Seer raises $55M as proteomics approaches prime time in disease detection, drug development
Omid Farokhzad was working in his Harvard nanotech lab in 2017 when he found something he knew would become his next company. Two years and, as of today, over $100 million in funding for later, he still hasn’t revealed what it is.
“We’ve seen them reported, but we don’t really know what their technology is or what it does.” Stephen Williams, the CMO of SomaLogic, the most high-profile potential competitor to Farokhzad’s company, Seer, told Endpoints News.
The technology was big, though, or at least Farokhzad thought so. Big enough for him to leave much of his work in Boston, where he launched two companies, opened his own lab and spent 20 years across Harvard and MIT, to spend most of his time building Seer in Redwood City, outside San Francisco. What he said was this: They were trying to best-of-both-worlds a scientific idea. They wanted to take a study of human proteins they said was often conducted with either slipshod speed or painstaking precision and make it both fast and meticulous.
Now, exactly a year after its public launch and just as SomaLogic eyes its biggest commercial expansion, Seer is announcing $55 million in Series D funding and its own pivot from a research-focused company to one trying to bring the proteome into the medical mainstream.
“There was the moment, where I said I need to do this myself,” Farokhzad told Endpoints, “and if I don’t I’m going to regret it for the rest of my life.”
Seer, SomaLogic, and a handful of other companies work on what’s become known as the proteome. For decades, it had been tossed around as a biological what-if, as seductive as it was unattainable. Perhaps instead of looking at DNA and RNA to understand the human body and its disorders, we could look at the proteins that genetic code makes, the ones that in most cases ultimately do the damage.
The hypotheticals were huge. Instead of predicting, based on genetics, if a 50-year-old had a predisposition to heart attacks or diabetes, a doctor could check if there had been actual changes in the blood that built to those ailments. Pharma companies could better understand how their drugs worked, and refine accordingly.
One obstacle was mathematical. With proteins, there are simply far more possibilities. DNA has four letters and bonds in very specific ways and structures: A-T, G-C, the double helix. Proteins are formed from up to 20 amino acids and can bond in myriad ways, and then change after their creation into a variety of states depending on what bonds to them.
“You actually need a lot of computing power to calculate the theoretical complexity you can create with proteins,” Farokhzad said. “You’ll never get there.”
The other problem was chemical. DNA and RNA can be replicated and amplified in a lab. Proteins cannot. Instead, researchers have relied on biological fishing expeditions. First, they used antibodies, which bind to proteins and make them easier to spot. But each antibody only binds to one protein. Using them to identify every human protein would be, as the New York Times’ Michael Behar put it last year, “like trying to catalog every fish in the ocean with a net that captured only a single species at a time.”
SomaLogic built itself on the discovery that nucleic acids called aptamers could be used to identify far more proteins at once. They built a database of 5,000 human proteins, and then use machine learning to see how changes in those proteins’ structures and quantity correlate with certain health factors. After decades of development and years of clinical tests, they launched their SomaSignal for medical practices in September and burnished it with a December study in Nature that scanned 17,000 participants for those 5,000 proteins and predicted for diabetes and heart attack, among other health issues.
Those who tested as high-risk could then be referred to preventative measures they may not otherwise have received.
Stephen Williams came to SomaLogic in 2009 after nearly 20 years in experimental and clinical development at Pfizer
Click on the image to see the full-sized version
“We tried to develop 13 models for 13 health issues, and we succeeded with 11,” Williams said, adding they failed only on weight predictions and current diet. “We really answered the question: Can proteins become a sole information source?”
Farokhzad’s critique is that this is a ‘biased’ approach: SomaLogic can only look for the proteins it has aptamers for and it also can’t look deeply into specific changes. He said his Seer will be ‘unbiased,’ rapidly identifying every protein – at least 20,000 – and all the changes that can occur within them and then use their own machine learning to find patterns. “Fast and broad,” he likes to call it.
“To really understand proteomics, you have to do it in an unbiased approach,” he said. “New biology is always driven by what we don’t know yet.”
Out in Redwood City, Farokhzad has set up a Seer headquarters with 40 employees and plans to expand to 80 with the new funding. He said they’ve spent the last year with their heads down trying to perfect the science. Afraid biobanks might contaminate samples with poor procedures, they’ve set up 70 clinical sites nationwide to build their database.
Other folks have used an unbiased approach – helping build databases like the Human Proteome Project, which now counts 19,823 proteins – but combing through all those amino acids takes forever. Farokhzad has promised to make it fast enough to work with drug companies (they already have partners) and unveil scans like those offered by SomaLogic.
If the tech holds up, it would have big implications for early detection, unveiling previously hidden changes in the body that presage disease. Farokhzard claims it could help clinical trials, naming Alzheimer’s and the hypothesis that some of the drugs that failed might work if they treat a patient early – at a stage we can’t yet detect. Like many in proteomics, he talks about it like a new frontier.
“We’re basically opening the floodgate of access to proteomic information,” he said.
In SomaLogic’s view, though, that grand vision isn’t necessary. All approaches are biased, Williams said. The question is whether you can make predictions that can be put to medical use.
“The question is whether you can measure something that is useful or actionable, not whether there’s some inherent truth,” he said. “For our business, we don’t need to understand what the biology is. We let the machine learning pick out the best combination of measurements.”
Still, Williams said that when their longtime partner Novartis – which uses SomaLogic’s tech to understand the mechanisms of their drugs, among other things – needs a higher resolution look at what SomaLogic technology turns up, they direct them to a different platform. Seer is developing a commercial scan like SomaLogic’s and, like SomaLogic, eventually hopes to put out a (somewhat controversial) chip anyone can use, but their easiest niche may be in drug development.
The industry will know soon enough. SomaSignal is only available in Colorado for now, but a bigger rollout is pending and they just hired a new commercial officer. After two years in relative obscurity, Farokhzad said Seer will soon be opening up, and a product launch is scheduled for 2021.
“Our technology risk is now largely put to rest and what lays ahead of us is basically product development and commercialization,” he said. “Now we’ve got to execute.”