By Jonathan D. Grinstein, PhD
When Molly Gibson, PhD, left her position as Kaleido Biosciences’ leader of computational biology and joined Flagship Pioneering in 2017, she had been mulling over the question of what the main drivers of biology were for rethinking the drug development process.
“If you could look at the millions of protein sequences to learn nature’s rules for conferring functionality and then generate novel proteins for specific functions, you could almost reimagine the entire drug discovery process and change the economies of scale, the number of programs any individual company could go after, and ultimately the outcome and the success rate of those medicines, which the field and the industry as a whole are in desperate need of,” said Gibson, who is chief strategy and innovation officer at Generate:Biomedicines and senior principal at Flagship Pioneering, in an interview with GEN Edge last December.
“The mission is being able to reimagine the productivity challenge in drug discovery and using machine learning (ML) to make the process reproducible where we can program molecules versus relying on this process of discovery and this bespoke trial and error process.”
At the time, biophysics dominated the field of protein sciences. It was the only way to understand and design new proteins. However, Gibson, whose doctoral training was in computational and systems biology, said that this approach has a lot of problems with scalability and brittleness.
So, in 2018, together with Gevorg Grigoryan, Gibson founded Generate:Biomedicines to develop a generalizable data-driven approach to understanding the relationship between protein sequence and function, with the end goal of developing proteins that had never been seen before.
“We did some early experiments to show that on toy systems like GFP, we could create things that are 50 times better, but also on really important problems, showing that we could optimize antibody properties, which underlie some of the most impactful therapeutics on the market, to create entirely novel therapeutics,” said Gibson. “Once we were able to show that data-driven approaches to proteins were possible, we could start thinking about generating entirely new molecules. So if you understood those rules, you could learn from them, and then you could use the rules and apply them in generative ways.”
There are two core components to Generate:Biomedicines’ technology. One is based on de novo protein generation, in which a computer suggests sequences with binding specificity without previous knowledge of binding on a target of interest instead of being limited to what the immune system produces. The second piece is around optimizing. Computationally thinking up a protein is one thing, but making it a viable therapeutic, which requires mastery of affinity, immunogenicity, and manufacturability, is another. To address this, Generate:Biomedicines has created an optimization suite that allows them to take raw native proteins and make them into viable therapeutics for desired targets.
Gibson’s vision goes beyond this initial drug discovery process and says they are innovating across the entire drug development pipeline. She said they have built out the company to develop drugs end-to-end and constantly think about how our technology impacts every part of that drug discovery process.
“We weren’t going just to reimagine how proteins are generated,” said Gibson. “In that initial process, we felt it was important to reimagine the whole paradigm, from therapeutic hypotheses all the way to when a patient is first dosed to when you have an approved drug.”
Fast forward to today, and Generate:Biomedicines has announced the largest Series C for a biotech company in 2023 at $273 million, bringing their total equity financing since 2020 to nearly $700 million and eclipsing the $226.5 million raised by Apollo Therapeutics just weeks earlier.
Behind their $370 million series B, the five-year-old company seems to have been able to put some of their money where their mouth is, as they also announced their first-in-human clinical trial for GB-0669, a monoclonal antibody targeting a highly conserved region of the spike protein in SARS-CoV-2. Also, Generate:Biomedicines expects to file a Clinical Trial Application by early Q4 2023 for its anti-TSLP monoclonal antibody in asthma, which is expected to enter clinical trials shortly thereafter.
The Series C haul will support Generate:Biomedicines in advancing their entire pipeline of 17 existing programs, largely focused on three different therapeutic areas: immunology, infectious disease, and oncology.
In addition to company founder Flagship Pioneering, all of Generate:Biomedicine’s existing series B investors participated in this round, including a wholly owned subsidiary of the Abu Dhabi Investment Authority (ADIA); Fidelity Management & Research Company; funds and accounts advised by T. Rowe Price Associates; ARCH Venture Partners; and March Capital. Additionally, This financing round attracted many new investors, including Amgen; NVentures, NVIDIA’s venture capital arm; MAPS Capital (Mirae Asset Group); and Pictet Alternative Advisors.
From idea to IND
In March 2021, Merck executive Mike Nally joined Generate:Biomedicines as CEO, and it didn’t take long for him to have the Kool-Aid coursing through his veins.
