Cellarity, Flagship's $50M bet on network biology, marries machine learning and single-cell tech for drug discovery
Cellarity started with a simple — but far from easy — idea that Avak Kahvejian and his team were floating around at Flagship Pioneering: to digitally encode a cell.
As he and his senior associate Nick Plugis dug deeper into the concept, they found that most of the models others have developed take a bottom-up approach, where they assemble the molecules inside cells and the connections between them from scratch. What if they opt for a top-down approach, aided by single-cell transcriptomics and machine learning, to gauge the behavior of the entire cellular network?
“If you look at cell behavior from the perspective of a molecular network underlying it, then you free yourself from the traditional approach of one-dimensional, two-dimensional, three-dimensional target-based or phenotypic-based drug discovery approaches,” Kahvejian, who took on the CEO role, told Endpoints News. “What it allows you to do is to use the network changes as your readout.”
Flagship dedicated $50 million to get the biotech started, which is how Cellarity has been funding the buildout of its platform and animal experiments to verify their initial hypotheses in the past two years.
By intertwining wet labs and a digital twin dubbed the Cellarium, Kahvejian believes his biotech hasn’t just “re-architected” therapeutic discovery, but also the organization of an AI upstart. Chad Nusbaum, founder of the Broad Technology Labs, leads the technical unit churning out data; while Milind Kamkolkar has joined as chief digital & data officer after pioneering the role at Sanofi.
“I wanted to build stuff. I didn’t want to just keep sourcing stuff,” Kamkolkar said of his decision to leave the pharma giant, where external partnership was the protocol for gaining digital competency.
It’s the complete opposite at Cellarity, as they are building a new engine that can be broken down into three layers. He calls the first “data ingestion” — channeling all the information generated by Nusbaum’s team with multiple methodologies and species into a database where scientists can plot and curate knowledge. Then they enter the exploration layer, interrogating the cell behaviors while analyzing how well existing and new compounds can perturb the cells. On the last layer, they visualize the findings by creating a satellite image of sorts.
Right now Cellarity has about 250 of these digital guides on different diseases, which they call Cellarity Maps. And they can encompass every step of the traditional drug discovery process.
“The machines are incredibly capable of parallelizing and collapsing what typically used to be a linear process to try to understand whether the impact of that drug actually does have toxicity or side effects,” Kamkolkar added.
With 40 staffers on board, Cellarity has gone broad with its tech platform, probing anything from epithelial barrier disorders and oncology to hematological disorders and neurology. The platform can accommodate multiple therapeutic modalities, Kahvejian said, and they’ve tested both small and large molecules. He isn’t disclosing a timeline for when they might steer their lead candidates into the clinic, but he’s not shy about the ambition to tackle “dozens of programs” at a time, and partnering as he sees fit.
As of September, Cristina Rondinone, the former head of cardiovascular, renal and metabolic diseases at AstraZeneca, has also joined as president to help grow the company and push it to the next stage, enabling downstream clinical development of leads.
The new hires will find themselves in a horizontal organization where no one domain supersedes the other, Kahvejian said, and where biologists, technologists, and the computational folks work together in an integrated and multilingual environment where insights are generated more quickly and are “actionable the minute they are generated.”
Kamkolkar recalled the surprise of a machine learning scientist when he found out that he was to spend time in labs and see how the data are generated.
“Yeah, you’re gonna have to go in labs,” Kamkolkar basically said. “It’s quite unique.”
