Lunit was founded in 2013; and 9 years after its inception, it recently went public on the Korean Stock Exchange. Prior its IPO, Lunit made a track record of receiving a total of $135M in investment, from leading healthcare VCs like HealthQuest, and industry leaders such as Guardant Health and Fujifilm. With the IPO, Lunit is looking to expand the commercialization of its cancer screening products with partners such as Fujifilm and GE Healthcare, which are already being used in more than 700 international sites, along with accelerating product development for its oncology solutions in partnership with Guardant Health.

‍

Good morning Dr. Ock, it’s great to have you here today. Would you like to start by telling the readers a little bit about yourself?

First of all, I really appreciate you inviting me to speak today. This is Chan-Young Ock, Chief Medical Officer in Lunit Oncology Group. Before I joined Lunit, I had worked as a medical oncologist in Seoul National University Hospital, and also worked with lots of biopharma companies. I started my career working on several clinical trials and realized that I was interested in developing novel biomarkers and have made it my mission to guide biopharma’s drug development in a more efficient and successful manner. I joined Lunit three years ago, and it’s really been a fantastic three years so far!

 ‍

It seems that you’ve caught the start-up bug. You were the Chief Medical Officer at MedPacto before holding the same position at Lunit, and you also run your own consulting group, Bang & Ock. What inspires you to work with all these different types of organizations, especially start-ups?

As I mentioned in my personal background, I am a medical oncologist and see many oncology patients who have no standout treatment options. This inspired me to seek novel agents and biomarkers to help them. I joined MedPacto five years ago as I was interested in developing a novel agent to help patients respond to immuno-oncology treatment, especially for those who may not respond to immune checkpoint inhibitor. If we discover additional agents like the TGF-β pathway, it might become helpful to more patients. I worked at MedPacto for one year and it was a great experience, but I also realized that I needed more clinical trial experience.

So, I went back to Seoul National University Hospital to run more clinical trials and see a lot of patients. Here, I experienced more than 100 clinical trials of all phases: 1, 2, and 3. I’m not sure if I’ve necessarily caught the start-up bug; instead, I just felt that we’re in need of biomarker development through the combination of work like DNA sequencing or IHC, especially in the immuno-oncology field. Lunit AI technology at that time, three years ago, was really quite impressive and promised to solve the issue of detecting actual immune cells with spatial AI analysis, which plays a main role in immuno-oncology. So, that’s why I joined Lunit.

Another company I have is Bang & Ock Consulting, co-founded with Dr. Yung-Jue Bang, a well-known senior medical oncologist. It was built to consult other biopharma companies looking for clinical knowledge in novel direct drug development, with the engagement being part of the agreement upon joining Lunit, so there's no conflict there.

I can really feel the passion coming through your voice when you mention these different oncology biomarkers. And so, for our readers who may be less familiar with Lunit and your product offerings, can you share a little about the vision and your product line?

Lunit’s mission is to conquer cancer through AI, and we provide AI solutions for cancer screening and to optimize cancer treatment. We have Lunit INSIGHT products, focused on detecting abnormalities in Chest X-ray or breast mammography, and Lunit SCOPE, the product I am working on. Lunit SCOPE provides an AI solution to help find more patients eligible for novel cancer treatment like immuno-oncology or some targeted agent and to enable efficient diagnosis for pathologists as well. Lunit INSIGHT products are currently in commercial use in over 700 sites globally, whereas Lunit SCOPE products are still in the development phase, mainly targeting the US market in partnership with Guardant Health.  

For more detail on the Lunit SCOPE product, it is an AI algorithm on a licensed platform using various types of pathology images, such as H&E or IHC. The first and leading product Lunit SCOPE IO has been developed by using more than 16,000 whole slide images with nearly 3 million cells annotated by board certified pathologists to detect immune cells and analyze their spatial distribution in the tumor micro-environment. Based on our previous publication in Journal of Clinical Oncology this year, Lunit SCOPE IO was able to predict responders of immune checkpoint inhibitors inNSCLC.

Also, recent studies have shown that Lunit SCOPE IO works in other cancer types as well, and we are aiming to apply this Lunit SCOPE IO model as the next generation tissue agnostic biomarker of immuno-oncology, which is why we named it Lunit SCOPE IO.

Another product we have is Lunit SCOPE PD-L1 and Lunit SCOPE HER2, which are IHC products. For these products, we quantify tumor cells of the target protein expression in a very detailed manner. For example, Lunit SCOPE PD-L1 has been reported to reduce the inter-observer variation among pathologists and may rescue more patients, 50% of the PD-L1 low population, for immune checkpoint inhibitor treatment. This can be applied to breast and bladder cancer as well, and we are looking to apply this model to multiple cancer types. 

