Last month, Dr. Elizabeth O’Day of Waltham, MA-based precision medicine startup Olaris, Inc. was kind enough to visit DRG’s Burlington, MA office to kick off our speaker series on the future of healthcare. We caught up with her afterwards to talk about the exciting work her company is doing using machine learning to identify new biomarkers for drug responsiveness in the metabolome:
Q: Where are we in terms of the maturation or development of precision medicine? What challenges do we need to overcome to make it really work?
Dr. O’Day: I think we’re actually on the cusp of making precision medicine real. The term’s been around for quite some time and I think we might have gotten a little ahead of ourselves because we saw a technology that could have impact. And over the past 5, 10 years we’ve been working to perfect that. You know, it’s all about getting the right drug to the right patient at the right time. We’ve made some strides there, primarily based on genomics profiling and stratifying patients based on different genetic mutations and risk groups, and then treating people accordingly. But we now have the ability to take that even further, to go end-to-end. Based on your molecular profile, we can say: here are the diseases you’re most likely to get, and we can monitor for them, and then if you co develop these diseases, we can tailor your treatment protocol for optimal outcomes. I think in the next ten years, what we all think of precision medicine as will actually become a reality.
Q: So preventive health merges with genomics.
Dr. O’Day: Exactly. Genomics is a great first step, and should be the default in treatment these days -- particularly for cancer. If you are diagnosed with cancer, you should get your tumor sequenced. That can help narrow the field of different treatment options available. If there are a hundred drugs, that maybe narrows it down to 50 or 20.
But that’s still not good enough. We need to be able to know what the best treatment is, not to just use trial and error with a couple of drugs. And that’s where the additional biomarkers, going beyond genomics, like proteins or metabolites, that’s where we can get additional insight and value to optimize the drug treatment paradigm.
Q: How do these non-genomic biomarkers interact with these treatments?
Dr. O’Day: Traditional genomics is only measuring a tiny subset of the 2 to 3 billion pairs of genes in our DNA right now. We don’t yet know how all of those things interact, so that having or not having a particular genetic mutation just tells us something about the risk that something will or won’t work for you. There are a lot more pieces to that puzzle that we need to solve on the genomic level, and then you have to input everything else – your age, your diet, your environment, what other drugs you might be taking, what other diseases you have. All of that data or information is going to impact whether a particular drug works for you or not.
The reason I’m so bullish about metabolites is because they are the closest readout to phenotype, so they become kind of a fingerprint for all those biomarkers influencing health – from the microbiome, what a person’s been eating, any of the environmental things that they might have been exposed to. All of that info gets captured in the metabolome. Coupled with really advanced analytical and machine learning methodologies, we’re now able to mine this data for patterns that can tell you, instead of, here’s 20 drugs you could try, here’s the best drug for you at this moment in time. And that’s going to change and we might need to update that as you go through treatment. That’s the beauty of metabolites in comparison to genomics, which is a static readout of your genetic mutations. Metabolites are constantly changing, and so we’re able to catch it as diseases evolve, and treatments start working or stop working.
Q: How do you figure out who’s going to be nonresponsive to a particular treatment? How do you identify the biomarker?
Dr. O’Day: We don’t know, and that’s been a sticking point. Previously, we’ve been focusing on a really small set of biomarkers to inform clinical decisions. Some of them have been incredibly useful, as glucose is to predict and monitor diabetes progression. There are other biomarkers out there that we haven’t discovered yet. And that’s what Olaris is all about, doing discovery to find these biomarkers and then validating them to guide treatment. Our methodology is simple: We take blood samples or urine samples from patients previously treated with a drug; Some patients will do well and some will do poorly, and we try to learn from that so that every patient helps inform treatment of the next generation of patients. It’s pattern recognition. We find patterns of metabolites in the people that did well that are different from those of the people that did not do well. Facebook could tell us apart if we from a picture of us. It’s not that I have a nose and you don’t, but our noses have different ratios. Machine learning can find these different ratios of metabolites and help us identify people most likely to respond -- and perhaps even more importantly, people who are not likely to respond. That can change the treatment paradigm.
We’re doing a new project for metastatic GIST, gastrointestinal stromal tumors. And if you get diagnosed with this disease, you first start on imatinib, and you take that until it stops working, and if that doesn’t work you take another drug called sunitinib, and when that doesn’t work you take yet another drug called regorafenib. You have to march through these different drug treatment options. Imagine you knew that these first line and second line treatments didn’t work for you, that really the best treatment was the third line one. You could skip all that ineffective treatment, which was not stopping your disease from progressing, and you can intervene when the disease is still treatable or curable while not exposing yourself to unnecessary side effects for ineffective drugs that cost a lot of money. We imagine a world where you skip all of that and just say, okay, Drug C is the best one for me, let’s get on it right away. And then maybe once that drug stops working -- because all drugs, especially for cancer, eventually stop working – maybe then, I go from C to A, so we optimize the treatment path for each individual.
Q: So that could be the solution to our cost problem too.
Dr. O’Day: I completely support rewarding innovation. Drugs that are transforming lives should be financially remunerated. I’m cool with that. What I’m not cool with is developing these drugs that only work for a fraction of patients and then not figuring out who should get them. If we keep moving in this direction the system will break, and we’ll have to transition to value-based pricing or contracting. I think we’re already starting to see that come together. Pharma companies want their drugs to work, and now that the diagnostics and other technologies are picking up I think there’s a beautiful opportunity for these things to come together.
Q: What disease targets are you working on, and which categories does this technology have the greatest potential for?
