Founder Stories: Najat Khan, Chief Data Science Officer at Janssen

This is the BioCentury Show. The BioCentury Show is brought to you by the 2nd BioCentury Bay Helix East-West Biopharma Summit. Join us October 2nd to 4th in Kendall Square to debate how to globalize biopharma innovation to benefit patients and achieve an ROI for investors.

Hello, I'm Steve Usdin, Washington editor of BioCentury. Today I'm speaking with Najat Khan about the intersection of biology and data and how it's transforming healthcare, especially the discovery, development, and deployment of medicines. Dr. Khan has a great deal of insight about how data is and will create exponential change because she's living and leading that change as chief data science officer and Global Head of Strategy, Portfolio and Operations for R&D at the Janssen Pharmaceutical Companies of Johnson Johnson. Najat, let's start with the big picture. J&J has leveraged chemistry and biology to build one of the world's most successful pharmaceutical businesses. How do you change the culture of a pharma company to integrate algorithms, AI, and all the other aspects of data science?

Great question. Great to be here, Steve. Thanks for having me. You know, if you think about science over the last several decades, it's been about evolution, right? There's so many different advances that have happened. And for us at Janssen, we have always been at the cutting edge of that innovation. But our focus is how do we apply it to discover and develop medicines for the benefit of patients? And I would say when it comes to data science, AI, machine learning, et cetera, that's no different. What we are doing at Janssen is we are looking at what are those really critical problems and opportunities we want to answer, whether it's in discovery or development, and then understanding with great rigor how do we apply data science to do that. Now, you mentioned that requires a change element. Absolutely. I would call it an evolution, right? And the way we're doing that is actually having data scientists working shoulder to shoulder with our chemists, our phenomenal chemists and biologists, operational needs, et cetera, to really create that change change from within by applying it to areas. And I want to emphasize this point, applying it to areas that really matter. We'll talk a little bit about that in how we develop medicines. Once you focus on medicines and programs and all of these approaches are going towards that, I think you can create that cultural change that's needed so that there's not only adoption but sustained evolution to drive that change going forward.

So we're going to get into some of the details. I want to ask you this one more kind of general question, which is, you know, there's a broad belief, especially among BioSentry's audience, that small companies are the engines of innovation. To the extent that that's true, is data science an exception? Are there advances that require the scale and the resources that only a very large organization like J&J and the other big pharmas possess?

That's a good question. Look, I think innovation happens anywhere and everywhere. And one of the philosophies that J&J broadly has always had is we tap the best innovation from anywhere in the world. We're really agnostic where that comes from. And again, we focus on applying that to build better medicines for patients. So when it comes to data science and AI, here's what I would say. There's innovation from a data perspective, there's innovation from algorithmic perspective, and then there's innovation from a deployment perspective and the application. We are partnering with many companies externally in order to both build our own algorithms, co-build algorithms, license datasets. All of that is something that we're doing with the ecosystem. And it's not just startup companies, regulators, right? Understanding the policies, which is constantly evolving, working in partnership as the ecosystem is developing and maturing itself. And I think what sets us apart though, I would say at Janssen is, you know, we have a data science team. I mean, building it from the ground up over the last 2 and a half years, it's been a labor of love. But that team is a new function in R&D that is working, as I mentioned, shoulder to shoulder with our chemistry groups, our biology groups, therapeutic areas, functions, so forth. Why is that important? The only way you drive innovation— again, that's, this is my philosophy— is how you apply it to something that truly matters. And that's what we're doing. So again, innovation external, innovation internal, but I think we are making a lot of inroads and pioneering the work to apply it. And you'll see some of the impact and examples I'll talk about.

Well, that's— let's get to that then. And let's start with examples in clinical development, things that are happening or things that could happen in the near future that are dramatically different because of the application of data science.

