DeciBio’s Peripheral Biomarkers Q&A with Dr. Ofer Sharon of OncoHost

Original source here.

OncoHost recently raised a $35M Series C funding round led by ALIVE Israel HealthTech VC with participation from other investors such as Leumi Partners and Menora Mivtachim. This funding will help expand OncoHost’s ongoing multicenter prospective PROPHETIC trial and aid cin PROphet®’s U.S. commercial market launch anticipated later this year.

Ofer Sharon, MD, MBA, is CEO and co-founder of OncoHost, a host response profiling company in the peripheral biomarkers space. A physician by training, Dr. Sharon leverages his previous experience in pharma drug development and entrepreneurship to guide OncoHost through the development of PROphet® and an ongoing clinical trial. Prior experiences include co-founding mental health startup Valera Health, serving as a Medical Director at Merck and AstraZeneca, and acting as New Technologies Scout at MedImmune.

Thank you so much for joining us today, Dr. Sharon. For our readers who might be less familiar with PROphet® and OncoHost, would you mind sharing a bit about your company, your product, and the vision?

OncoHost was founded at the end of 2017, with headquarters in Israel, north of Tel Aviv, and a CLIA-registered laboratory in Cary, North Carolina. It started in the lab of Professor Yuval Shaked, head of the Technion Integrated Cancer Research Center at the Israeli Institute of Technology. For the past 15 years, Yuval has been studying the concept of host response—the physiological reaction of the body to anti-cancer treatment—in oncology. What he uncovered during the studies, which other investigators have also concluded, is that host response plays a critical role in tumor resistance to treatment. In many cases it may even facilitate tumor growth and progression.

We analyze protein patterns in the plasma and combine this analysis with bioinformatics and machine learning to address three important clinical questions:

1. First, is the question of response: will the patient respond to treatment?
2. Next, why does resistance occur? The resistance mechanisms involve a complex interplay between the host, the tumor, and the therapy.
3. Finally, what is the next step in treatment, and what kind of combination strategy could be taken in order to mitigate those resistance pathways? The tool is designed for clinicians and, obviously, the patient. At this stage, this is not a tool for drug discovery or combination discovery. The idea is to support clinicians in their complex decision-making regarding patient management.

The first market we are targeting is late-stage immunotherapy-treated cancer patients, and the first indication we are tackling is non-small cell lung cancer.

It sounds like OncoHost has great momentum. Digging a little deeper, in regards to vision, what is the role you see OncoHost playing in clinical care overall?

‍I, and others, believe that precision medicine should be the next step in oncology. We’ve reached the point where we have a lot of excellent tools in our hands. Obviously, more tools are needed and more will be developed, but we understand today that no two patients are the same, despite very similar clinical characteristics. We need to start thinking of each patient as an individual and much more deeply analyze the resistance pathways and mechanisms that are relevant for each patient.

To gain a deeper understanding of cancer and move to the next stage of cancer therapy, we need to rethink the approach to precision medicine. The precision medicine pioneers focus on the interactions between the therapies and the tumor, which makes a lot of sense when looking for targets. It also makes some sense when looking for biomarkers, but misses a very big chunk of the picture when they are not taking into consideration that the interaction between therapy and tumor takes place in a very complex biological system that is the human being.

OncoHost sits at a very interesting intersection, observing the interplay between three elements: tumor, patient, and therapy. We do this on a proteomic level, which allows us to get a downstream view of the biological processes that are taking place. Having said that, I strongly believe that proteomics alone doesn’t tell the full story. Moving forward, we will develop the company along two main roads: the first is developing a product—expanding to more indications and treatment modalities, looking at the earlier stages of disease, which is the classic way you would expect a company like us to develop. The other road is to add more capabilities to the proteomic platform, including tumor and tumor-related biomarkers, and microbiome and single-cell analysis technologies.

