Generate: Biomedicines says the $370 million it raised in Series B financing this month will enable it to live up to its name.
That name, the company says, summarizes its expansive mission of harnessing data and machine learning at scale to rapidly generate countless protein therapeutics in numerous modalities for a variety of diseases.
With protein-based therapeutics accounting for most of last year’s top 15 best-selling new drugs, and a growing number of drug developers focused on turning proteins into treatments, how does Generate expect to stand out?
“We’ve focused exclusively on this protein therapeutics domain. Given the complexity of proteins, trying to distill out generalizable principles and ultimately have a certain level of command over them, we think, will create a huge opportunity as we go forward,” Generate CEO Mike Nally told GEN Edge.
“Being able to, in some ways, create completely de novo proteins with desired attributes opens up a huge opportunity to have a profound impact on global health.”
The market for protein therapeutics is expected to grow through 2027 at a compound annual growth rate of 6.86%–to $290.69 billion from $182.69 billion this year, according to a report issued in June by Market Research Future.
To achieve that growth, protein-focused drug developers like Generate will need to surmount longtime hurdles to therapeutic development, Nally acknowledged.
“Nature has only sampled a small fraction of potential proteins over the course of history, so our ability to truly have command and explore the totality of protein space opens up entirely novel amino acid sequences with desired therapeutic categories,” Nally said.
Over the past three years, Nally said, Generate has studied every known protein structure, every protein-protein interaction, and roughly 190 million amino-acid sequences encoding protein structure and function. That research has yielded “tens of thousands” of completely novel proteins and amino-acid sequences with functions sought by the company, then integrating its wet lab work with computational capability.
Generate doesn’t call itself a synthetic biology company, though Nally said the company uses that tool extensively to program DNA molecules and build them into proteins.
The company’s platform integrates protein science expertise with structural biology, machine learning algorithms designed to analyze many millions of known proteins and peptides, looking for statistical patterns linking amino-acid sequence, structure, and function, data science, and computational infrastructure—all that is supplemented with Generate’s own proprietary experimental data.
The platform can also generate antibodies designed to bind specific epitopes on desired targets, allowing in silico generation of potent antibodies on demand. It can also generate functional and agonistic antibodies, as well as antibodies to integral membrane proteins, multiprotein complexes, and other long difficult to hit targets.
“Sometimes you struggle with the binding affinity. Sometimes you struggle with protein expression or immunogenicity challenges,” Nally said. “There’s a host of manufacturability and developability parameters that often plagues different protein therapeutics. With the technology, we have ways of optimizing on all of those different parameters to ultimately come up with the best therapeutic alternative for the targets we select.”
Generate’s development approach, which it calls “generative biology,” is designed to facilitate testing and learning which proteins work best against which disease targets. The company projects its approach can shave two years off traditional protein-based optimization processes.
“The vast majority of protein spaces are non-functional, so being able to reverse that that broad landscape and find different functional alternatives with desired attributes, I think, opens up the prospect of ultimately getting to where you can instantly program and instantly generate a desired therapeutic with the attributes that would be unique to an individual’s condition, if you had complete mastery over that domain. That’s the first logical extension of the technology,” Nally said.
“Obviously we’re a long way from that today, but we think we’re on a pretty exciting road where those sorts of discoveries can become more and more prevalent as we apply the technology.”
For each individual target, Generate has been able to generate 100 variants—a limitation, until now, of the company’s ability to produce and validate targets.
“We’re now investing heavily in automation, so we’re moving from 100-variant sets to 10,000-variant sets, on the way to 1 million-variant sets,” Nally said.
The average protein is 100 amino acids long. Given the 20 amino acids that constitute the protein family, the potential universe of variants is 20100—more than the number of atoms in the observable universe, estimated at between 1078 and 1082.
“Being able to traverse that entire landscape allows us to come up with countless permutations in terms of functional proteins across that entire landscape,” Nally said. “We’re learning around how different sequence diversity and different functional diversity can refine our overarching computational approach. That learning loop is extraordinarily powerful, and it allows us then to really optimize on all of these different parameters to ultimately come up with optimal medicine.”
Nally joined Generate in March from Merck & Co., where he spent 18 years, most recently as Chief Marketing Officer of its Human Health division. Nally previously served as President of Merck’s Global Vaccines division and before that headed Merck’s business in Sweden and the U.K.
Nally is also CEO-Partner with Flagship Pioneering, the venture capital and business acceleration firm whose best-known spinout is Moderna. Flagship launched Generate in 2019 by combining a pair of exploratory projects.
