Leveraging AI in healthcare is tough business, especially in the high-stakes world of oncology. Biotech startup Valar Labs is aiming high but starting small with a tool that could accurately predict certain treatment outcomes and save patients valuable time. The company has raised $22 million to expand into new cancers and treatments.
Every cancer is different, and for many cancers best practices have been established after years of testing, but sometimes a particular treatment plan needs to be followed for many months to see if it is effective.
Bladder cancer is one of them, Valar's co-founders explained to TechCrunch. BCG therapy, a common first-line treatment recommended by oncologists, has about the same chance of working as flipping a coin, which is actually a pretty good chance. But wouldn't it be nice if you didn't have to flip a coin in the first place? That's the problem Valar is trying to solve.
CEO Anirudh Joshi said the teams met at Stanford University, where they were researching AI support for clinical decision-making — in other words, helping both patients and doctors decide whether to choose between two treatment options or 12.
“What we learned is that for the majority of cancer patients today, the treatment plan is really unclear,” Joshi says. “You have options, but you don't know what's going to work. You just have to try. So our idea was to make an informed decision. In bladder cancer treatment, only one in two patients responds to standard treatment. If we knew which patients would respond to which treatment, we wouldn't waste a year on treatments that don't work.”
Valar Labs co-founders (left to right) Damir Vrabac, Anirudh Joshi, and Viswesh Krishna. Image courtesy of Valar Labs
The first test they developed, called Vesta, focuses on this specific situation. And it's not just a theoretical software solution: The team worked with 12 medical centers around the world to study more than 1,000 patients to understand exactly why some patients respond to certain treatments.
This process has two components: first, a visual AI (or computer vision model) that is trained on thousands of histology images of cancer patients. Thin slices of affected tissue are increasingly scanned and inspected by experts, but the process can be somewhat rough.
“This ultra-high resolution image tells us a lot about what's going on at the cellular level in the tumor,” explains CTO Viswesh Krishna. “We run our models on this image and extract a huge number of features, similar to a genomic panel. We generate thousands of histological reads. [i.e. important image features]and then we take the most important ones that a pathologist might look at but can't really quantify. A pathologist might notice that they're different, but they can't measure those differences.”
An example of a processed histology slide. Upon closer inspection, individual features and cells can be seen outlined. Image credit: Valar Labs
Joshi is careful to add that it's not meant to replace pathologists, but rather to augment their capabilities. Think of it as a smart microscope that helps experts make precise measurements of cellular damage, immune responses and other structures that indicate how well a disease is progressing or being contained.
“At the end of the day, it's always the physician who is in charge. This is just more data, and physicians are happy about that. And by doing tests like this, it gives them a grounded, outside perspective, and patients are really happy about that,” Joshi said.
The team says the imaging component was trained on a large amount of data, making it generalizable to many fields and cancers. Counting lymphocytes in breast cancer tissue is roughly the same task as doing it in skin cancer tissue. But that number, or any other quantifiable biomarker the model can identify, indicates a patient's likelihood of responding to treatment, and is highly specific to that particular condition.
So the second element of Valar's system is something that needs to be tailored to the specific clinical situation. And to that end, the company has demonstrated that in the specific case of bladder cancer and standard treatment plans, its test predicts success much more accurately than any other metric.
Risk factors like age, health history, and smoking status play a role in predicting certain treatment outcomes, but these are “very crude,” Joshi noted. Valar claims its AI model “outperforms all of these variables.” [in predictive power]and are independent of them,” meaning that they can be used in addition to, and not just instead of, standard risk factors.
He also noted that it's important to keep results interpretable. The last thing doctors and patients need is a black box. So when it says a patient will respond well, that's backed up by evidence that “their immune system is doing thing A, their nuclei are doing thing B, etc.”
Image courtesy of Valar LabsImage courtesy of Valar Labs
Founded in 2021, the company has put a lot of effort into building the image model and the first clinical model of the aforementioned BCG therapy for bladder cancer patients. As Valar pointed out in a recent presentation, the test can identify individuals who are three times more likely to be at risk of not responding to BCG and therefore would likely be better served (by their care team) to try a different approach. Even saving a month of wasted effort could be life-changing for some people.
As anyone who has been through cancer treatment can tell you, each day of treatment is not only incredibly precious, but also difficult to come by. While Valar may not be able to provide certainty (which is nearly impossible in oncology), it could be a powerful arrow in a caregiver's quiver.
With its first product release on the horizon, Valar closed a $22 million Series A funding round led by DCVC and Andreessen Horowitz, with participation from Pear VC.
“The funding came at a perfect time,” Joshi says, “allowing us to complete this validation and the funds will help accelerate the commercialization of Vesta while we also begin to expand into other types of cancer.”
The founders said they hope to steadily expand using a commercial lab model common to genomic testing these days, with COO Damir Vrabac saying, “This is very similar to other tests we've introduced before and will not disrupt the healthcare system in any way.” The hope is that this will ultimately lower overall healthcare costs by passing the costs on to insurance companies and avoiding unnecessary and ineffective treatments.