If we look further down the road, maybe five years from now, the way we understand cancer and the way we look at the patient is going to be very different from today. First, it’s not necessarily going to be indication-specific. Rather than looking at a patient as “has non-small cell lung cancer”, we might look at them as a patient with X, Y, Z mutations whose immune system might react to anti-PD-L1 or a combination of other targeted therapies, based on personalized analysis of resistance mechanisms. In order to do that, we first need to understand that we are looking at the continuum; cancer is not a one-time event, and the patient who starts the treatment on day one is not going to be the same patient after 2 weeks, 2 months, and 4 months. Things change, cancer is dynamic, and we need a way not only to diagnose and to understand which treatment is correct as a first-line, but also to monitor. So for OncoHost, that means there’s going to be a lot of activity in the near future, both on a clinical development level (with the clinical trials that we are conducting), and on a business development level as we add more tools to the platform to enable a wider view of the disease.

Absolutely, very well-said. It definitely does make sense to take into consideration not only the tumor environment but also the host characteristics when looking at treatment, and I’m very excited to see how these clinical trials continue to develop. I see that you’re also the founder of other startups such as Valera Health. What inspired you to pivot from your time and experience within the industry, within pharmaceuticals, to biotech entrepreneurship?

Well, I’m not sure that I pivoted from pharma, necessarily. I think that pharma, for me, was a place where I got the chance to stop, think, and gain tools that you don’t get as a young biotech entrepreneur: understanding the processes, strategy, and complexity of drug development. So I don’t see myself as a corporate guy that became an entrepreneur; I see myself as an entrepreneur trying to work in the corporate space for some time. My real passion in life is in building companies. I think this experience, the understanding of the market, the strategy, the capabilities, what is needed to develop a drug and market it, can only be learned by working in pharma; they are the experts, after all.

Definitely. It sounds like your end goal was always to build your own companies, and you needed a place to gather experience and insights early on.

Yes, and I also happened to meet my co-founder for Valera at Merck—he was my colleague, and we used to work together, so I think that was a win-win for the pharma company when I left. I wasn’t a great corporate guy, and I got my experience before moving on to the biotech industry.

That certainly sounds like a win-win. Pivoting a bit to the peripheral biomarkers space. Overall, where do you think the peripheral biomarkers industry currently is, and how do you envision OncoHost helping make the next big advancements in the space?

I don’t think we’re even scratching the surface in terms of what is needed and what can be done with biomarkers, and there are many reasons for that. Moving forward, I think that we are looking at two big challenges.

One challenge is the pharma companies: pharma companies are trying to develop drugs that follow the paradigm of one size fits all. If you look at the inclusion criteria of most Phase III clinical trials for immunotherapy, you will see that you can include patients from the age of 18 to the age of, I don’t know, 65 or 70. An 18-year-old and a 65-year-old are not the same person, and they are not going to respond the same way to drugs. If you look at the success rate of clinical trials, we see that the failure rate is over 80%. Only a week ago, I think, the Phase III anti-TIGIT drug tiragolumab failed. Those Phase III trials are not failing because the drugs are ineffective; they are failing because we are not finding the right patients for those drugs. It’s going to be a major challenge to change this one size fits all approach, but tying yourself to a niche group of patients is a problem in the long-term.

The second challenge is the fact that clinicians and oncologists are used to fighting cancer like it’s the last fight, which in many cases, unfortunately, is true. This means that you’re willing to accept low response rates in the hopes that it will extend your patient’s life, so we get used to celebrating successes by measuring overall survival and progression-free survival (PFS), even if they’re only marginal, but we don’t really talk about patient quality of life and extending life without adverse events. We need to look at response rates; it doesn’t make sense to develop drugs that only fit 15-20% of the patients—what about the rest? I think those are the two major challenges that the biomarker industry is going to face.

At OncoHost, I’m trying to find the value proposition for clinicians so I can say, “This is how we’re going to enhance your daily decision-making.” When we first started, we could have gone the more common route and worked with biopharma to earn initial revenues, but instead, I went a different direction. I am building a tool for clinicians to use as another layer of information when making critical decisions regarding patient treatment.

So it seems like a lot of companies are trying to position themselves more upstream of the clinic, like partnering with pharma, but you see opportunity working with clinicians to tailor and assess the treatment plan. What differences in OncoHost’s PROphet will drive interest and adoption over competitors’ solutions?

