Using AI and Data Science to Lead Advancements in Medicine

Novartis – Bulent

MICHAEL KRIGSMAN: We’re speaking with Bülent Kiziltan, Head of Causal and Predictive Analytics at the AI Innovation Lab of Novartis. They’re a huge company. This will be a fascinating conversation into data-driven drug discovery and development. Bülent, how are you? It’s great to see you. 

BÜLENT KIZILTAN: It’s great to see you, Michael as well. Thank you. How are you? 

MICHAEL KRIGSMAN: I’m well. Bülent, tell us about Novartis and tell us about the AI Innovation Lab. 

BÜLENT KIZILTAN: Novartis is an American Swiss multinational company. One of the larger pharma companies in the globe, with more than 100,000 employees. AI Innovation Lab has been recently created, over a year old now. The company had the visionary roadmap to transform Novartis into a data science company several years ago. And in that context, Novartis understood that we need to really push in the air innovation space. 

And we know that it is only possible to be at the frontier of AI innovation, if you strategically intersect academia tech and pharma at the same time. And with standard and traditional data science operations, that has not been possible in the broader sense of the pharma and biotech. So AI Innovation Lab is positioned strategically, to intersect the different disciplines connecting to academia. Basically, tapping into the know how that’s being created in that space. 

And co-developing, co-innovating in that space with leading institutions and the pioneers of the field. Also, partnered with tech giants such as Microsoft Research, to really bring in that technology into the operations. And then also, cut across disciplinary lines and business units internally to drive impact at Novartis. 

MICHAEL KRIGSMAN: And Bülent, you’re Head of Causal and Predictive Analytics at the AI Innovation Lab. So what does that role encompass? 

BÜLENT KIZILTAN: At the Innovation Lab, we have three pillars. One of them is focused on image analytics and visualization, that is led by my colleague. Another pillar is NLP, which is also very specific to natural language processing. And everything else that we do in the AI data science and machine learning sense and advanced statistics, falls under the responsibility of my team. And in addition to what we do, traditional data science, we are also building capabilities into causal inference, causal discovery space. 

MICHAEL KRIGSMAN: We’re talking about data. So when you discuss causal and making inferences, that implies data. So how does data come into play in the drug discovery process? 

BÜLENT KIZILTAN: Data has been already in the very center of our decision making process, drug discovery and development process. So while we decided several years ago to transform the internal culture, and build a roadmap for Novartis to become a data science company, what we meant by that is, how do we make sure we can leverage data even more effectively and make sure that our decisions are driven by data? 

And this is valid for drug discovery, drug development and everything else that we do within Novartis. So specifically in the drug discovery space, the Novartis Institute for Biological Research has already large amounts of data. Close to 20 petabytes of data already, that has images, scans, chemical information and compound libraries that have information for more than– close to two million compounds. 

So how do we really combine all that information, and makes sure that the whole discovery and development process that takes painfully several years, several billion, how can we make it more effective, shorter? And make sure that we don’t have to do everything in the lab, and we can do that in silico, namely with a computational modeling. 

So that’s the direction we’re heading. Data plays a pivotal role in that whole process. The goal with data science and AI innovation is to expedite that process, and tap into areas of information that we were not able to do so effectively before. 

MICHAEL KRIGSMAN: So fundamentally, then the difference relative to the past is using computational methods as opposed to lab-based methods. 

BÜLENT KIZILTAN: I would say in addition, we are basically augmenting the whole process. And also, sometimes we can simulate information and data, that we can use to add to the informational evidence that was created in the lab. So we are basically partnering up, replacing the laboratory data that is coming out of that effort. And in addition, we are also building capabilities that can help us to effectively combine information, that are coming from what we call different modalities. 

For instance, if we have images, if we have chemical compounds, the traditional method of analysis and combination, was to run analysis separately in those two different modalities. Whereas with new and innovative methods, that is coming from the domain of data sciences, we can effectively and more efficiently combine that information to extract new insights. 

MICHAEL KRIGSMAN: So is the role of data changing, or is it– and I’ll say “merely” gathering different kinds of data? So is the use of data fundamentally changing relative to the past? 