Dr. O’Day: We’re focused on metastatic breast cancer as a starting place. With metastatic disease, you only get one shot to stop or curb the spread of the metastasis, and this is where I think our technology can add a lot of value. I was just in San Antonio for a big breast cancer meeting and there was a patient there who had metastatic breast cancer. She’s been in a trial for some of the drugs that we’ve done studies on. And she was begging me to get access to our test, but it’s not ready for prime time yet. We’re still doing validation. And she’s like, please, you know, I’m 38 years old, I have two kids, I just want to see them grow up. She gave me a bracelet I’ll probably wear for the rest of my life. These people just need time and there are drugs that can help them and we have a solution to help optimize that. So that’s why we’re focused there.
But the technology is agnostic to disease or drug, so we’re looking to expand in other areas where we might add value. The second disease indication that we’ve started working on is neurodegeneration, particularly Parkinson’s. We’ve partnered with biopharmas and now we’re working with the Michael J. Fox Foundation to see if we can uncover a diagnostic for Parkinson’s, because right now patients spend years trying to figure out what’s wrong with them, and often ten years passes before clinical symptoms develop. So we need to be able to diagnose and intervene earlier to give these drugs a chance to work.
Q: So that’s a case where a diagnosis is entirely symptomatic, there’s no known biomarker.
Dr. O’Day: Exactly, so you rule out a bunch of other diseases first. A lot of these neurodegernative diseases are quite subjective or only really diagnosable post-mortem, so there’s a great opportunity for metabolomics here to play a role in figuring out what you have and how to stop it in a timely manner.
Q: Are you talking to pharmas at all?
Dr. O’Day: We have both internal projects and ones we’re partnered with pharma on. That’s a big part of our business. The world we want to create is one where every new drug that comes to market has one of these biomarkers of response on the label, so that you can tell who the drug works for and doesn’t work for. We have to work with pharma to make that vision a reality. My goal is to partner with every major pharma company and let them access our technology to accelerate their drug development process.
Q: Have you engaged payers?
Dr. O’Day: We’re trying to be very proactive with payers. You know, you couldn’t design something more complicated than the current healthcare system in the U.S. It has pitted people who share the same ultimate goal of delivering the best treatments to patients against one another. Innovators and insurers are too often pit against one another, where the innovator has developed some cool technology and wants to get paid for it, but the payer doesn’t know how to value it. We have a partnership underway with a national payer and a big regional payer, who are together trying to see what value our test could have for their members and then use that to drive value-based pricing for diagnostics. We want to be able to say, ‘Hey, we did the study and we know our test can save you and your members $30 million a year on MS prescriptions.’ We’re trying to work creatively with payers now and build communication earlier on.
Q: How do you use machine learning in your work?
Dr. O’Day: Machine learning or AI is a popular term, but these technologies have been around forever. They’re just mathematical algorithms that take historical data and look for classifiers that let you distinguish two groups. So I have a dog, his name is Pythagoras -- I call him Pi for short, because math jokes are my jam. He’s the love of my life. He’s half French bulldog, half Boston terrier. He has a little black dot on his head. He’s about 30 pounds and he’s black and white. I could create an algorithm that used those features, like black on the head, some rough weight ratio, or comparing black to whiteness, and then apply this model to all photos of dogs and be able to pick out Pythagoras from other photos. That’s how facial recognition works and that’s exactly like what we’re doing for responders and non-responders, finding these classifiers in patients who have gone through treatment and being able to predict responsiveness to a drug.
Q: How do you get the training data?
Dr. O’Day: Through our amazing collaborators. That’s been the biggest bottleneck in our business. I had to convince some doctors to share patient samples with us to do a collaboration, show that it worked and then that led to other collaborations with access to other samples. But that’s the key to all of this, and the training data that these projects are built on will determine the clinical utility or usefulness or validity of these products. That woman in San Antonio, I don’t want to take her blood and just run it through our test because our test is not validated yet. We’ve done a couple of very small training sets and we’re in the process of doing a very large validation study. If that works, I will have confidence in our machine learning algorithms and in our model to predict for a patient. But until we have that data set it’s not right for me to make those decisions.
Q: What a gut punch.
Dr. O’Day: She’s been in my dreams every night since then. We’re super lucky to do this, but these are real people. Our technology, our science, our dedication could save lives, let them spend more time with their kids. That’s not lost on my and that’s why I don’t take vacations.
Dr. O’Day: Yeah, I think so. To be quite fair, as Olaris has picked up, those other things have certainly had to be toned down, but I remain very passionate and committed to them. We need to educate people more about how medicine works and the role they can play, so I’m very committed to any engagement opportunities and helping people understand how they can take ownership of their health. That would improve outcomes in and of itself. And it’s a nice feed forward loop of helping people to do what’s good.
Q: You attended [then-Vice President Biden’s] Moonshot Summit. Did you get to meet him?
Dr. O’Day: I did meet him and (his wife) Jill a couple times. He’s a good guy. He’s really passionate. The Biden Cancer Initiative is no longer active but it was a great thing that brought together the different stakeholders in cancer. And we had some really honest conversation about why things work or don’t work.
Q: How has your estimation of the potential for curing cancer changed over the intervening three years?
Dr. O’Day: My brother had cancer when we were kids. He’s fine. But since I was a little girl, I’ve been saying I was going to cure cancer. It’s the only thing I’ve ever wanted to do. And over time, my understanding of what curing cancer will look like has evolved to changing cancer from this very scary potential death sentence to us realizing that, okay, as we age, this genomic instability happens, and cancer’s likely going to develop, but we can be smart about it and stay one step ahead of the cancer by cycling in different treatments as they’re needed and not treating when we don’t need to, and by being able to monitor treatment so we can optimize it. I think of it as increasing mileage, so that patients can just keep living their lives. Cancer will be in the background, but they don’t have to be as scared or uncertain that their treatments are working or not.