That's a great question. So here's what I would say. Let's take clinical development and break it up into 3 main areas. One is how are you designing the trial? So you're going after the right patients, driving this concept of precision medicine that we've been talking for a while now. The second is generating the right evidence. So you really understand whether the therapeutics, you know, what is the therapeutic index and so forth. And then the third is around how do you execute the trial, the operations aspect of it. And I think what is exciting about what we're doing at Janssen is we are compounding the impact of data science across all those three areas. So let's start with trial design. You know, what we do now is we are actually using real-world data in order to design our trial in addition to all of the other approaches that we've used. Now you might ask, why is that important? So let me give you a couple of examples. One is understanding who are the right patients that would actually benefit from the therapies. So in our, one of our expect vaccine trials, what we have done is we've used real-world data machine learning algorithms to be able to understand what are the features of those that are most at risk for this specific disease. This is invasive E. coli, leads to sepsis, impacts those that are immunocompromised and, you know, at 65 and over. But we have found also novel biomarkers. And why is that important? We leverage that to actually design the trial. That changed, you know, the inclusion/exclusion criteria of the program. That's changed the sample size and just overall the design of the program. And it was not done in isolation, shoulder to shoulder with our clinical development leads. And now the program is running, it's in, it's in phase 3. There is no therapy or no vaccine for this disease at all today. It's a huge unmet need and really important examples of how we're ensuring that we're going to the right patients at the right time. So that's one. Another example I'll also say is in prostate cancer. You know, there's lots of therapies in prostate cancer. One of the things that's really important is to ensure that those that are high risk are getting the right therapies. So we have developed AI algorithm based on histopath. So histopath images, Steve, you know, these are— everybody's getting biopsies, right? And what we have done is leveraging AI, you can actually get a lot more data out of the histopath images itself. And in fact, we can actually predict, and we've done it with one of our external partners, you can predict those that are more at risk. Why is that important? That is putting precision medicine in practice, right? You're actually stratifying who's high risk versus not. The other thing I want to emphasize is this is using old data, data that's already collected today for new purposes. What I mean by that is what this would look like is you basically have the histopath image. There's an algorithm that's running and it would prompt and say this patient is at high risk for X, Y, and Z and therefore would benefit from this medicine. That's really important because you're not really adding burden to the health system, which is already burdened today, and leveraging data current data for new insights. So that— those are a couple of examples just in the first bucket of trial design and just how to be smarter about that. The other thing that I'd also say is a lot of the times, you know, when we talk about precision medicine, there are patients in the real world but they're not picked up on time, they're not detected. So how do you drive early detection of diseases? So I'll give you a couple of examples. One is in pulmonary arterial hypertension, which is a rare disease. And the other is in AL amyloidosis, also relatively rare disease. Today, most of these patients, if you look at the real world, get diagnosed 3 to 4 years too late. So even if you have therapies on the market, the outcomes can only go so far because you're being diagnosed late. So using very similar approaches with a couple of external partners, we have developed AI algorithms that can actually diagnose the patient up to 3 years in advance by looking at the subtle differences in their echo and their ECG. Now, echoes and ECGs have been collected for a long time and they're collected pretty early on in the patient's journey, but they get missed over and over again. So again, this would be an algorithm that would be running in the background. You do the ECG or echo and it would flag this patient might have XYZ disease, so PAH or AL amyloidosis. We have received breakthrough device designation for both of those algorithms from the FDA. So again, the point around partnering with regulators, supremely important. And we're focused on approval hopefully next year and then deployment after that. So I'll just pause for the first part if you have questions, and then I was going to go to a couple of the other ones.

No, I think those are clear enough. Let's continue the arc.

Yeah. The other thing I would also mention, the second arc is around how do you develop the most holistic evidence generation? So today we have great RCT clinical trial data, can you leverage real-world data? Now, Steve, this has been— this question has been going on for like a decade.

And again, there's so many meetings where people are talking about real-world data.

It's exactly— here's what I would say. Yeah, the way we think about it is, A, you have to have great rigor in how you use real-world data because it's generated in a certain way, not for the specific trial. But more importantly, it helps you with internal go/no-go decisions. But then also it's an and, it's not an or, right? It's not like RWE is going to replace RCT. You need both. Now you might ask why? So a couple of examples. One is with one of our assets, nipokalumab, we are looking at a lot of different indications and really understanding which ones are rare versus not. What is the prevalence in the real world? What is the journey for a patient in the real world? Super important so that we're designing and developing the medicine with the best knowledge on the ground knowledge possible. And that's what we So there's one indication for that asset where we have received orphan drug designation. And one of the big inputs there was using real-world data to really understand what does the patient journey look like. Another in multiple myeloma where there are single-arm studies, we actually leverage a prospective real-world registry in order to compare the active arm versus the current standard of care. And I say prospective for a reason. Retrospective real-world data sometimes has gaps, data missingness. I'm sure you've heard of it, right? Data provenance, all the cool terms we throw around. But the reality is, you know, prospective data actually can be better quality. And that's something we used because we wanted to have high vigor in what we did. That's super important. It was actually submitted as part of our BLA filing, as well as to the European regulators as well. So you can see the value to that, which is really understanding how standard of care is doing versus, versus the experimental therapy.