BÜLENT KIZILTAN: I would say it’s evolving, and it’s on a continuous path of evolution. How we can really use data, and create data and record data. So this is why the whole data science domain is very interdisciplinary. That we need to work together in tandem with our partners from the technology domain. 

Also, work with academics to bring in the cutting-edge know how which they’ve built in that domain. And combine that effectively to come up with new insights, come up with new methods of combination. But also, we have been exploring with methods to produce new types of data to augment the whole process. 

MICHAEL KRIGSMAN: How much data do you work with? 

BÜLENT KIZILTAN: It depends on the use case. Sometimes it’s very limited data, sometimes we work with a terabyte to petabytes in terms of size. But really, it depends on the use case, what area, which disease area, which a specific drug that we are working on. And in the health care domain, we basically need to build capabilities, which allow us to extract information both from small and limited data as well as big data. 

And this is a whole spectrum of information content, if you will, when it comes to what type of data and the size of the data that we work on. The future of innovation is not only with big data, but also, we need to be able to extract effectively information from small data. So it really depends. 

MICHAEL KRIGSMAN: We do tend to think of large volumes of data when it comes to activities like machine learning, but you’re also working very closely with sparse data, where you don’t have that much data and yet you still need to drive and draw certain types of important conclusions. 

BÜLENT KIZILTAN: That is correct, Michael. Specifically in health care, more often than not, we are limited by the amount of information that we can gather for a specific use case. We cannot just go about building methodologies in vacuum, and then try to apply it to different use cases or problem sets. 

So we are in a constant evolution in the domain of data science and AI, where the methodologies and innovation front is not merely driven by the methods, but it’s driven by what type of data is available. And that’s why we as a group at the AI Innovation Center, we are building capabilities that not only allows us to use big data, if you will, but also can extract effectively and efficiently new insights from limited data. 

You can augment the data with different methodologies, but also use some really mature applied mathematical methods to combine that information, and basically build innovative approaches to extract a combined insight from the whole spectrum of information that you have access to. 

MICHAEL KRIGSMAN: Can you give us some context around the kinds of data that you look at, or in general in the pharmaceutical industry that’s being used? 

BÜLENT KIZILTAN: Yeah, so the information can come from compound libraries in the drug discovery space. It comes from Web Labs. It comes from publicly available sometimes patient records. There are images. There are sometimes videos that are being recorded along the process. 

So whatever information we can get our hands on, depending on the use case, can vary a lot. There is genomic information. There is blood work. There is laboratory analysis that we can tap into. There’s also a large amounts of publicly available data, that we use to augment the whole inside prediction process. 

MICHAEL KRIGSMAN: So you’re gathering this very broad spectrum of data both from within Novartis, as well as external reference sources of a variety of different kinds. And all of that goes into the drug discovery and development process, when you’re doing your analysis. 

BÜLENT KIZILTAN: That’s correct. And Novartis sits in a very unique place where it has over the last two decades, recorded over 2 million patients year’s worth of data, that is accessible internally to our data scientists and biologists. 

In addition to the data that’s being created and procured by the Novartis Institute for Biological Research, which has access to chemical data, images and biology– specific to biology, the amounts of data that we’re bringing in. But also, we’re tapping into publicly available data, to augment the whole predictive process. 

MICHAEL KRIGSMAN: And what kind of infrastructure needs to be in place, to manage all of these different kinds of data? Again, whether at Novartis or in general inside the industry as a whole. 

BÜLENT KIZILTAN: Making data accessible and preventing data scientists who spent most of their time on data wrangling and data curation, I think is one of the major goals of every larger pharma company, because there are lots of regulations. 

So companies have been investing billions of dollars in building that infrastructure that will make that data fluid internally, meaning, directly accessible and ready to analyze. So data scientists can use the innovative technologies and architectures, to partner with domain experts to produce additional insights into the whole process. 

MICHAEL KRIGSMAN: Is that generally an IT function, or who’s responsible for building that infrastructure? Again, just in general across the industry. 

BÜLENT KIZILTAN: Yeah, IT is definitely a major player and partner in that whole process. But depending on the strategies, priorities and resources of companies, which business unit and function it is under can really depend. We have a specific data and AI function, where we partner very closely with IT. And we do cross-functional work, in order to build and make that infrastructure really usable by all our data scientists. 

