Stealthy Insitro opens up—starting with Gilead deal worth up to $1.05B

Daphne Koller’s Insitro launched last May with a blog post detailing an ambitious goal: to harness data and machine learning in the creation of a new approach to drug discovery. The post painted a broad vision of what machine learning has already achieved in other industries and how it might transform the way biotech works, but without getting into the weeds. 

Now, Insitro is ready to share a little more—starting with a Gilead partnership worth $15 million upfront, but that could total more than $1 billion. 

Insitro and Gilead have inked a three-year pact to discover and develop treatments for nonalcoholic steatohepatitis (NASH), a space it seems all of biopharma has piled into, but has yet to see an approved drug. The duo will use Insitro’s technology, dubbed the Insitro Human platform (ISH) to create disease models of NASH and discover targets that affect the progression or regression of the disease. Of the drug targets the platform identifies for NASH, Gilead may advance up to five, taking care of all the chemistry and development.

RELATED: Daphne Koller, A-list backers unite to create machine learning drug R&D startup

In addition to the $15 million upfront fee, Insitro could pick up another $35 million in near-term operational milestones, as well as another $200 million for each of the five programs Gilead chooses to take forward. It also has the option to co-develop those programs with Gilead in the U.S., in which case Insitro would receive milestones and royalties on ex-U.S. sales and a share of profits in China. 

“The Gilead team has a tremendous depth of scientific knowledge in NASH and the biology of the disease, as well as a very large amount of clinical samples from different clinical trials run over the course of years. They’re providing that into the partnership, as well as down the line, the chemistry that needs to be done to develop a compound against a target,” Koller told FierceBiotech. 

Insitro listed a marquee group of investors at launch— ARCH Venture Partners, Foresite Capital, a16z, GV and Third Rock Ventures—but kept the amount of capital it had raised under wraps. It’s now revealing that it raised $100 million in its series A financing, and that Alexandira Venture Investments, Bezos Expeditions, Mubadala Investment Company, Two Sigma Ventures and Verily also pitched in. 

RELATED: Can adding Novo's GLP-1 to Gilead's NASH combo help hit fatty liver disease? New trial aims to find out

To get into the weeds, Insitro wants to use in vitro systems—or “test-tube experiments” involving cells outside the human body—to predict what drug developers would see in a human, in vivo, clinical model. It is these models that would propose drug targets and predict how patients, or specific subgroups of patients, will respond to specific treatments.  

It’s a way to get answers without performing illegal, immoral, or just plain impossible experiments on people. 

“We want to take data from humans, which is the thing we really care about but can’t experiment in. So, [instead] we will generate very large amounts of data that is a surrogate of that, if you will, in an in vitro system and use machine learning to align them,” Koller said. 

RELATED: Rewiring R&D: The FierceBiotech JPM 2019 breakfast debate

Those “test-tube experiments” would make up an experimental system based on stem cells taken from humans and differentiated into whatever kinds of cells that Insitro is interested in looking at. That way, the company can “measure a broad set of cellular phenotypes in a scalable way” and “get a broad perspective on what the cell state is for cells that have different genetic burden of disease,” Koller said. 

Insitro work generating data on purpose—as an input for machine-learning models—contrasts with the prevailing mindset that data is a “byproduct,” something that is generated as a result of clinical trials. 

“There’s a growing amount of human genotype and phenotype data out there, some from public repositories like the UK Biobank and some from clinical trials like what we would end up partnering with Gilead,” Koller said. “But those data, while incredibly valuable, are oftentimes relatively small-scale from a machine-learning perspective … they’re datasets, not experimental systems, so you can’t intervene in them.” 

What she means is that Insights drawn from these data tend to be statistical associations that aren’t predictive: “You don’t really know that when you perturb any one of them [genes] whether it’s actually going to make a difference in the phenotype—not until you get into a clinical trial, at which point you’ve invested an untold amount,” she said. 

Insitro’s approach would predict the effect of such a perturbation, as Koller puts it, on patients before the clinical trial stage. It could save money, time and other resources, as well as improve the success rate of drug discovery. 

The Gilead collaboration is just the first step. Insitro is thinking about pursuing more partnerships and wants to become a drug development company in its own right. Koller couldn’t elaborate, but said that Insitro is in talks around one or two deals it might want to pursue in other indications. 

“I think as we are building out this hopefully very robust, broadly applicable platform, we would want to be able to make use of it in ways we are not able to entirely pursue on our own,” she said. “And that’s where we benefit from having partnerships with in-house expertise with specific disease biologies we might not currently have.” 

Similarly to Insitro’s deal with Gilead, other partnerships could bring clinical data that would help in the development of its in vitro models. 

“So we’re not just creating models in a vacuum, we will be able to develop models and test their ability to make predictions based on the clinical outcomes of real patients,” Koller said.