And it also allows you to preserve randomization, right?

Or some sense of— Yes, because we actually do comparative analysis to make sure the subjects in the prospective arm, you know, external control arm or real-world arm, are similar, comparable to those in the active arm. And quite frankly, the data generation is just better, right? You have less data missingness. And I want to say, like, you know, throughout that entire journey, we go back and forth. We have a lot of, you know, FDA and others have been great about pre-specification discussions where we can talk about the data elements, being very transparent of the data you have, what you don't have, what are the missingness. I think that's another big important aspect of how you do a lot of AI algorithms in real-world data. It cannot be a black box. You want to be proactive about what goes in there versus not.

We've got a couple minutes before we have to take a little break. I want to ask you, I had some specific things that I looked into, some examples of things that J&J's done. One of them is the use of nocturnal scratch as an endpoint Yeah, that seems to me that's a really good example of a way that you can use digital tools to create a meaningful endpoint that wouldn't have existed in their absence.

Can you talk about that? Excellent question. I was going to go there next in terms of evidence generation from digital endpoints. So nocturnal scratch, you know, this is something that affects a lot of different diseases, dermatology, many different dermatology diseases such as atopic dermatitis and others. And so what we have done is we have a combination of both a passive monitoring device, again, all consented with patients. It's a device that's in your home that can actually, looking at the radio frequency, detect at night, like as you're having itches or changes, it can even measure gait for Parkinson's. It can measure heart rate as well. And why is that important? Because there's a lot of variables that are on you, right? But actually having ones that are passively collecting really helps us generate more data in your home, which is where patients generally are in terms of how they're doing. Another example, which I'm actually really excited about, is for Alzheimer's, we're trying to pick up, detect patients for our therapies, for our assets focused on tau using digital biomarkers. And one of the ones that we're using is a 10-minute test. Basically tests your gait, it tests, you know, your ability to play games, so cognitive abilities, and then also augmented reality. So you're supposed to place objects in a certain area and then you're supposed to go back and remember where they were placed. And this approach can actually detect very, very early signs like preclinical Alzheimer's. So just think about the value of that, right? It's in the home. So you can actually lead to more decentralized trials, both from the nocturnal scratch as well as for Alzheimer's as well. We're doing it in other areas. And it can detect subtle differences much earlier in advance, allowing you to develop therapies and actually see the difference of the therapeutic index much, much earlier on. Well, thanks.

We're going to have to take a quick break now. We'll continue when we come back. We'll talk about some of the policy issues. We'll talk about some of the issues about using data for drug discovery also.

Sounds great. Emerging biopharma companies in the US, Europe, and Asia face the same dilemma. How do you globalize to deliver innovation to patients and reward investors? A handful of biopharma companies will build global operations. Most will achieve global economics through partnering and licensing. Others will capture a share of the economics by enabling global development and financings. This October in Kendall Square, BioCentri and Bayhelix once again bring together decision-makers and investors to address these industry challenges and to form trusted cross-border relationships.

Register today at biocentrieastwest.com. I'm back speaking with Najat Khan, and I want to continue the discussion that we started there with the use of data science for running clinical trials, for improving the way that clinical trials are run, particularly around goals around diversity, equity, and inclusion, which are really important to all the companies and to all the patients and regulators right now.

