While companies like OpenAI and Anthropic continue to popularize the idea of using ordinary language to ask artificial intelligence agents for answers to their questions, to write their proposals, or to draw pictures, a London startup called Basecamp Research has raised $60 million to tackle a new frontier: an AI that not only answers any question related to biology and the biodiversity of the natural world, but produce new insights that humans could not achieve on their own.
“There is an enormous data gap that exists today where people are training [biology] models,” said Glen Gower, the CEO of Basecamp Research, in an interview. “Some of the top pharma companies the world are training models that simply don’t see enough of the natural world.”
The funding is coming on the heels of some notable momentum for the startup. To date, Gower said, it has inked more than 100 partnerships across 25 countries, with organizations to expand its database with primary-source information, and some 15 that are using its AI to help build new products. Early examples of those products are Procter & Gamble, working on formulations for new fabric dyes that are more sustainable.
Notably, Basecamp Research also claims that its foundational model BaseFold has outperformed AlphaFold 2 — whose creators at DeepMind just today won the Nobel Prize for Chemistry — when it comes to predicting accurately large, complex protein structures and small molecule interactions.
The startup’s approach to building an AI for biology is incredibly ambitious: it is building its models from the ground up.
Gower and his co-founder Oliver Vince are both biology PhDs who met back in their undergraduate days at Oxford. The name “Basecamp Research” comes from time they spent living on an ice cap, Vince said, doing DNA sequencing using hardware they had built themselves.
“We pioneered the first mobile DNA sequencing laboratory,” he said.
Basecamp Research has adapted components of that hardware “into very small units” to collect data for the newer startup, he added.
There have been hundreds of books, thousands of pages of research, and megabytes of other data generated over decades in the field of biology. The problem is that much of that data is outdated, very unstructured and simply inconsistent. So to build its AI, Basecamp Research is meticulously gathering primary data first-hand to build its models from the ground up, the aim being to produce an AI that will be able to have better insights into biology than any human can, simply because of the breadth of data that can be brought to bear.
“We use a combination of exploration — literally going around the world to pick up data, understand Hot Springs, volcanoes, those sorts of things — and combine that with an artificial intelligence program that is focused purely on training massive language models to build, effectively, a ‘ChatGPT’ for nature,” Gower said. The startup has also amassed what he said may well be the “largest compute cluster” also dedicated to the natural world to power this.
Just as ChatGPT’s superpower is in recalling and formulating natural language responses to questions that are asked of it, the same goes for what Basecamp Research is setting out to do. The difference is that the breadth of information in the world — Vince estimates that we have only managed to capture some 1% of information about our world’s biodiversity — means that we mere humans just don’t even have the capacity to ask the right questions at this point. Or, as backer Andy Conrad of S32, previously CEO of Verily Life Sciences at Google, puts it: Basecamp Research’s platform can “address questions that the biopharma industry hasn’t even known to ask.”
“So rather than something that understands the language of text or speech, [our platform] understands the language of DNA, understands the language of biology, and therefore can go past what humans can do in the biological design space,” Gowers continued. “We are traditionally very bad at understanding DNA, and therefore these language models, if given enough data, can really, really, really excel.”
The Series B, led by the Paris firm Singular, is coming alongside what Basecamp Research is describing as a “multi-year collaboration” Dr. David R. Liu and the Broad Institute, the major biomedical research center that works across MIT and Harvard. The plan will be to use the funding to continue building out the startup, both through partnerships with other biomedical and research organizations, and by amassing more data to expand its models.
Beyond this, however, Basecamp Research’s roadmap includes helping organizations with drug discovery and other large challenges that touch on understanding and making better use of the natural world.
While there are commercial deals being linked, its work with the Broad Institute sheds light on what form this might take. Right now, the labs run by Dr. Liu are looking at “novel fusion proteins and other large molecules,” used to create genetic medicines, and they are using datasets from Basecamp Research to develop these.
What is less likely, it seems, is an actual “chatGPT”-style interface for the startup. Right now, Gowers said they see more opportunity of working on a B2B basis rather than channelling resources into building a product to engage with the general public. That’s not to say that might not be on its roadmap down the line, he added.
(This appears to also be the approach that others building large “science” models are taking: Jua, which is building a large physics model, has initially targeted organizations that need better insights into weather patterns.)
The startup is not disclosing any valuation, except to note that this Series B — which also includes participation from S32, redalpine, André Hoffmann, Vice-Chairman of Roche, Feike Sijbesma, Chair ofRoyal Philips and Former CEO of DSM, and Paul Polman, former CEO of Unilever — is an upround. For some context, Basecamp Research has raised $85 million to date, with previous investors including Hummingbird, True Ventures, strategic backer Valo and others. PitchBook put its last valuation, from 2022, at a very modest $71 million.