“One of the things that became quickly apparent was that the early lessons that the technology was starting to suggest, if true, would transform the way drugs are made,” Nally told GEN Edge. “We’d migrate from the artisanal approaches of immunizing a humanized mouse or llama and move toward a paradigm of increasing programmability. If we could master the programming of proteins, we could drive biology in new, unique, and desired ways.”
So, over the last two years, Nally has worked with the rest of the leadership team to build up the company from Gibson’s and Grigoryan’s initial ideas and advancements, and they have demonstrated the ability to de novo generate an antibody and made the manufacturing pipeline reproducible and robust.
When I spoke to Gibson nine months ago, she said they expected two INDs in 2023, which looks like it will happen.
The first-in-human lead program is an antibody with pan-SARS-CoV-2 specificity, which allows them to maintain efficacy with new variants, which Gibson hopes will allow Generate:Biomedicines to demonstrate that their AI-generated molecules are safe and effective in humans. This is a relatively newer program that, under Nally, was taken from an idea to their first-in-human lead candidate, intended for specific immunocompromised populations (about 2% of the global population), in a speedy 16 months.
Of the IND expected to be filed in the next few weeks, which is for asthma, Gibson said back in 2022 that it demonstrates the potential to create optimal therapeutics where they’ve taken a molecule that had picomolar binding down to femtomolar binding.
“[Experimentally], this is incredibly challenging because it’s almost impossible even to measure that type of binding—you can do it computationally and then validate at a lower throughput,” said Gibson. “We completely changed the dosing paradigm of this molecule for patients because we have a tighter affinity.”
Since Nally was brought on, there have been several important strategic partnerships for the company—one with Amgen that happened around 18 months ago and another with MD Anderson Cancer Center that was signed this spring—and the pipeline has matured to the point that the company anticipates adding 10 programs annually.
(Not) lost in space
Generate:Biomedicines is just at the very onset of applying their technology, which is modality-agnostic. According to Nally, they’ve used it across antibodies, T-cell engagers, cell therapy, antibody-drug conjugates (ADCs), enzymes, peptides, and growth hormones. But Nally has kept Generate:Biomedicines focused.
They’ve invested in people, scaling up from about 30 to 60 at the time of Series B, and today there are about 280 people. They’re also trying to spend their money smartly, not reinventing every cog in the drug development wheel. For example, their first two programs are antibodies, so Nally said that they have opted to go with tried and true CDMO partners available for manufacturing, allowing Generate:Biomedicines to deploy their capital where they can add distinctive value on the molecular generation side of things or to advancing candidates.
“As we move things toward and into the clinic, our dollars will go to the toxicity studies, scaling up and manufacturing, getting IND readiness, and then running those clinical trials,” said Nally. “The runway on that will be multiple years. We think over the next 12 to 24 months, we should have four to six clinical assets and four to six well-characterized molecules that sit right behind those ready to move toward the clinic.”
While Generate:Biomedicines plugs in their AI technology wherever possible, they also have a reverence for experimental biology. They’ve got just as many, if not more, scientists in the wet lab as in the dry one.
In June of this year, Generate:Biomedicines announced they were expanding beyond their initial location in Somerville, investing in a cryogenic electron microscopy (cryoEM) laboratory in Andover, MA. It is among the largest privately owned cryoEM laboratories in the United States at 70,000 square feet and houses four industry-leading cryoEM setups, a full end-to-end wet lab, and data processing capabilities enabled by machine learning.
“Why cryoEM is so important to us is that, if you think about the protein-protein interaction data, each one of those points is an X, Y, and Z coordinate,” said Nally. “Feeding in the amount of data you gather from just one image of a protein-protein interaction that gets fed back into the learning loop of the machine learning aspect of this is enormously powerful.”
One thing I always hear when talking to AI-driven drug development companies is some idiom—typically “scratching the surface” or “tip of the iceberg”—communicating how big of a space has just been unlocked and how little has been investigated.
All of this idiomatic repetition has made me consider if there is a better way to describe the potential of this new approach to drug development. Perhaps more fitting is to think of AI-driven drug discovery as only having escaped the Earth’s atmosphere and making it past lunar orbit, but there’s a long way to go before leaving out the solar system, galaxy, and the undefined edges of the unimaginably vast universe. Many of these pioneers are aware of the possibility of being lost in space, so I’d expect the number of AI-driven companies and the amount of money thrown at them to continue to grow if they can stick to their focused slice of the heavens.