Yeah, that's a great point. I mean, and that's the third pillar I was mentioning on clinical development is how we run our trials. So a couple of things that we're doing. First, using real-world data, you know, US, ex-US, and then machine learning algorithms, we can actually create these really great heat maps. It's called Trial360.ai. It's this internal approach that we have where you can see where the patients are, all de-identified across the US and across other regions. Why is that important? It gives you a data-driven approach of where we should put our sites to find patients, putting our sites to where the patients are versus what prior knowledge tells us. It opens the aperture on where you can go. That's really important for diversity inclusion because these datasets also allow us the machine learning algorithms that we have, index in areas that have high diversal, you know, diverse representation, socioeconomic status, etc. And this is not just an idea. We're now scaling this across 50 trials in our pipeline. That's the one point I want to mention. All of these examples for scaling across every single asset that's important in our clinical development pipeline, which, as you mentioned, it's quite large and diverse. And we have areas of impact in, for instance, in immunology. There are trials that we have accelerated recruitment by about 1.5x. There are other programs where we have seeing, you know, 50% improvement. And I, I want to emphasize one thing, and also in terms of diversity, we've seen the same, especially in fully remote trials that we have developed. So for another Immunoderm indication, we have trials that are fully decentralized. And Steve, what was really exciting was patients can actually use their phone to take images— you know, this is for atopic dermatitis and other Immunoderm diseases— and send it to their physicians. That's what allows it to be fully decentralized. We recruited the first patient within 45 minutes of going live, and more than half of the patient population are people of color. So really thinking creatively as to how you can make the trials much more representative of those that are being affected, affected by that, by that disease. So I think you'll see more and more of that happening using AI for good to go to where the patients are, to reduce the patient burden, increase representation in our trials across the board.

Sure. So we're talking about AI. You can't talk about AI today without talking about generative AI, large language models. Yes. Everybody's talking about ChatGPT, and I think the public, one of the good things about it is it gives the public a visceral understanding really of some of the possibilities and maybe some of the concerns. Start with what's your, what, you know, how does that fit in with what you're doing at J&J? How are you using those tools? How do you think things are going to, where do you think things are going to go?

Because they're moving very fast.

Yes.

No, a great question. Look, it's really become dinner, evening cocktail conversations across the board. And quite frankly, like you said, it's given the broader population really a firsthand look into how an application of generative AI can be useful for them. Right. And so that's really helped across the board. I mean, people are noticing it more. I would say in our case, look, we've been using AI and machine learning, deep learning to talk about real-world data, all of that across fully scale across development, and then a lot around discovery. But here's where it's helping us. Number 1, we're actually starting to use generative AI approaches in drug discovery. What do I mean by that? So for instance, if you think about the molecular invention space, right, how do we design the right molecules, therapeutics, generating new approaches, new molecular entities that you can then test? That is something that we can now do and starting to do with generative AI for both small molecule and biologics. Why is that important? Listen, there's still a lot of work to be done. It's not prime time yet because, you know, generating thoughts from the English language, which we have developed, is very different from generating, you know, new molecules based on the language of biology and chemistry. We're still learning. We're still editors of that, of that area. But it allows you to have new diverse molecules that haven't been developed before. There's an approach called diffusion models that what it does is if you have a structure of a protein, it helps you to actually generate what the what the therapeutic, right, the ligand as we call it, could look like very quickly. And that has the potential to actually accelerate how we make our medicines. So generative AI has a very, very important role to play there. It's not fully done yet. There's a lot of work to be done, but we're playing heavily in that space. Another thing I would say for generative AI that gets me excited in development, think about all of the CMC dossiers, the clinical development protocol. Think about a lot of the documentation that we need to generate. In some diseases like multiple myeloma, IBD, and others, we've done it for some time. So you can actually use generative AI approaches to create the first draft and then be a copilot for you as the clinician is iterating on those documents. Because so that we have our study teams and others really focusing on the most hardest problems to solve versus some of the, like, you know, less interesting things, more mundane things. How can we free up their time? And then the other thing I would also say when it comes to generative AI, remember you mentioned the question around digital endpoints. So one thing that we have started to do is actually use generative AI to generate data that the algorithms can actually train on. So for instance, when we do this for depression, you can use generative AI to actually say, okay, if somebody is depressed, what are the words they might be using? What are the languages they might be speaking, the phrases to actually train your models and iterate much, much faster. You can use it to generate code. You can even use it, and we're using this in our platform Med.ai, you can also use it for asking scientific literature questions, hypothesis generation that includes our internal and external data. So our scientists can really quickly generate data and different hypotheses. So I wouldn't say the opportunities are endless. We're being very selective of where we want to go that's going to actually help us make our make our medicines. One thing I want to say, we've all heard about generative AI and hallucinations. Sometimes the answers are wrong. It depends on what are the, you know, what datasets are used to train it. That's what I want to mention. It's not what you get from it is not a final product. You need a human in the loop to validate things, but it helps you as a better starting point.