Lunit SCOPE HER2 quantifies HER2 expressing tumor cells, especially in the HER2 low population, which is a trending topic nowadays. We are trying to figure out if Lunit SCOPE HER2 can expand the indication into the HER2 ultra low population, so this is a very interesting area as well.

All these products sound very interesting and sound like they can really benefit pathologists, oncologists, and researchers. Can you tell us what the path to the clinic looks like for all these products? Because as you know, digital pathology is a very novel field and with it being a novel field, there comes many obstacles with adoption.There's low scanner penetration, skepticism from older school pathologists, and of the IT infrastructure investment that is needed in many hospitals. Some digital pathology companies such as Paige and PathAI have released their own slide viewing platforms to boost adoption of their algorithms. So, how does Lunit plan on overcoming these obstacles to reach the clinic and how do you see it developing in the next couple years?

That’s a great question. The reality is that it is still challenging and there are many barriers to work through to implement digital pathology into the routine clinical practice at this time. The main players in this field are mostly pathologists and usually have that old school skepticism. Most believe that microscopic examination is much more efficient and handier for their use, especially for routine clinical practice. Compared to digital pathology which requires more planning for budgets and logistics including scanning, visualization, and using mouse scroll - which is not handy from their end. Based on our observation, only 2 or 3 out of 10 big centers in the United States have implemented digital pathology systems; and among them, only 2 or 3 out of 10 pathologists in these institutions actually use digital pathology for their routine clinical practice.

Although there is still a long way to go, things are constantly changing and evolving. We believe that digital pathology will be eventually implemented at least for archiving the sample or core materials for future analysis, mostly in big centers and especially for oncology research. But on the other hand, we are not aiming to replace most routine diagnostic pathology right out of the gate. We believe that biomarker development is somewhat different, and that AI based digital pathology image analysis is critical, right now, for certain clinical oncology such as novel IO treatments or detecting patients in the ultra-low HER2 population.

There are close to 1 million cells in a single slide, which cannot be fully assessed by a pathologist. There are also countless possibilities of the interaction network among cells which can’t be objectively assessed by human pathologists. So, in conjunction with the novel drug development, such as an immuno-oncology agent or antibody-drug conjugate, investigators and pharmaceutical companies have been looking for more than just sequencing data to understand tumor biology itself and its association with the immune system in detail. So, in summary, we’ll say that digital pathology for routine clinical care may expand in the future, but digital pathology for novel biomarker analysis for drug development is happening right now.

It will definitely be exciting to see how the digital pathology evolves over the next few years and perhaps if any developments expedite its placement into the clinic. Lunit SCOPE PD-L1 has already received the CE-IVD mark of approval, and you mentioned that you were looking to expand into the US market. Is Lunit also looking to attain FDA approval and what is your prospective timeline for this?

Yeah, the US market will definitely be our first target market, especially with our partnership with Guardant Health. Stay tuned for more news on that soon. For FDA approval, getting FDA approval for routine clinical care is not our first priority. However, receiving FDA approval, along with novel agents in development, as a companion diagnostic (CDx) biomarker is our utmost priority!

That definitely makes sense for the short term. Alright so our last question to wrap up the digital pathology side is: Where do you see digital pathology heading in the next 10 years?

10 years is hard to predict. I’m still of the belief that even after 10 years, most big institutions will implement digital pathology systems, but will reserve their use for special cases rather than routine clinical care. For example, for assessing the tumor microenvironment in a spatial manner, or quantifying proteins like HER2 with a continuous value rather than the crude classification of 0, 1, 2, and 3 as we have now.

Moreover, it can be applied to measure efficacy of certain treatments objectively for various types of disease, like pathological complete response measurement after initial chemotherapy or for liver and skin diseases. So that kind of specific agenda might be needed to use digital pathology even in 10 years. But I believe that there’s more on the biomarker side such as the association with the lymphocytes and tumor cells, and we are still working on that.

And now we have all the data and the AI capabilities to understand what H&E analysis is telling us. So, we are working on, in depth, the biology of the tumor to understand the future risk of cancer, what kind of future risk it may pose, and how we might best target using AI based technology.

At the rate at which technology is advancing, many would have expected adoption to come sooner, but your answer really speaks volumes about the barriers to the clinic. It also serves as the perfect segue into our AI section. The most critical step in creating an effective AI is to accurately describe the problem and outline the variables of interest. Can you help our readers understand the specific problem that the Lunit SCOPE is addressing? As well as any problems you faced?

