Niven Narain and how AI and Machine Learning Are Changing Drug Discovery
Harry interviews Niven Narain, the co-founder, president and CEO of Berg, a Boston-based biopharma company driving the next generation of drugs and diagnostics by combining patient-driven biology and AI to unravel actionable disease insight. Narain has overseen development of Berg’s clinical stage assets and pipeline and forged strategic partnerships with industry academia and US and UK governments. He says Berg’s philosophy is to combine a systems biology architecture with patients’ demographic data and clinical outcome data, and then apply Bayesian artificial intelligence algorithms to drive better understanding of diseases.
Harry Glorikian: Welcome to the Moneyball medicine podcast I’m your host Harry Glorikian. This series is all about the data-driven transformation of the healthcare and life sciences landscape. Each episode we dive deep through one-on-one interviews with leaders in the new cost-conscious value-based healthcare economy. We look at the challenges and opportunities they’re facing and their predictions for the years to come.
Okay welcome to another edition of Moneyball Medicine. Today I have Niven Narayan who is co-founder president and CEO of Berg, a Boston-based biopharma company driving the next generation of drugs and Diagnostics by combining patient driven biology and artificial intelligence to unravel actionable disease insight. He has overseen development of Berg’s clinical stage assets and pipeline and forged strategic partners with industry academia and US and UK government’s.
Niven is most passionate about improving patient care and enabling increased access to innovative medicines to improve healthcare outcomes.
Niven welcome to Moneyball Medicine podcast, it’s great to spend time together again.
Niven Narain: It’s great to be on again, Harry, it’s always good to catch up and I think it’s such an important continuous dialogue you know given how quickly technology is moving in healthcare. So, again happy to be on.
Harry Glorikian: I had the pleasure of learning about Berg and coming in and taking a look at your systems and being brought up to speed, on what you guys are doing during the writing of Moneyball Medicine. But since then you know and maybe for the people listening for the first time and who don’t know the company. Can you tell me a little bit about you know this whole concept that you have of a artificial-intelligence, drug discovery model engine and where we were back what two plus years ago and where you are now?
Niven Narain: Yes, sure you know, so the company was really founded on this the philosophy that we should at this point in developed and this is about ten years back. We took a good hard look of how could we use biology in a more fundamental sense to drive a greater understanding of diseases. But importantly how our disease is different than a healthy, an otherwise healthy individual or a healthy cell or a healthy tissue. And the approach that we took at that time was really to combine a systems biology architecture with a combination of a patient’s demographic data, their clinical outcome data.
And then we wanted to look at a novel way of how do we analyze this data, because obviously this is in the late 2000s, you know early 2010’s. And our decision at that point was to take an agnostic approach to not bias ourselves by what was known already, so looking for example that you know Jiwa studies and the to known or traditional pathways. And our approach is really to bring a new data topology and new data ecosystem together, where one could look at genes and proteins and demographics and a patient’s, clinical story overall and then feed this data architecture into a Bayesian artificial intelligence system.
And this Bayesian AI system is really well positioned to analyze this type of data, because what we’re trying to get at is not just a correlation. So, a lot of analytical methods look at how A is correlated to B, and how that correlation may you know may predict a greater depth of understanding. But what we’re really after is, how do we understand the elements within a patient’s biology to link a causal inference between a mutation of a certain gene or a dysregulated expression profile of a protein in a given pathway.
And then using that as a pivot to correlate that you know, wow this is what is it could be responsible for the onset of prostate cancer or Parkinson’s disease or why certain individuals don’t respond to a certain drug. So, this entire, you know this whole approach was really it was really novel at that time in the sense that, we were allowing the data to guide us to the hypotheses instead of you know the traditional sense of taking hypotheses and going through a lot of data generation processes.
So, since we’ve last had you know such a forum, two years ago. We’ve advanced significantly on our pancreatic cancer drug, which was then, we were still wrapping up our phase one solid tumor approach. And you know since then we’ve now embarked into a face to pancreatic trial, that trial is really a precision oncology trial. So, we were collecting tissues and samples and you know blood your own etc. on these patients, were able to build a biological profile on these patients. We’re able then to map that profile against whether or not the patient has a response or not.