We took the long way in many ways, taking time and care in the way we approach the data. First, we are running an ongoing prospective, multi-centered clinical trial which means we are paying with time and taking on a significant financial and clinical risk. In this prospective trial, we are following up with patients for up to two years, and an additional trial we are going to open soon will follow patients up to 5 years. The risk and time are worth it to us because we think the quality of data from a prospective trial provides a level or reassurance that is different from that of companies mainly working with blood specimen banks and retrospective samples.

Second, I would say that OncoHost is not an AI company; OncoHost is a company that deals with biology and cancer mechanisms and happens to use AI as one of the tools to develop the products. This means that in our reports, there will be nothing that cannot be explained biologically. When you’re working with data and deploying computational tools, you can get signals that may look amazing, but they’re not related to biology in any way. False discoveries are a huge caveat for “AI” companies. If we want clinicians to believe us, and to be able to understand what we do, we need to bridge this gap between mathematics and biology.

As a biology company, we can understand and explain why we present a certain combination, why we present a certain resistance pathway, and what’s going on there. This is not an AI black box; everything is completely transparent. Based on what I’ve seen in publications and abstracts of companies, this is a big differentiator. Another differentiator is the fact that our published results will always be based on blinded validation. Let me explain what I mean: in many cases, we can see excellent results and excellent sensitivity and specificity, but when you read the fine print, you will see that the validation was done using methods like “leave-one-out” or “cross validation”, so you’re essentially training and validating your algorithm on the same cohort of patients. It’s no wonder the results are excellent.

On the other hand, when we are presenting data, it’s valid validation. It's a “one touch one go” experience: there’s a cohort of patients that is locked to our R&D team. We don’t know what the outcomes of the patients are when we train the algorithm. Once the team tells us they’re ready to validate it, we open the database, and they run the algorithm on the database’s cohort of patients. Whatever the outcome is are the final results. There’s no second chance here. And in that sense, I think this improves the quality of data. This is complicated to explain to investors and clinicians because true blinded validation first of all requires a lot of patience and is kind of risky. We are willing to pay that price because we’re trying to answer important questions that may affect clinical decision-making. I am happy to report that the clinical results we have at this stage are much better than expected. The signal is very powerful and based on true blinded validation.

Thank you for that explanation, and I love that you said that OncoHost is not an AI company, it is a biology company because at the end of the day, AI is a technology that helps us solve problems in the clinic. Could you explain to our readers how exactly OncoHost is leveraging AI in PROphet®?

Yes, but maybe I’ll start with the problem first. When you’re trying to use machine learning to train an algorithm, you have two major risks to avoid. First is the risk of overfitting, which we just discussed—training and validating your algorithm on the same population. We avoid this by running prospective clinical trials with blinded validation.

The other big issue is a little bit more complicated to explain, but I’ll try. When you’re developing an AI-based product and looking for features that will differentiate between two groups of patients—in our case, the responders and non-responders—many companies including us, initially, try to look for signatures, a group of features that may be genes, proteins, transcriptome, or a combination of those, that can differentiate between those groups of patients. The thing is, if you try to do that with a relatively small number of patients, you may find a signature with several features that provides you with excellent results, but when you bring in a new cohort of patients, you will find a completely different signature. So which signature is the right one? If you look at what is being published, you will see that many people are looking for signatures out of a dataset of hundreds, sometimes thousands, of features. In those cases, the odds of what we call “false discovery” are very high.

We decided to develop a completely different framework to solve that: minimize feature selection and look for “universal” signs of resistance in the protein patterns. Rather than looking for a specific signature, we are measuring the level of biological “noise” in the patient’s body. This allows us to create a tool that does not suffer from this issue of false discovery, which is a big problem in AI. This proprietary approach allowed us to create a tool that does not suffer from false discovery, a big problem in AI, and is also accurate, precise, and reproducible.

Thank you for that. I hear you, I know that overfitting is a big problem in healthcare AI in general because it tends to introduce a lot of bias into the AI system, so it’s very interesting to hear how you are staying aware of it and actively trying to mitigate it. I wanted to shift gears back to the peripheral biomarker space and probe a bit more about the technologies. In your publication in The Journal for ImmunoTherapy of Cancer, we see that you use an ELISA-based array (SomaScan) for proteomic profiling; however, your company has also mentioned a mass spec-based proteomic profiling technology  (PROMIS-Quan). Could you elaborate on OncoHost’s use of ELISA vs mass spec? Do you anticipate OncoHost adopting both ELISA and mass spec in its analytical technology, and your general thoughts?