And well, and it sounds like if you're, if you're trying to generate hypotheses that actually hallucinate, what's called hallucinations might actually be springboards for people to think of things in different ways, things that they hadn't thought of.

That's exactly right. You know, I call it a brain in a box for me, which is essentially like all of these thoughts and all of this information. How do you actually leverage that? Give me a better starting point, right? Give me ideas, you know, and be my copilot. And that's, that's how we are looking to leverage it.

But some of the things that you, you mentioned also, I think, raise concerns for people. You know, when you're talking about, for example, the example of depression, One of the big concerns about anything having to do with data now is privacy.

Yeah.

The question about does privacy even exist in a world where people voluntarily keep a device in their pockets 24 hours a day, practically, that tracks their movements, their thoughts, really their desires? How do you think about that? How do you think about that in the context of what you're trying to do in, well, in diseases like depression?

Yeah, I mean, it's a great question. Look, I would say one of 60 to 70% of the time that's spent in any of these examples that I mentioned, it's around data validation, data privacy, data engineering, all of the, you know, some people will call it the backend stuff, but the most important aspect around transparency of data, fairness, AI and ethics, reducing bias and being very transparent on what that is. And to me, I think that's how you increase the level of trust. 'Cause like, here's the thing, the reason why whether it's depression example or any other, right? You're generating data, these new datasets you're generating. Why is there a concern? Because if it's not explainable, right? If you don't understand how the data is being generated or if how the algorithms have been developed, and more importantly, Steve, how the algorithms are being deployed and how are they being monitored? If the input data is changing, who's doing that, right?

Like, well, and that, yeah, I think that, that, that's the key because people are worried about being manipulated by it, right? They're worried that somebody's collecting this very intimate data about their personal lives and about their healthcare status and what's going to be done with it. And how do they have confidence of what's going to be done with it? So that flows into another question, which is Congress, federal regulatory agencies, they're already jumping in now. They want to do something about generative AI, large language models, their concern. I think that the concerns are kind of diffuse. I'm not sure that they've actually identified them, but what tends to happen in these cases is that drug developers, people who are trying to advance medicine, are not at the table typically when these kind of discussions are happening. And I'm wondering what you think should happen. What should regulators, what should legislators think are the opportunities, what they should do to advance things, and what are the things that they should avoid to avoid unintentionally disrupting or hindering biomedical progress?

Yeah, look, I would say taking a step back, you know, in healthcare we're used to— I think this is one of the positives when you said external innovation inside and outside— we're very used to patient consent, HIPAA, you know, like putting that patient privacy and patient consent first in everything that we do. That is something that's the fabric of how pharma companies and biotech companies have existed for a long time. Why is that important? That's important because how you collect data, you know, all of that, all of the examples I mentioned, we are consenting patients, you know, um, it's all de-identified and that's really, really important and aggregated. So you're really ensuring the patient privacy is preserved. And not just, you know, the US has one regulations, you know, different countries in Europe and have others. So we have a full team that's constantly focusing on that. Here's what I would say when it comes to regulators as an example. You know, at the FDA, for instance, recently published a white paper on the applications of AI machine learning and discovery and development. And what I loved about it is, A, it was a really good white paper, but B, they had a call to different, you know, practitioners, pharma, biotech, et cetera, to actually give feedback. So that's, that's the two-way dialogue that's actually happening quite a bit. For instance, the FDA's the Office of Digital Health, you know, the RWE subcommittee. So that really helps in terms of going back and forth and learning together. One thing that's really important, it's easy to write policies, but it's really important that they're fit for purpose for these types of questions we're trying to answer. So the examples that I mentioned and others have been working on, using that as a test case to say, how do we make sure the algorithms, the policies, the privacy bills, etc., are fit for purpose, is extremely, extremely important. When you think about policy from a global perspective, for instance, I was at the OECD last year and we were talking about how do you enable data flows, doing it in the the right way across countries. Big question to answer, right? That comes to data standardization. So we did have, you know, many different players, including pharma companies, J&J, at the table to really talk about what's needed. I think it's about the dialogue because, you know, the blueprint for none of this has been written yet. We have to do it together. And the earlier conversations we can have— the white paper is a great example of calling for feedback— will help us iterate and to get it right.