And that’s really important because that then allows us to truly engage with patient stratification modules or so, as we go into late stage registration on pivotal trials, we would then be able to create you know protocols. Where we can engage companion diagnostics or engage the molecular profile analysis, before allowing a patient to come into the trial. So, it allows us to be more precise, allows for more predictive you know modeling in the drug development process. But you know something I care about it also allows us for patients who are at the end stage of their lives to for us to conduct more ethical clinical trials.
Because if we know that our drugs probably not going to work for that patient, it’s in the best interest of both parties to not offer that patient that drug. So, in pancreatic cancer we’ve made significant strides both on the drug development and a diagnostic component. We’ve advanced a really exciting technology and epidermolysis bullosa where in the end stages of wrapping up of phase 1, trial down at the University of Miami and we’re now in the planning stages of a phase 3 registration trial, in that indication which is a rare a childhood disease of the skin. It really creates a lot of blistering and postures and impaired wound healing.
So, an extremely deleterious disease to the skin and otherwise the psychosocial effects and kids, on that realm also for the psychosocial component we have a drug that’s now in the phase 3 planning phases for chemotherapy induced alopecia. We’ve just wrapped up the trial, it early in a year at Cedars-Sinai and Memorial Sloan-Kettering that asset is, it really is gonna seek to fill an unmet need in cancer, we’re for most almost 60% of chemo therapies induced alopecia which is hair loss. And that really gives a patient of stark awareness a stark, acute reminder that they have cancer.
They can feel it, they can see it and that’s psychosocial component I think is so important. So, advancing this clinical asset into an enabling trial we’re extremely excited about that. So, really you know late-stage plans for these three assets in pancreatic cancer, chemotherapy induced alopecia and EB. And then on the heels of the clinical development we then also have made, you know pretty significant progress on a pipeline. So, we have two more second-generation cancer drugs and development that are now marching towards IND-enabling trials.
We have a really exciting a novel drug target for lark to meet mutated Parkinson’s disease, and we’ve now seen from a recent publication that came out of about a month ago that, some of these mutations may behave like the idiopathic kind in other parts of Parkinson’s. So, the company has made strides you know clinically but also developmentally in the cancer and neurological diseases. And so really this platform which is interrogative biology has really helped to fuel and guide late stage developments in our clinical assets, reposition, I’m sorry reposition some of the known assets and then really fuel a de-novo pipeline of drugs.
Harry Glorikian: Tell me with the platform and this approach of using artificial intelligence, and your Bayesian AI system basically, does it shorten the timeline? Does it identify new pathways; can you do it with a lower you know with that with lower number of people for lower cost? What are all the, why do it this way? What are the benefits of this?
Niven Narain: Yeah, so if I I’ll answer your question in a three-prong sense, Harry. One philosophically and scientifically, I think doing it this way allows us to not throw away the data that doesn’t you know necessarily satisfy a statistical significance or alpha. I don’t think disease you know cares about what satisfies statistical significance or traditional ways of looking at data. We only you know, we for the most part include the data that that satisfies this point of five significances. But there are lots of data and I think the point I’m trying to make is that disease is not very neat, it’s very complex it’s very messy.
And when you look at it from a mathematical in a statistical perspective we have to allow all of the correlations and all of the implications of that data to have a voice. And so this approach allows but you know by taking a Bayesian AI approach, which there are really no cut offs. There’s no preconceived hypotheses to say well we’re gonna just have a cut-off of 80% of the data or 60% of data, we feed all of the data into the system. Clinically it’s important, because we’re putting literally when you know big hot button term is patient-centric. What does that really mean you know how do you really define that?
And I think for Berg it’s being a patient-centric by starting the process of drug development with human tissue samples. Learning as much as we can about the clinical records, learning as much as we can about the components of the biology within those samples, and allowing the math to give power give rise to that biology. So, he can teach us more about what’s going on in the medicine. So, dynamically we learn about the disease much more fundamentally. Scientifically we take a much broader unbiased approach. Clinically we’re allowing for more fundamental insight into what’s going on into disease.
And then when you add on the business perspective of it you know because you’re learning more about the disease and the patient profile that you’re studying, you’re probably gonna you know produce drugs that are much more predictive towards a given population. Which really is defining and exemplifying what precision medicine is from a pure business operational excellence perspective, we don’t need a thousand people to discover a drug. We don’t need five to seven years and the average 150 million according to the Demasi, you know the recent Demasi numbers.
