AI represents a next iteration of this beautiful analogy—a new superpowered set of ‘bicycles’ for the mind. There is perhaps nowhere more exciting or impactful to build such tools than in healthcare—where progress is often so constrained by human expertise.
Valar Labs is taking on exactly this challenge, and bringing AI into the toolbox for better cancer care. And we are thrilled to announce today that we have co-led Valar’s Seed and Series A financings, alongside partners at Pear VC and DCVC.
AI in healthcare: bicycles for expert minds
Where will AI impact healthcare first? Can startups build bicycles for healthcare minds? Which bicycles are worth building? Our team at a16z bio + health has outlined several frameworks for thinking about AI applications in healthcare (here, here, here), and one I reference frequently is a simple 2×2 diagram of tasks that are easy vs hard for humans, and easy vs hard for AI. Innovation in all quadrants can drive impact, but the top quadrant is what I’ve called the ‘magic’ one, where hard-to-train AI (read: AI with a moat) can help humans achieve tasks that we could simply never do on our own (even if we had infinite time). This is exactly where we believe Valar Labs’ AI diagnostic products sit.
Software as a cancer diagnostic
There are nearly ~2M new cancer diagnoses annually in the US, many more globally. In order to make every single one of these diagnoses, a biopsy of the patient’s tumor is laid onto a slide, stained in standard ways, and histologically classified as cancer upon review under a microscope by a pathologist. During the COVID pandemic, a huge number of these slides were digitized—for virtual review by pathologists. This means that every cancer patient already has images of their tumor histology, well preserved in their medical records and typically readily accessible (often fully digitized!) for their care team.
After a cancer diagnosis is established (and molecular testing is performed), the next step is for oncology care teams to recommend treatment options. To select a therapy, oncologists rely on clinical trial results and treatment guidelines, which typically provide data on an overall response rate (ORR), or the fraction of all patients with a given cancer diagnosis who respond to a specific therapy. But this data is still hard to translate to individual patients: for a therapy approved with an ORR of 35%, there are unfortunately two patients who won’t respond for every one patient who will. Identifying the patients who are unlikely to respond could save them valuable time (to try other therapies)—and spare them harsh side effects.
Let’s go back to those histology images that every patient already has on file…what if these same histology images contained far more information than what they are telling us today? What if the slides have hidden insights buried within them, invisible to human pathologists, but accessible to AI? What if these signatures could be used to predict how a patient’s disease would progress? Or whether he or she would respond well to a particular therapeutic regimen?
To answer these questions, the Valar team has partnered with oncologists, cancer centers, and research teams all over the world to build several large registries of thousands of patient specimen images, carefully paired with clinical response data. They have rigorously trained AI algorithms on these retrospective data—and published results showing that AI can indeed learn what humans simply cannot see. Valar’s AI diagnostic tests can predict how aggressive a bladder cancer patient’s individual cancer will be—or whether a pancreatic cancer patient is likely to respond to the FOLFIRINOX treatment regimen—or if an ovarian cancer patient is a high- vs low-likelihood responder to platinum-based chemotherapy.
The strength of these early data have made clinical research partners excited to engage in an early access program for Valar’s first clinical-grade, CLIA-validated diagnostic product (Vesta) in bladder cancer. There is still a lot more to learn, but we’ve heard the feedback again and again from oncologists: if this information was knowable, every patient and provider would want to know.
The Valar team: moving code to clinic, fast
The Valar co-founders—Anirudh Joshi, Damir Vrabac, and Viswesh Krishna—met as researchers in Andrew Ng’s AI Lab at Stanford. They were intensely motivated to use AI to do something big for patients, and were moving fast to learn everything they could about cancer care.
Perhaps even more impressive than their technical speed has been their extraordinary depth of understanding of the clinical scenarios that they are trying to impact. To earn this understanding, they have cold-emailed and ultimately mind-melded with dozens of oncologists and pathologists. They have spoken to biopharma companies developing new therapies. They have built connections with regulatory experts in the diagnostics and reimbursement world. They have become fixtures at national oncology conferences, presenting data and collecting feedback.
One leading oncologist we spoke with said he loved collaborating with the Valar team because of their “complete humility and willingness to learn about what we do clinically.” These are rare traits in a team that is also so cutting-edge in their own AI, software, and execution capabilities. And these are exactly the combination of skills that will be required to develop, validate, distribute, and scale the best new ‘bicycles’ to every oncology mind, and to every care team in the world.
Join the mission—Valar is hiring!