Yes, let’s start with Lunit SCOPE IO. It’s kind of impossible to assess all the immune cells manually in the spatial region of cancer. The Lunit SCOPE IO detects tumor cell lymphocytes in the tumor tissue and the stroma. This enables us to detect the immune phenotype between immune inflamed (immune cells are in close proximity to cancer), immune excluded (immune cells are unable to penetrate the tumor, but present in the stroma), or immune desert (no immune cells are present in the tumor microenvironment). This immuno-phenotyping allows us to predict the actual clinical outcome of the immune checkpoint inhibitors.  

Lunit SCOPE PD-L1 and HER2, focus on detecting detailed target expression by profiling cells as objectively and quantitatively as possible. Again, this aims to eliminate the human error of missing certain regions of protein expression that could allow for immunotherapy. We segment the cancer area to remove the first positive tumor cell outside the cancer area and enrich the tumor cancer area. Then, we calculate the proportion of PD-L1 positive tumor cells versus PD-L1 negative tumor cells for TPS scoring. We can also calculate the CPS combined positive score as well, which is the proportion between PD-L1 positive immune cells and PD-L1 negative immune cells. HER2 is similar and we classify tumor cells based on a 0-3 positivity scale and can calculate the proportion in each phenotype.

With regards to challenges, there are a lot. Quality control of the slide image is one of the most important. Inappropriate quality during staining and scanning the slide may impact AI analysis much more than it would impact human pathology's examination. AI is only able to comprehend what it has been trained to see, and any outlier patterns or rare types of cancer cells may confuse an AI. To overcome this, we have to continuously train our AI to tackle specifically hard cases with poor image preparation or feed more data with rare types of cancer, like neuroendocrine carcinoma, to boost the performance of the AI. Another challenge is that they only give us a small size of the biopsy tissue which means we may only derive useless information without representing the driver mutations and there could be intra- and inter-tumoral heterogeneity issues. So, these innate issues should be resolved by multimodal approaches, not just pathology. For example, radiology image analysis can help us understand more.

These are very important challenges to consider, and it looks like Lunit has plans to overcome these challenges. I recently read a scientific paper that said the Lunit SCOPE PD-L1 identified 50% more patients that were eligible for immunotherapy, compared to pathologist screening, while some competitors can only boast of around 30% increase. So, we wanted to hear from you, what steps have you taken to ensure that your algorithm is the most effective and how you know that this is not a problem of overfitting?

So, claiming how much an AI model can expand a certain indication is mostly dependent on what specific indication or market they focus on and not just the maturity of the technology. We and our competitors have been very positively benefiting each other to progress the field of AI in digital pathology.

Additionally, we know that this is not a problem of overfitting because the algorithm identifies certain areas of the cancer that the pathologists might have missed. Pathologists may only focus on certain areas of the slide, but an AI very diligently understands the entirety of the slide images so they can capture PD-L1 positive tumor cells outside of the field view of a pathologist. Anyhow, we are always aware of the possibility of overfitting for any AI model, therefore we usually collect the data from multiple data sources and validate it using as many different data sources as possible.

Yeah, that’s a wonderful and diplomatic answer. And that wraps up our AI section as well, too. Again, I want to congratulate you on your recent IPO. That's a huge step to take as a company and I wanted to see if you had any advice for entrepreneurs that are looking to start their own venture, or others who are currently running their start-ups and looking to take their companies public.

Thank you. However, we think that the IPO is just one of the many starting points on the way to our final goal to conquer cancer through AI. The company was founded almost 10 years ago, but I joined Lunit three years ago. I’m not a founding member, but I believe one of the keys to our current success so far was that we have been focused on a very specific disease of high human needs, cancer.

‍

I would agree. It’s very important not to lose sight of your mission after you IPO. Before we conclude, do you have any final thoughts or comments on anything that you'd like to speak about that we haven't had a chance to touch on yet?

 Nothing from my end at this time, but I really appreciate that you invited us for this interview, and please reach out if you have any further questions!

Thank you so much, Dr. Ock for your time, this interview has really been a pleasure and will be very informative to our audience.

‍

Comments and opinions expressed by interviewees are their own and do not represent or reflect the opinions, policies, or positions of DeciBio Consulting or have its endorsement. Note: DeciBio Consulting, its employees or owners, or our guests may hold assets discussed in this article/episode. This article/blog/episode does not provide investment advice, and is intended for informational and entertainment purposes only. You should do your own research and make your own independent decisions when considering any financial transactions.