We’re able to really drive lean operationally efficient discovery programs, because it’s very data heavy it’s very technologically heavy and you know our scientist or our operators that are on every disease or every target. They’re able to really dynamically interact with this data in a sense where, they can you know concede and touch it and feel it in a way that it allows that data to really come to life. So, we’re able to of course spend a lot less money on a traditional discovery program. We are reducing the trial and error.
We’re allowing the data to guide us to where we need to focus in on, and then very quickly the discovery teams you know work with development teams to determine what is the best platform, a development platform to put this and should it be, you know a protein base drugs, is it a biologic. Should we look at you know RNAI or CRISPR based technologies should we you know look at a small molecule screen very quickly. So, all of this is done in a modular sense very quickly and I think that’s just been a huge advantage to how efficient predictive and cost-effective we can get from a pure concept to a validated drug target or a validated diagnostic.
Harry Glorikian: So, if you were to put some sort of rough percentage increases or time savings or people savings. Like, what would you sort of give it a rough estimate of compared to the traditional model?
Niven Narain: Yes, so I’m just gonna use really generic you know numbers and I’m gonna just use the VC model. So, the average series A, in the VC is you know from a VC back company from concept to proof a principle, you know let’s say proof of principle to the IND, average is about 22 to 25 million, and that takes about two to three years. Berg is able to cut that in more than half and build a model from concept to a validated disease target or a validated you know diagnostic in about six to nine months. So, that’s even more than 50% and that’s just using a VC model as you know as a denominator or predicate.
Some may say that’s an unfair model to use, if I can use an academic model which of course numbers are lower, but the time is longer. So, the two levers are time and cost if we use a Big Pharma model the infrastructure is bigger, the cost is being a because of a measure of that infrastructure that the cost is higher, but the time doesn’t change that much. So, you know when you look at the lean and the rapidity of the lean nature of what we’re doing in the rapidity to the validation. It’s a stark contrast from what’s or traditional senses and even with the advent of technologies over the past three to five years.
Because to our listeners you know some may say, well gee is hey you know biology has come a long way and it has the emerging technologies have enabled like CRISPR Cas9 and sort of enabled more rapidity and innovation. That’s true but we still have to then validate all those models as a measure of what these validated phenotypes are, because at the end of the day these discoveries have to then go into a funnel and either creating an IND to do first in man trials, reposition an asset. Whether that’s a phase two or phase three or a diagnostic asset, where we now have to go back into retrospective or clinical prospective trials to validate this this biomarker in a patient population.
So, the way that we’re going to validate this is not changed, it’s still the clinical trials. How do we either make the clinical trial more predictive more lean and effective, or how do we get as much information upfront? So, we know we’re triaging the biology against the disease phenotype, the population against the outcome the proposed and desired disease outcome, and then the market size relative to my up for an investment in cost. So, it’s you know I think these methodologies allow also, I think Harry you know one of the points I’ve appreciated over the past couple of years. It allows companies like Berg to go into diseases that are ultra-weir or rear with a higher degree of confidence you know knowing that, these methodologies allow us to get to a go or no-go decision much quicker.
So, in diseases like EB or other rare diseases that triage process allows us to study these types of diseases, where in other cases it’s a you know the investment is a risk.
Harry Glorikian: From what I’m hearing from you, do you believe that this sort of technology trend and I have seen many come and go over time this fundamental approach of utilizing machine learning and AI for drug discovery is going to be, how things are done in the future?
Niven Narain: I think absolutely, I think what’s gonna calibrate and position how AI machine learning is going to be used most effectively is outcome. Until we don’t develop the first drug to be guided or the first drug to be developed with AI, either is a repositioned drug which is you know like our BPM three, one, five, one, zero or a de novo development that’s just flat-out protein or a small molecule that has come out of a machine learning or an AI system. That then is the world’s first pivot to development. Berg is if I’m not mistaken has validated the world’s first clinical Diagnostics and in prostate cancer.
