Third, obsessive instru­men­ta­tion: every single step of Zymergen’s manu­fac­turing and testing process is heavily instru­mented, collecting fine-grained data in real time. This data is fed into manu­fac­turing machine learning systems that self-optimize to avoid errors and work better, faster, with less energy and fewer steps — saving material amounts of time and money. This data is also fed into Zymergen’s core AI organism design platform.

It is important to note that Zymergen is optimizing its platforms and manu­fac­turing across multiple organisms at the same time. Typically, industry uses a few types of bacteria such as S. cerevisiae (yeast) and E. coli (bacteria), but Zymergen is able to use other microbe models in addition to adding fungi, gram-positive and gram-negative bacteria as potential hosts.

The result is a true data-driven self-optimizing feedback loop that improves the cost, speed, quality, and reliability of Zymergen’s design software, manu­fac­turing processes, and final engineered biology outcomes. With every single customer and every single process step for every customer, Zymergen has built and continues to build a sustainable, self-reinforcing advantage. This kind of lead is extremely hard to surpass, even with lots of money and time, because Zymergen’s advantage improves geomet­ri­cally over time, leaving any competitor starting from today or even a while back at a permanent disadvantage.

Zymergen has created a systematic approach to engineering biology, using AI, robotic and machine-learning driven manu­fac­turing, hyper-scale big data, and massively scale out compute. Fortune 500 companies now rely on Zymergen to deliver microbes that ensure that industrial-scale, high-value products are manu­fac­tured in a repeatable, precise, high quality, and low-error rate process.

Investing from the Beginning, Helping Throughout

We knew of Zymergen’s initial work because a co-investor in some of our other deep tech companies had been the graduate thesis advisor to Zymergen’s CTO. An angel investor introduced us to Josh Hoffman, the CEO of Zymergen, about the same time, several years ago.

When the DCVC team heard the Zymergen pitch, we knew our investment thesis that deep stack leaders (especially those with ground-breaking AI advantages) transform industries matched Zymergen’s approach to engineering biology. We have invested twice since our seed investment, in fact leading the round prior to Softbank’s, because we believe the company transforms many pivotal biology problems into solvable engineering problems with multi-billion-dollar outcomes. The founding team came to us with a vision to unite genomics, robotics, hyper-scale data, massive compute, and artificial intel­li­gence to improve the industrial production of materials, and they continue to execute on that vision.

DCVC is pleased that we were able to contribute materially to Zymergen’s current and ongoing success. From day one, we were involved in technology strategy, helping analyze and decide on software and systems imple­men­ta­tions, providing expert knowledge from our Equity Partners (our network of world-class tech­nol­o­gists, scientists, and entre­pre­neurs who share in common in our firm’s results) and key intro­duc­tions from across our portfolio of other cutting edge companies.

We played a hands-on role in recruiting key employees, delivered critical Fortune 500 customers from initial intro­duc­tion through making sure they signed contracts, advised on important intel­lec­tual property and patent issues based on decades of experience, and offered management unyielding support and a sounding board while still delivering frank and construc­tive advice. At elite gatherings of both research and industry leaders we host, we made sure Zymergen had a chance to recruit talent, make alliances, and pitch customers.

When Zymergen needed substantial financing to capitalize on its early progress, but didn’t want to take months to persuade outside investors, we stepped up to lead the key ​“inflection” financing. And, DCVC played an essential and active role in assembling the partic­i­pants for and then closing this current round of funding.

Looking to the Future

At one of our annual DCVC events, Josh described how Zymergen’s technology would impact drug delivery. The coatings on drugs are not capable of releasing a steady dose throughout the day. What if you could create a peptide enclosure for drugs that could precisely release ther­a­peu­tics all day long?

What if you could create surgical glue that was both hypoal­ler­genic and antimi­cro­bial? Or bandages that healed as they were absorbed harmlessly into the skin? What if you could create polymers for flexible optical electronics? The possi­bil­i­ties for truly novel microbial-derived products are endless and inspiring.

But the problem space seems daunting.

Today, these innovations seem beyond reach because of the infor­ma­tional complexity of the microbial genome, and the time to test hypotheses about how to put it to use. However, Zymergen’s ability to iteratively search, engineer, and test organisms at orders of magnitude greater speed than has existed to date is already delivering break­throughs.

We can’t talk about these break­throughs yet, but if you don’t read about them here, you probably will in a major news outlet soon enough.