MICHAEL KRIGSMAN: And what about the question of cloud versus on-premises? Does that come into play at all in terms of how you collect, gather, use, interpret the data, cloud versus on-premise? 

BÜLENT KIZILTAN: Yes, that is a question that’s being addressed by our partners from the technology world. Certainly we have a major partner, which is Microsoft Research and AI Innovation Lab. We are working with them, and use their cloud Infrastructure for our data pipelines. But we have internal infrastructure as well, that will allow us to run a quick exploratory studies, but also scale it up. So we basically are investing on both ends. 

MICHAEL KRIGSMAN: Yeah, both ends meaning both cloud and on-premise. 

BÜLENT KIZILTAN: Internal infrastructure, yes. 

MICHAEL KRIGSMAN: Bülent, for this type of drug discovery and development, does real time data come into play at all? 

BÜLENT KIZILTAN: So we definitely work a lot with what is called a longitudinal data, which is data recorded at different times. But those are primarily recorded, and then we do a retrospective analysis on those data. So we don’t do real time analysis. But there is infrastructure in place with other functions, where real-time data might be important. 

MICHAEL KRIGSMAN: Can you give me an example of where real-time data might be important? The reason I’m asking, is because one would think with drug discovery, you’re collecting this data and then that data is analyzed after it’s collected, where you interpret it and then produce some type of conclusion. And that’s not how we tend to think about real-time data. 

BÜLENT KIZILTAN: That’s correct, so in the drug discovery and development space, I would say all the data that I’ve worked with exclusively have been recorded, but there is a temporal aspect to the data that plays a very important role in the analysis. But there are functions internally, a potential that relates to market analysis that need to tap into real-time data potentially for their analysis and prediction. Which might use some real-time data, but I have not personally worked on those types of data. 

MICHAEL KRIGSMAN: And obviously that’s distinct from the process of drug discovery and development itself. 

BÜLENT KIZILTAN: That’s correct. So in the drug discovery and development domain, the processes typically last a very long time. And one of the goals of AI and data science, is to shorten that process. So most of the data is being recorded, but they have a temporal aspect that we really tap into and use in our analysis. 

MICHAEL KRIGSMAN: And so really then, an important goal is the shortening of that longitudinal period, in order to derive conclusions more rapidly and bring drugs to market more quickly. 

BÜLENT KIZILTAN: That’s correct. The cost goes up linearly, with the increased time that is spent on a particular drug development and discovery process. And the reason why it might take long, at least one aspect is to generate evidence and enroll cohorts, and patients into the studies. 

With machine learning data science, we can shorten the period by creating additional evidence, using some of the methodologies that we are developing, but also target specific patients and cohorts that will be more suitable for specific studies. So machine learning, definitely helps with shortening the study timeline and also driving down the cost for drug development and discovery. 

MICHAEL KRIGSMAN: Do you know whether there has been research conducted about the effectiveness of machine learning, in order to shorten that time period to bring drugs to market? 

BÜLENT KIZILTAN: Yes, the whole process, drug discovery to development is a very long process with multiple moving parts. And machine learning does currently play a role in most of those moving parts, and has the potential to critically and positively affect the other remaining moving parts. So we are investigating and actively investing into building capabilities, to augment every moving part in this whole pipeline that contributes to that process. 

MICHAEL KRIGSMAN: What are some of the data challenges that you face? Whether it’s gathering high quality data, consistent data or anything else that you run into, or that people in the industry as a whole run into when collecting all of these various kinds of data. 

BÜLENT KIZILTAN: As you already mentioned, Michael machine learning, specifically the standard methodologies rely on the amount of information that’s in the data that’s being gathered. So the more data the better. In the health care domain and biotech, the amount of data is a constant problem, because it takes a long time to gather the data, to make it clean enough for the data scientists to be used. 

But also, accessing the data across different business units can be challenging, because of the stringent regulation. So everything above that you mentioned, getting clean data, getting the amount of data that we require to produce evidence, but also accessing the data which relates to the investment into the infrastructure, are all important aspects of the whole process. 

MICHAEL KRIGSMAN: Do you face a different set of data challenges than folks who were doing traditional lab-based work? 