We are seeing many companies now that are developing proteomic profiling tools, sometimes combining mass spec with other capabilities. For me, the idea was to find a partner to collaborate with that could provide us with this panel almost as a commodity. Though mass spec is very interesting and part of our research, technical challenges prevent us from using it commercially today. However, looking into the future, it is something I definitely would like to see incorporated into the platform.

Today, we run our analysis and get results from SomaLogic’s SomaScan platform, which we think is an excellent platform in terms of accuracy and reproducibility. We did a lot of validation there and also found partners on the other side to work with us on development. We are very excited about this collaboration and what it holds for us in the future.

That makes sense. Quick follow-up, when I read the paper, I saw that SomaScan is capable of reading around 7,000 proteomic markers. But, and correct me if I’m wrong, the paper said there were 3 specific signatures of interest that had been associated with host response profiling, so I was wondering if you think that PROphet® will only be used with such high plex technologies, or if you also anticipate it being used with lower plex technologies that may be cheaper.

Good question. For now, the objective for the company is reaching the maximum accuracy and precision that we need to provide value for our customers. This is the main focus for the company. With that said, we see a lot of potential for lower plex technologies for a pipeline product that will enable a PROphet®-based monitoring tool that will require serial testing.

At the end of the day, combining different capabilities and gaining a better understanding of cancer is great, but we also need to think about cost. I think that using technologies that are lower cost is key and part of our strategy.

Cost can definitely be easier to optimize later, once the technology is more developed. We read that in the Times of Israel that your platform will also be sent into space to monitor astronauts’ immune systems. Would you mind telling us a bit more about that, and what you would like to achieve?

It actually came as an initiative from Professor Lawrence at Sheba Medical Center, the largest hospital in Israel. The idea is fascinating, but it is an “N” of 1, so we will analyze and present the results but need to be cautious with their interpretation. The astronauts are back; we are now waiting for the blood samples to arrive. The idea is to test the impact of microgravity on the immune system, so out of those 7,000 proteins that we are measuring, we are focusing on those that are related to immune modulation to see what the change is before and after the flight. We are looking at other parameters as well. It’s scientific curiosity, and though it’s fun to participate, we have to remember the sample size is one, so we’ll have to see.

Excellent, real exciting stuff. Out of curiosity, was this one of the use cases that you anticipated, or did this come out of left field?

No, this is not going to be a fun answer, but the way I chose to move this company is very clear: to benefit this triad of patients, payers, and oncologists. When I used to work for pharma, I spoke to many patients, and I saw how much guesswork, frustration, and fear there is. Taking some of those uncertainties out of the mix is an important goal for me. I didn’t think of space; I thought only of cancer, plain and simple.

That passion for patient care is really admirable. Before we wrap up, do you have any last thoughts to share with the readers?

I have many, but I think that specifically for precision medicine, I think that it’s super important to do two things. First is to build confidence with the clinician community in what is being developed and how it’s being tested because their acceptance of it is key. Otherwise, we’re going to find ourselves for many years to come looking for specific drugs for specific targets, only to discover that they are only relevant to a very small number of patients. I think it’s key to develop a regulatory-approved tool and help stakeholders get access to it sooner rather than later.

And second, to the many companies that are out there that are trying to develop a product with a very specific angle in the biomarker and precision medicine industry: I urge them to come and talk to us or to others to try and think about vertical integration, because we need to look at this story with a much more holistic approach than we do now. We shouldn’t make the same mistakes we did with drug development, developing a drug for one single target; we need to take a holistic approach. So to the readers, I would say, look at consolidation, look at combinations of technologies in order to gain a better understanding of cancer.

Very well-put. We’re very excited to see where precision medicine and the peripheral biomarkers space takes us next, and how technologies and medicine will take a more holistic approach to treatment of the patient. Thank you so much for taking the time to chat with us today; it’s been a pleasure.