With biotechnology on the verge of a “ChatGPT moment,” Ginkgo Bioworks has ambitious plans to position itself as the OpenAI or Anthropic of that wave.
The biotech company recently rolled out its own large language model (LLM) for building proteins, as well as an API that allows researchers and developers to access that model and other synthetic biology AI systems. The announcement builds on a partnership Ginkgo announced with Google Cloud last year.
The rollout comes as several new companies have sought to bring the same fundamental generative AI tech behind models like ChatGPT to programmable biology, which could aid in drug discovery and other breakthroughs.
Ginkgo sees two of its key advantages in this race as its proprietary data and automation tools. It wants its API to serve as a marketplace for developers to build on top of, similar to what big LLM companies like OpenAI and Anthropic offer for the tech industry, according to Ankit Gupta, head of AI at Ginkgo.
“We’re basically making this ecosystem by which Ginkgo becomes a tools provider to the entire industry,” Gupta told Tech Brew. “What we’re doing here is basically making a model API akin to the OpenAI or Anthropic APIs, so that others can now build great tools on top of them and access those model predictions.”
Ecosystem era: For Ginkgo, there’s a lot riding on this new chapter. The company has undergone mass layoffs and seen its stock price crater as the once high-flying startup struggles to turn its technology into sustainable revenue.
Gupta said the rollout of the model and the API represent a broader shift in Ginkgo’s business model from tackling biology problems directly to supplying researchers and developers with the tools to do so themselves.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
“I see this being the start of a pretty significant transformation,” Gupta said. “The history of Ginkgo has been, ‘Come to us with your hard problem, we will solve it for you and charge you for that.’ And I think we increasingly see now that we’ve built a lot of incredibly powerful technologies to solve problems, and…we should try to make these available for others to solve their various problems as much as we can.”
Fine-tuning needed: Despite the flood of new protein design models, Gupta said there is currently a lack of tools and robust APIs to serve as a foundation for biologists and pharma companies to fine-tune specific biological or drug discovery models.
For instance, a company might want to train a base protein foundation model with additional focused data so that it can better predict certain properties of antibodies, Gupta said. The fine-tuning process works with the exact same architecture as language-based LLMs, but with bio data substituted for big tracts of internet text, Gupta said.
Despite the current rush to build out biotech-specific foundation models, however, Gupta stressed that this kind of synthetic drug discovery is still in the very earliest stages.
“Drug discovery—especially protein and biologics drug discovery—these are very, very, very, very hard problems. And we are really just at the very early days of any AI meaningfully making a difference here,” Gupta said. “I do think that the future of this discipline will involve every company competitively developing therapeutics to be using these tools, and either they’re going to spend a lot of money doing it themselves, or they’re gonna find providers that can enable them to do so.”