BÜLENT KIZILTAN: I would say yes and no. Some of the different challenges that we face when we basically go into gathering the data, coming from different laboratories, coming from external vendors and also internal clinical trials, the challenge is to really combine information that is coming from different streams. 

The internal biases might be very different. The noise can be very different. Whereas in the lab, the pipeline can be very homogeneous, even though it can be biased. But once they identify that specific bias, it can be addressed immediately. Whereas when we work with multiple data sources, the underlying problems and biases can be more complicated. 

MICHAEL KRIGSMAN: In a lab, obviously it’s a much more controlled setting, whereas you’re taking all comers, so to speak. 

BÜLENT KIZILTAN: That’s correct. So as you said, so things can be controlled in a lab, but it’s still a painful process for the people in the lab to do things manually. 

Whereas we are trying to automate that process as much as possible, we basically tap into methodologies that can be semi-supervised as they call, or unsupervised or are self-supervised. So we take incremental steps in adding levels of sophistication when we’re building the methodologies, but we take incremental steps, which is, I think, very critical in the process. 

MICHAEL KRIGSMAN: Now you mentioned bias. How do you ensure that the level of bias is reduced to a minimum, when you’re working with these large data sets? 

BÜLENT KIZILTAN: That is an outstanding question. The domain is actively investing into building well-defined strategies, to address biases when it comes to data collection and analysis. So there are two types of biases that we can talk about. One of them is inherent to the data, and we can test that with a placebo group. 

Or if we have an additional set of data that we can compare, the data that we’re working on, we can come up with a metric that can give us an insight into what type of biases there are in terms of the content or the data itself. And then we have biases that might come from the methodology that we’re applying to. 

And there we also have in place some approaches which we can use to measure effectively, what type of biases there might be. So uncertainty prediction here, plays a critical role in understanding what we’re predicting. And this is kind of an ongoing and dynamically evolving area within data science. 

MICHAEL KRIGSMAN: It sounds like the bias issue both from an inherent bias standpoint as well as a process standpoint, it sounds like these are both very important issues to your team and to the industry in general. 

BÜLENT KIZILTAN: Very much so, and this is specifically important in health care and biotech, because the decisions that we make based on the models will affect patients’ lives. So we are investing into really identifying those biases, and try to come up with ethical ways of defining our pipeline and processes. 

MICHAEL KRIGSMAN: The ethics issue is a very thorny one as well. 

BÜLENT KIZILTAN: That’s correct. 

MICHAEL KRIGSMAN: Do you want to expand on why the ethical aspects are challenging? 

BÜLENT KIZILTAN: Yes, so AI and ethics has as well as the other areas of AI been evolving. But one of the reasons why it has not matured as quickly as the other areas, is we did not have investment and partnership from social sciences coming into play and partnering with engineers and scientists on the specific issue. 

We have seen that kind of partnership evolving and becoming more mature, but we still need to find ways to intersect with social sciences and academia, and other players that can really contribute to that whole effort. It’s not only a technical issue, but there are topics that really touch the technical aspects of what we do. 

But has to do with social sciences and understanding the society, and the technology and the long-term ramifications. So it is a truly interdisciplinary domain, and that collaboration has not evolved as quickly as the technical aspects of AI. 

MICHAEL KRIGSMAN: Well, certainly when it comes to issues such as equitable access to health care resources, types of diseases that are studied, the costs and so forth, health care is very different from virtually every other domain for obvious reasons. Right, because it’s hey, if I’m sick, I don’t care how much it costs. But at the same time, I don’t want to have my health insurance go up. So obviously the social aspects are extremely complex, needless to say. 

BÜLENT KIZILTAN: Very much so Michael. Health care biotech pharma is a very complicated domain. There are multiple partners that have to be actively engaging into the discussion, and development of the methodologies that we are kind of pushing at the frontier. 

We cannot be only driven by academia. We cannot be only driven by technology. We definitely have to partner effectively with domain experts coming from different disciplines. And it’s truly an interdisciplinary domain, which may it’s sometimes very difficult, but it’s worth investing into and it’s exciting. 

MICHAEL KRIGSMAN: Tell us about the composition of your team. You’ve mentioned this interdisciplinary aspect a number of times, so how does your team reflect that emphasis? 

