Generate Biomedicines, Inc. Common Stock 12th Annual Cantor Fitzgerald Global Healthcare Conference

Yeah. So, Eric, first and foremost, a huge thank you for the opportunity. Generate is a company that was founded in 2018 with a premise that, at the time, was somewhat unique. Could you use machine learning-based approaches to design protein therapeutics? I think today that sounds a lot more normal, but certainly back then, to the protein establishment, this was almost heresy. The way proteins were designed historically was bottoms up, using biophysical-based means. Our co-founder, Gevorg, his work at Dartmouth led him to this belief that a data-driven approach could lead to superior answers. The company was founded with that premise. The longer-term vision is that biology ultimately could become programmable. Proteins are code, and if we could understand that code, we could ultimately program biology to drive desired effect. We did not recognize how fast the field would change.

Clearly, when the company was founded, this is a pre-AlphaFold moment, a pre-generative AI moment, and yet the early results we have seen from the technology allows us to manipulate proteins in ways that historical technologies have just struggled. The company has now put five molecules into the clinic. Our lead program is in phase III of development, for the treatment of severe asthma. We have a number of other programs in the cancer domain. The technology is very scalable, so we can have the capacity to start somewhere between five and 10 programs a year. We work not only for our own programs, but also with key partners like Amgen and Novartis. We are scratching the surface of the potential of the technology and excited by what is to come in the years ahead.

Well, I think it is actually this intersection, right? What we are seeing is these models are really, really powerful. I think that is what the recognition of AlphaFold was actually about, was these models can understand what is an unimaginably large search space exquisitely well. The average protein is about 200 amino acids long. There are 20 natural selections. The combinatorial possibilities is about atoms in the universe cubed. That obviously is a very hard thing to randomly explore, which is the way we have tried to design proteins in the past. What distinguishes Generate is actually a virtuous cycle of an integrated dry and wet lab. The reason that is important, if you think about where generative AI has had the greatest effect, it is in domains where you have rapid verification cycles. You can think about gaming as where DeepMind started the company.

You can think about domains like coding or domains like math, which we are seeing breakthroughs today. All of those have a rapid feedback loop. Biology does not have a rapid feedback loop, and so the lab becomes that feedback loop. A lot of our effort is not only pushing the frontiers of computational design, but also experimental throughput and pace. So we are able to survey this space in very distinctive ways. Now, I think, Eric, you were kind of alluding to some of the hype in the field. These tools will not solve therapeutic challenges, biology, in the next 5 years. I know some people are saying it is going to cure all cancers on that sort of a timeframe. I also think what is obscured from the public's view is that the lead times in biology actually understate this current state of the art for these sort of technologies.

It takes us 10 to 15 years to make a drug. The mere fact that our lead candidate is in phase III, about 5 years from design, is emblematic of how fast the field is moving. That molecule was designed on a model that predates the original release of ChatGPT. So if you think about where the current bleeding edge is, it has advanced massively. So while we agree that there is a lot of bluster in the space, we also believe that the potential of these sort of technologies could be transformative in a number of domains in drug discovery.

Eric, I would also underscore the type of data that we use to train our models is very differentiated. The scale in which, just so underscoring some of the things that Mike said, the scale at which we can measure is on the order of billions. The cryo-electron microscopy that we have, we have 4 cryo-electron microscopes that look at an atomistic resolution of protein-protein interactions on a dynamic basis. So these are not just static structures. So we are able to actually understand what is happening at binding sites and use those data on a three-dimensional perspective to help train the models. So while models may commoditize at the end of the day, the types of data will inform and lead to ultimately better results, which we think will be differentiated.

We've had to tear up our infrastructure three times in an 8-year company. We were a digitally native company. The hire number five at Generate was a bioinformatics hire to create a digital infrastructure for the company. The founding of the company predated the release of the LLMs. That sort of technological innovation forced us to rethink our digital infrastructure. And again, around the turn of the year, we started to recognize that a whole different agentic approach to science was going to be the future. And so part of it is, and I think this is the benefit of being an insurgent company versus an incumbent. Part of the reason I was always plagued at Merck was there are deep-seated cultural barriers to rapid technological change moments.

In a startup type environment, you have both a selection bias for the people that select into that domain, but also the company was founded with this kind of digital orientation. We embrace new technologies at a rate and a pace that larger companies, I think, simply struggle to do. Lastly, but I think importantly, the infrastructure also has been purpose-built. What large companies are trying to do right now is retrofit. It is kind of like an old home where you are trying to modernize an 1800s Victorian house. The reality is sometimes it is easier just to build from scratch. What we are finding is that given the digital infrastructure that is required to pursue biology in this way, there are huge advantages to being in this insurgent moment.