We've only got a couple of minutes left. There's a lot of stuff I want to ask you that we didn't get to. But one thing I do want to get to is the idea of how do you recruit and nurture the right kinds of talent in biopharmaceutical companies and data science. You're an example people kind of point to and say, well, it's a non-traditional career path. You started in organic chemistry, you got a PhD, and then you're now you're running one of the biggest pharmaceutical companies in the world's data science program. But seems to me that there needs to be a wider, maybe wider view of the, of the whole thing. And maybe we need to be looking and thinking, how do you get people in from other fields completely outside of science or who don't have PhDs? What, what you're thinking about how, how to get the right people into, into this world so that we can fully integrate and take advantage of data science and biology that interface.

I think it has to be an and, the data science and biology. You know, as organizations mature, it has to be on an equal footing. And that's not always the case today because that's how you get the best talent to come. Now let's talk about talent. From a talent perspective, I always said, and you know, I have the privilege of building the team here at Janssen R&D, bilingual talent. So Steve, what do I mean by that? There were some people that were completely from a computer science background science, a math background, not a science background at all, but they have learned the science, right, working with our amazing scientists and biologists, etc. It's a great team sport. Then you had people that actually have more of a background in, say, oncology or neuroscience, and then really picked up and upskilled on the, you know, computer science and data science side. So I think I, I'm a realist, right? You're not gonna— it's very hard to find people who know both. But it's really important in our industry to understand not just the domain knowledge, but what we were just talking about, regulations, like, you know, how do you actually get something approved. So actually having people that are bilingual, you know, for me personally, it's funny, you know, when I was doing my undergrad, I was the kind of person who was like, I want to do science, I, you know, want to learn about computer science, and I also want to learn about business. Because if you don't know how to apply disciplines, you're not going to have the impact on patients that you want, right? And there are more and more universities that are actually having these bilingual disciplines as a graduate degree or undergraduate degree. But the thing I want to emphasize is it's, you know, I'm a big believer of discovering talent and making sure you give them the support you need because a lot of this, you know, backgrounds of folks on my team were not exactly one or the other. But what is most important is the piece around inclusion of data science on the teams. So what I mean by that is actually bringing in data scientists scientists early when you're strategically planning on designing a trial, or you and I were talking in the break, redefining a disease. I mean, think about it. We talk about Alzheimer's, Parkinson's, that's named after the physician who found the syndrome. That's not the biology of the disease describing it. So now we have data scientists sitting with amazing scientists and they're looking at multi-omic data, EHR claims, et cetera, all de-identified, to be able to subset those patients based on their biology of the disease. What is the driver mutation? We actually have two machine learning-assisted targets in oncology that we identified working with our scientists and that are now about to be in our pipeline, in our clinical stage pipeline, progressing really quickly early next year. That's how you change. If you don't change the notion of redefining diseases based on the biology that's driving it, you can only do so much, right? It really increases your probability of success afterwards. So being around the table, culture of inclusion, inclusion of ideas, inclusion of backgrounds, that's what creates success in what I would say is self-disrupting by adopting what AI can do in terms of good for how we make medicines versus, you know, working in silos. So background is important, important, but it's also important of the how you work in a data— in a day-to-day world. And, you know, I'll say my team is half women, Steve. This is an area that usually has 15-16% female diversity. That doesn't happen by accident, right? I have so many people that actually reach out to me on LinkedIn in terms of they're interested. They've seen some of the examples we work on. Diversity of thought and background makes a huge, huge difference in how you drive, apply, and have the impact from data science.

Well, thank you. Thank you so much. It's been a really energizing conversation. We don't have any more time, but I hope we can do it again and soon.

Thank you. Thank you for having me, Steve.

And thank you all for watching.

The BioCentury Show is brought to you by the second BioCentury Bay Helix East-West Biopharma Summit. Join us October 2nd to 4th in Kendall Square to debate how to globalize biopharma innovation to benefit patients and achieve an ROI for investors.