Koch's 'disruptive' VC bets $110M on one of Flagship's new, big computational startups
David Berry joined Flagship as a Harvard and MIT wunderkind in 2005 and by now he’s been at the firm almost as long as anyone not named Noubar. He’s played a hand in some of their biggest startups, including Seres, Omega, KSQ, Evelo and, of course, Moderna.
Flagship is known for its flash, but over the last couple years, Berry has been quietly building a computational startup in Boston that he thinks can remake drug development in a way that, for all the buzz, machine learning and artificial intelligence have yet to do.
He’s managed to convince at least a few investors of that vision, raising $100 million in a Series A, $50 million in an undisclosed financing and a $190 million in a Series B despite remaining far from the clinic. On Tuesday, Valo revealed another investor who bought in, adding $110 million from Koch Disruptive Technology, the venture capital arm of the famed and infamous holding company Koch Industries.
Koch Disruptive Technologies has invested in a broad swatch of tech companies but not yet waded deeply into biotech. In backing Valo, they are buying into what Berry says is a new way of thinking about drugs and tech and data.
“As we look at the process that exists in drug development or drug discovery today, it’s effectively a point-to-point process and each of the steps has its own data, its own way of making decisions,” Berry told Endpoints News. “As the drug moves through its various steps, it’s as if it’s been thrown over a wall.”
AI and machine learning have largely been deployed at individual steps of the drug development process: used by Atomwise, for example, to screen massive libraries for the best molecule, or by Insitro to find the difference between diseased and healthy cells, and the places where developers might be able to intervene and turn one to the other. Separately, large pharmas and small biotechs like BlackThorn have used machine learning to design trials and find the best patients.
Berry argues that we need a better system — a better platform — that can integrate all those various sources and applications of data from the outset, allowing companies to overcome the hurdles that tend to appear as a drug goes through development. And that data should be grounded in humans (as opposed to mice or monkey experiments) to increase the odds that basic lab insights actually translate into the clinic.
With about 110 employees, large amounts of human data, including what they claim is the “largest and most-detailed cardio-metabolic dataset in the world,” and reams and reams of cloud space, they’ve built a platform they call Opal to do that.
For example, Berry claimed, they’ve developed an algorithm that allows them to predict the toxicology of any given compound with 86% accuracy. They also use 3D physics software to simulate and design molecules, adopting a similar approach to the well-partnered computational biotech Schrödinger, which has long argued that the AI algorithms many startups use to screen for molecules struggle because they rely on 2D images of 3D molecules.
“When you ask the computer to try learn that a 2D image represents a 3D structure — that’s a very hard thing for a computer to learn,” Berry said.
To find new targets, the platform and Valo’s scientists combine “omics” data – genomics, proteomics, even the much less talked about metabolomics — with data that track how patients change as they age and their diseases progress or wane. “Diseases are dynamic,” he said.
Valo will use the new proceeds to continue hiring a few dozen more employees and progress the platform. With two years of runway, it will also give them the cash to get into the clinic, Berry said. Although they haven’t gone deep into details, Valo did release their first batch of cancer targets earlier this year: NAMPT, PARP1, USP28 and HDAC3.
Their targets for neurodegeneration and cardiovascular disease remain undisclosed.