One was FL56, an effort led by general partner Avak Kahvejian to develop an algorithmic drug discovery platform based on insights from Gevorg Grigoryan, PhD, of Dartmouth College, who with colleagues discovered that protein structure formed according to a “language” enabling the engineering of novel proteins that folded and functioned without physical descriptions.
In the other exploratory project, FL57, Molly Gibson, PhD, a principal at Flagship, partnered with general partner Geoffrey von Maltzahn, to test whether advances in machine learning in natural language processing and image processing could be applied to amino-acid sequences of proteins.
Flagship combined the two projects in 2019 into a single company. Two of Generate’s four co-founders, Kahvejian and von Maltzahn, initially served as co-CEOs. Gibson joining as Chief Innovation Officer, and Grigoryan as Chief Technology Officer; both remain with the company. The company’s board is chaired by Flagship’s co-founder and CEO Noubar Afeyan, PhD.
Gibson said she joined as a co-founder company after concluding that machine learning could be applied toward generate proteins much as it has been used to generate novel paragraphs.
“The right language”
“We thought in this case that machine learning was actually the right language to understand biology and then also be able to generate it in very similar ways,” Gibson recalled.
“If we learn from the immense amount of biological insights that nature has given us and evolution has provided, then we can learn those principles and then use them with generative ML algorithms similar to text or art, and generate proteins that function for the way that we want.”
“That was the founding insight that led us to really build this integrated computational and wet lab team to pursue that vision,” she added.
Generate emerged from stealth mode in September 2020 when it completed a Series A totaling $50 million, funded solely by Flagship.
Generate says its platform is disease-area agnostic, with the ability to rapidly generate for any therapeutic need antibodies, enzymes, peptides, even cell and gene therapies.
“Our ability to understand protein-protein interactions, our ability to understand protein function, and generate those protein functions allows us to build up the entire system of gene and cell therapies,” Gibson said.
Nally said the company has worked to find targets in oncology, immunology, and infectious disease, but isn’t limiting its R&D to these areas.
“From a focus standpoint, we’ll start there, and we’ll look to partners where they may have deeper biological expertise and other domains to exploit the technology to its fullest,” Nally said. “We have a capability in protein engineering that we think is pretty unique, but we know we’re going to need partners that can bring either disease-area biology expertise, manufacturing expertise, or clinical development expertise to complement our core capability.”
At the same time, Generate plans to develop its own expertise and structural capabilities for larger-scale growth. The company has begun moving employees into a 70,000-square-foot leased facility in suburban Andover, MA.
“We’ll have a structural biology core of cryo-EM capability that we think will be really important for our future learning, but also, we’ll expand our protein production capability there,” Nally said.
On the move
By the end of the first quarter or early second quarter of 2022, Generate also plans to move its headquarters out of Cambridge, MA, a shade north to Somerville, MA, into 70,000 square feet of leased space within the 1.3 million-square-foot Boynton Yards lab/office/apartment campus. The Somerville site will feature expanded wet lab space.
The Somervile and Andover facilities are part of a company expansion that will see Generate grow from its current 80 employees to about 500 over the next two years.
“The vast majority of our workforce today is and will continue to be scientists. We’re going to greatly expand our computational core. We’re looking to bring in the best machine learning scientists that have an orientation toward protein engineering. We’re also looking to extend our wet lab capability deep into protein sciences preclinical development,” Nally explained.
“We’re also starting to think through how we extend it to process development into regulatory clinical development. Then we’re going to build out to our G&A [general and administrative] functions,” he added. “But the vast majority of our investment will be on the scientific side.”
That investment, he added, will help Generate fulfill its plans to build a pipeline of preclinical protein therapeutic candidates as early as year’s end, with “several” to be advanced into the clinic by 2023.
Flagship Pioneering led the company’s Series B financing, which is expected to give the company a capital runway of “a couple of years,” Nally said.
Joining Flagship in the round were several institutional co-investors, including: a wholly owned subsidiary of the Abu Dhabi Investment Authority (ADIA), the Alaska Permanent Fund, Altitude Life Science Ventures, ARCH Venture Partners, Fidelity Management & Research Company LLC, Morningside Ventures, and funds and accounts advised by T. Rowe Price Associates.
“We were really fortunate to have just a great group of investors that believed in the foundational tenet that computers will help us understand biology in a more profound way. I couldn’t have asked for a better group of investors to support what’s a pretty big ambition,” Nally said.