So, we worked with the Department of Defense to just you know literally from Ground Zero to take the health records and the biological records you know predicted. We have found some markers that show the separation between benign prostate hypertrophy and prostate cancer, you know less aggressive versus more aggressive prostate cancers. We validate this is now in retrospective prospective trials and over 1500 patients. So, this really shows that this process can work. I think that if we take a step back and think about the journey of the drug, the drug developer, the physician and the patient. How is this technology going to help each stakeholder, and what is the pathway to commercialization governed by? And it’s governed by payers and regulators.
So, I have seen firsthand, I think all of us should be able to widely accept that the FDA are the regulatory agencies have made leaps and bounds of trying their best to try to understand these technologies keep up with them, engage workshops, engage these conversations to say, okay how did it really work. What changes do we have to make? What do we need to teach within the agency, there’s new awareness of how we review a review process works? Scott got leave has just, he’s amazed me, because he’s a physician but he’s I think he’s demonstrated in a really short time that he’s not gonna allow yesterday’s biases to carry over into tomorrow’s approval process. And the payers, payers are paying in making investments in technological companies to really try to figure out, okay if this is really true how do you help me make my process more efficient. Because right now approximately I’m spending about sixty to eighty percent of my reimbursement monies on approximately twenty percent of those who recovered. So, when you look at the pressure points within the system which the two pressure points and the levers are, how do we engage the regulators to help us get these products approved. Because if the products are not approved this is just a bunch of fancy science.
It sounds harsh, but it’s true and if the payers are not gonna pay for it, then you can still get a drug or technology approved what’s gonna be adoption and implementation. So, those two big levers have made such tremendous leaps and bounds in the past three years, that it allows folks like me folks like you know, companies like Berg’s to really have a lens of hope that the investment in the technology and the investment in a time, the investment in these types of approaches. If you can create the right products that show that you’re safe, it’s safe, it’s validated you have a process of showing that these diagnostics or these drugs really gonna create a step change.
Unlike five years ago Harry, if you remember the conversations at the conference’s, there were whole sections of conferences that dealt with, well how is the FDA gonna look at it, how are regulators gonna look at it or payers gonna understand it. You don’t see those tracks at conferences anymore, you see FDA representatives or representatives from pairs speak on panels, right next to CEOs, right next to leading scientists or clinicians. The conversation is here; I think the future is really exciting. I think we need to continuously educate each other. We need to, I don’t think we’re all speaking different languages anymore I think we’ve actually found a language of machine learning in AI.
I think what we really need to do is now you know bring together a lens in a concentration around how do, we all together advance these technologies as safely, as quickly as responsibly and ethically as possible. Because the next generation of healthcare is absolutely gonna be based on using mathematics, using machine learning analytical methods, artificial intelligence, virtual reality, augmented reality to you know to allow the patient story to be told in a way, that allows drug developers to create drugs that we can’t even imagine today.
Harry Glorikian: So, there I would say let me challenge you on that, so I’m not challenging the payer the regulator there’s always struggling to keep up with everything that we’re doing. But you know we’re gonna create a new company using machine learning AI and so forth. The hardware is advancing at unprecedented rates, right. The software is improving every time you turn around. So, what do what do we need to do to? I mean totally different set of employees in my mind right and a hybrid, I need somebody who understands the biology.
And then I need somebody that can actually write the code, and then I need that upgradable on a regular basis. Because otherwise if NVIDIA is new chip is ten times more powerful than the last chipset, well the guy who comes after me leave me in the dust, because his processing capability is that much better that much faster. Now I know the fundamental data is what drives these systems, but you know I’m just where do we need to be what do we need to be doing from an implementation hiring perspective, capabilities perspective in your mind.
I remember when I interviewed you the last time, you said you know at one point we needed to go back and rewrite some of the stuff we were working on, because we got some new blood that came in and showed us a new way to look at it. So, how do you balance those things for companies that are coming up that want to be the next Berg?
Niven Narain: I think you have to say, look we’ve made our very healthy share of mistakes along the way. It’s not as you can imagine not been an easy road, in anything it’s never an easy road but it’s never an easy road when it’s uncharted and innovative you know territory. So, if you just take I think the only analogy I can think of in my mind to, when you think of the future is you take a piece of paper and a pencil. And a piece of paper and a pencil, makes a note. Now you upgrade that and there’s a typewriter, you upgrade a typewriter you got Microsoft Word.
