While companies like OpenAI and Anthropic continue to popularize the idea of using ordinary language to ask artificially intelligent agents to answer questions, write proposals, and draw pictures, Basecamp A London startup called Research has raised $60 million to tackle new frontiers. We are building AI agents that not only answer all your questions about biology and the biodiversity of the natural world, but also generate new insights that humans alone could not achieve.
“There are huge data gaps where people are training today. [biology] Glen Gowers, co-founder and CEO of Basecamp Research, said in an interview: “Some of the world's top pharmaceutical companies are training models that don't fully understand the natural world.”
The funding comes on the heels of the startup's notable momentum. To date, Basecamp Research has formed more than 100 partnerships with organizations in 25 countries to expand its database of primary source information, Gowers said. About 15 of them are using AI to help build new products. Procter & Gamble uses this model to design enzymes for detergents that remove stains at low temperatures. Colorifix is working on new, more sustainable textile dye formulations.
In particular, when it comes to accurately predicting the interactions between large, complex protein structures and small molecules, Basecamp Research has proven that its foundational model, BaseFold, is the AlphaFold 2 (DeepMind's developer just won the Nobel Prize in Chemistry today). claims to be better than.
The startup's approach to building biological AI is incredibly ambitious, building models from scratch.
Gowers and co-founder Oliver Vince, both with PhDs in biology, met while undergraduates at Oxford University. The name “Basecamp Research” comes from the fact that they lived on top of the ice sheet and used hardware they built to sequence DNA, Vince said.
“We pioneered the first mobile DNA sequencing laboratory,” he said.
Basecamp Research has adapted its hardware components into “very small units” to collect data for new startups, he added.
The field of biology has produced hundreds of books, thousands of pages of research, and petabytes of data over the decades. The problem is that much of that data is old, unstructured, and simply inconsistent. That's why Basecamp Research builds AI by meticulously collecting primary data directly and building models from scratch. The goal is to create AI that can gain better insights into biology than humans, given the wide variety of data available.
“We literally travel around the world collecting data, using a combination of exploration to understand things like hot springs and volcanoes, and then using that to train large language models to build. Combined with an artificial intelligence program that is purely focused on “ChatGPT'' for nature,'' Gowers said. The startup is also amassing what he said could be the “largest computational cluster” dedicated to nature to power this.
Just as ChatGPT's superpower lies in recalling and formulating natural language responses to questions, the same is true of what Basecamp Research is trying to do. The difference is that given the breadth of information in the world (Vince estimates that we only know about 1% of the world's biodiversity information), we mere mortals have no idea about this issue. It means they don't even have the ability to ask the right questions. point.
Or as Andy Conrad, S32 backer and former CEO of Google's Verily Life Sciences, puts it: Basecamp Research's platform “can address questions the biopharmaceutical industry didn't even know to ask.”
“In other words, it doesn't understand text or audio language; [our platform] Because we understand the language of DNA and we understand the language of biology, we can go beyond what humans can do in the realm of biological design,” Gowers continued. “We're traditionally very bad at understanding DNA, so these language models can be really, really, really good given enough data.”
The Series B, led by European company Cingular, is a “multi-year collaboration between Basecamp Research and Dr. David R. Liu and the Broad Institute, a leading biomedical research center spanning MIT and Harvard University. ” is carried out in parallel with what is expressed. The plan is to use this funding to continue building the startup through partnerships with other biomedical and research institutions, and by accumulating more data to expand the model.
Beyond this, Basecamp Research's roadmap includes supporting organizations tackling drug discovery and other big challenges that involve understanding and better utilizing the natural world.
There are commercial deals involved, but the startup's work with the Broad Institute sheds light on what form this will take. Currently, the lab run by Dr. Liu is researching “novel fusion proteins and other macromolecules” used to create genetic medicines, and is using Basecamp Research's datasets in their development.
Although unlikely, it appears to be an actual “ChatGPT” style interface for startups. Gowers said the company currently believes there are more opportunities to work on a B2B basis rather than putting resources into building products that engage the general public. But that doesn't mean such products might not be on future roadmaps, he added.
This also appears to be the approach taken by other companies building large-scale “scientific” models. Jua, which builds large-scale physical models, initially targeted organizations that needed better insight into weather patterns.
Basecamp Research does not disclose its valuation, other than that this Series B is an upside round. For context, the startup has raised $85 million to date, with previous investors including Hummingbird, True Ventures, and strategic backer Valo. PitchBook gave it a very conservative valuation of $71 million after 2022.
Series B also included S32, redalpine and Roche Vice Chairman André Hoffmann. Feike Szybesma, Chairman of Royal Philips and former CEO of DSM. Paul Polman, former CEO of Unilever.