You upgrade Microsoft Word, you have these technologies and machine learning that has a speech recognition capabilities. We’ve just gone through four platforms of simply writing and that’s just simply writing, just putting a word down to a recording, a recorded piece of instrument. That instrument went from a paper to a typewriter, to a software to now an Augmented software, and but it empirically has changed and has been altered over time. Because it started out with the hands and the eyes and the brain. But then we added in the mouth at the end now and now with speech recognition is, it’s using you know language in a different way.
It’s combining more empirical components, that’s exactly what we’re doing in biology. Because we started out you know looking at you know an individual genes as we looked at gels, we looked at you know animal models, now there’s AI and machine learning and how is it all gonna keep up is, I would submit to you in your challenge that it’s not gonna be easy. But what I would also you know balance that recognition of that challenge is that, unlike where there were only a few companies you know who would create word processors, you know whether it was word or other processors.
There’s so many companies did the critical mass of individuals and entities there to dealing with the issue is whether it’s software hardware or education. And I should really emphasize the educational component because I think it was a nature commentary a few months back, where I said the PhD programs of the future they can’t be just you know, I think the days of just getting a PhD in computer science or a PhD and molecular biology. The individuals we’re gonna make the biggest change in the future, those individuals who really know math and biology or know CS and biology or know CS and medicine, but it’s gonna be a hybrid system.
I agree it’s gonna be biology plus or it’s gonna be math plus, and that’s really what the employee of the future is gonna be most successful. And I think that is gonna take, I mean I think we’re aware of because we’re having a conversation. So, that’s people check the box on that.
Harry Glorikian: I’m not sure it’s everybody .
Niven Narain: That’s fair, but the educational process has to change. I think you’re seeing, I mean unfortunately right now it’s you know I named kind of the same names, and they’re really the leading institutions you know Stanford, Columbia, Oxford, Harvard you know Carnegie Mellon etc. There many others, but we still have not met that mass, you know critical educational sea change that is bringing together this hybrid, this fusion of Technology if you will. So, I think that’s one extremely important component.
But having said that I don’t think, it’s we’re out doing Moore’s Law in so many ways we’ve outdone it in software, we’re all doing it in hardware for sure. And I think on the educational component since the forums and platforms and the access entry points to education have been completely revolutionized. Because of things like the Khan Academy, because of it you know things like AI, you know some of the platforms that the Gates Foundation and others. And there are many others those are just you know some of the ones that come to mind very quickly, but you need not go to a classroom anymore to learn. You need not be a part of a formal community anymore to learn, you literally can learn off of a computer-based interchange.
Now the practical components of that have to be played out you know obviously within the community. But I think since that’s changed so much Harry, uh the point of bringing this together the enhancements, the Corrections, the course changes or the course Corrections that are gonna be inevitable. I think it gonna happen much quicker in the next few years and they would have happened ten years ago. So I think, I’m a bit more hopeful that folks being able to learn from the mistakes, the mistakes you know frankly that the company is like Berg’s made and others, which I think we need to be very transparent open and frank about things that we’ve done well, things we haven’t done well.
You know I think one of the big mistakes we performed early on is we were so tunnel vision into the technology, that we didn’t bring in some of the endpoint stakeholders. I think we brought him in a bit too late, if we had brought them in earlier like some senior members of the pharmacist societies, some you know you know doing a partnership with Pharma earlier. You know speaking to payers earlier, engaging folks like you know like Medicare or you know the NHS or you know providers help us really understand what really matters, how do we develop technology is in a much broader sense.
I think we would have potentially you know gotten there faster or had more robust data. But having said that it was a first you know we were doing things for the first time. And you know looking back on the ten years I think what’s gonna help the next ten years, be more effective for our company and for many other companies and groups is that, we have to have these conversations and share -. It’s so important to share what we all think is the right thing to do. It is gonna be even more important to share what we think is not the right thing to do or frankly just a wrong thing to do. And I think we have a moral responsibility to speak up more about that.
It’s like you know people don’t like to publish bad data. Well, we need to start to talk about bad processes or wrong processes, because it’s just gonna help the community get there faster. And of course there’s competitive intelligence and you know companies are competing against each other, but if you think that longview you’re only helping yourself. Because of you if the payers and the providers and the regulators, they get it more effectively and they get it in a less you know timeframe, it helps everyone that’s charging for it in that same direction. It helps the entire community. So, I think that’s the way we need to look at it.
Harry Glorikian : Well, if you look at technology companies right, they come up with standards. All the AI research is being published by all the players and they’re competing more on the data that they have that’s proprietary to them, but not the algorithm not the code, that’s really reducible. So, that’s not the necessarily the protectable asset.
Niven Narain: No, I think the algorithms are, I mean they’re there many companies and groups who are just frankly using you know open source software, sharing their software you have great academic groups like Atul Butte, group at UCSF Eric Schadt at Mount Sinai, Andrea Califano, Chaz Boudreaux Oxford. They, I mean these guys literally share all their data and they’re very open about how they do processes. And I think those are four names that I admire, because it’s showing that, you know intellectual property is really important you know you know patents help to preserve your right for a defined period of time to sell a product.
And that’s really important for commerce, but in order to move the needle significantly and create a sea change in innovation, I think there’s a key difference between the innovations that’s necessary to make big steps and big changes towards the scientific discoveries. Because that alone if everyone can share and get that part of it right, then now it’s incumbent on a company or a group to then innovate how do they create novel products that are protectable around that. And those are two really different layers of innovation, they oftentimes get lumped together and that’s where a lot of issues and problems come out.
But if we can understand that you know this is really a multi-layered process of innovation, where it’s like a pyramid and at the bottom, everyone’s got to play well together and be open and be transparent. And that allows us all to be better and then out of that funnel of that initial baseline of innovation, now it’s incumbent on individual groups to productized when you productize and of course you can have IP and patents around that. That’s really important because otherwise where’s the incentive and then the top layer above that is then the commercial and the reimbursement and the proliferation of the business models that actually have a repetitive and a sustainable model of revenue to fuel the ongoing second, third, fourth generation products of that initial innovation. And if we could think about it in those layers I think you know we can make a hell of a lot more progress.
Harry Glorikian: So on that note, I want to thank you for joining me today on the podcast. And look forward to future interactions and hear more updates on Berg and where it’s going and how you’re changing outcomes for patients and driving technology forward. So, thank you very much for spending the time today.
Niven Narain: Well, thank you Harry and I know in closing I just like to say, I think you’ve done a fantastic job of allowing the voices, you know multiple voices to be heard. Because I think that’s really important, I know every time we talk or every time you make an introduction to someone else. I always get a different lens and that’s really important for me as a scientist, as a CEO, as a human being. So, really I think your podcast and you know obviously your books that you put out and the narrative that you were helping to create within this industry for all of us.
I think is really unique, because you’re touching CEOs, you’re touching the senior academicians, you know pairs you know folks from government and you’re bringing that conversation together. So, I think this is a really cool outlet to make us really think about what we’re doing, so we can be better at it. So, thank you Harry.
Harry Glorikian: Thank you very much. And that’s it for this episode, hope you enjoyed the insights and discussion. For more information, please feel free to go to www.glorikian.com. Hope you join us next time, until then farewell.
How to rate MoneyBall Medicine on iTunes with an iPhone, iPad, or iPod touch:
- Launch the “Podcasts” app on your device. If you can’t find this app, swipe all the way to the left on your home screen until you’re on the Search page. Tap the search field at the top and type in “Podcasts.” Apple’s Podcasts app should show up in the search results.
- Tap the Podcasts app icon, and after it opens, tap the Search field at the top, or the little magnifying glass icon in the lower right corner.
- Type MoneyBall Medicine into the search field and press the Search button.
- In the search results, click on the MoneyBall Medicine logo.
- On the next page, scroll down until you see the Ratings & Reviews section. Below that you’ll see five purple stars.
- Tap the stars to rate the show.
- Scroll down a little farther. You’ll see a purple link saying “Write a Review.”
- On the next screen, you’ll see the stars again. You can tap them to leave a rating, if you haven’t already.
- In the Title field, type a summary for your review.
- In the Review field, type your review.
- When you’re finished, click Send.
- That’s it, you’re done. Thanks!