BÜLENT KIZILTAN: Because data science and AI is very interdisciplinary, we have been focusing on attracting talent that can bring in their diverse backgrounds into our operations and pipelines with their contributions. So we are attracting talent from very different domains, ranging from physics, mathematics, bioinformatics, chemistry, you name it. We have even people with social sciences background, that have built some analytical skills on top, that are contributing to our ongoing efforts. 

MICHAEL KRIGSMAN: So you’re bringing people together, with very different diverse set of backgrounds. 

BÜLENT KIZILTAN: That is correct. And one thing that is very important in data science, is we cannot focus only on the capabilities that they bring on board. Mainly because capabilities, if they are not nurtured, they can become obsolete in six to nine months. 

So we want to attract talent that can really learn, adapt and is truly interdisciplinary and is curious. This is why when I go out and talk to my peers and colleagues, I want to focus on curiosity and the potential to learn, in addition to the capabilities that they bring to the table. 

MICHAEL KRIGSMAN: So it’s not just a matter of having the scientific background, but they need the right type of mindset and culture really. 

BÜLENT KIZILTAN: Absolutely, sometimes that becomes more important. With the potential certainly to learn the technical aspects of the work that we do, which is very technical, I think the mindset plays a pivotal role in becoming successful in the domain of data science and AI. 

MICHAEL KRIGSMAN: So they need the right technical chops, the right technical skills for managing and dealing with the problems you address, as well as that cultural and mindset view. What are some of those core technical skills that as a baseline, they at least have to possess? 

BÜLENT KIZILTAN: Certainly a good level of understanding of the mathematics of the work that we do, the machine learning and computational aspects of what we do, are kind of the minimum technical capabilities that we look for. But in addition to those, we want to attract talent that can learn, adapt to a changing environment. 

Can ingest data, and information quickly and execute, because what is relevant today might become obsolete in six to nine months, as I just mentioned. So in addition to the mathematical and computational core capabilities, we’re looking for talent that can adapt, learn and execute in a short time. 

MICHAEL KRIGSMAN: So that adaptability is key, because drug discovery and development is evolving at a very rapid pace. 

BÜLENT KIZILTAN: That’s correct, and data science in particular on top of drug discovery, is evolving even faster. So when we sit in the room, more often than not, I mean, we feel like we don’t know anything about the topics that are being discussed in the world of biology or chemistry. 

I certainly come from an astrophysics background, but we need to be able to effectively interface, and ingest and understand the information is coming from domain experts. And implement that, and integrate that into the methodologies that we’re developing. So being able to really understand information, digest it and implement it into the work that we do, I think is critical if you want to be successful in the future. 

MICHAEL KRIGSMAN: That’s very interesting. So you have your domain expertise, but then you’re drawing data, capabilities, skills, techniques, analysis skills, capabilities, techniques from other domains as well outside of biopharmaceutical. 

BÜLENT KIZILTAN: That’s correct, so our approaches and technical roadmaps are driven and should be driven by the information that’s available to us. Mainly, it should be data driven, but we are basically blind, if we don’t interface with the domain experts effectively. We would be purely empiricist, if we don’t use that domain information. And that plays also a pivotal role in building a comprehensive machinery, that will give additional insights for whatever problem that we’re tackling. 

MICHAEL KRIGSMAN: So again, not necessarily at Novartis specifically, but the data is obviously very highly domain specific, but the analytical techniques may come from other fields, is that a correct way to state it? 

BÜLENT KIZILTAN: Absolutely, and this is where the innovation frontier is. In fact, I’ve used some of the techniques that have been developed in the domain of astrophysics, applied mathematics into new problem sets. And this is one of the two ways innovation can happen. One way to innovate is to build and develop new methodologies, but the other means of innovating is to apply advanced methodologies to new problems. And this is exactly what we do. 

MICHAEL KRIGSMAN: Interesting, so innovation can come from how you gather the data, the types of data, where you’re gathering that data. But innovation can also come from the analytical techniques, that you’re bringing to bear to operate on that body of data. 

BÜLENT KIZILTAN: Yes, the second one is definitely correct. We can use methodologies that are developed in different domains, apply to new sets of problems. And we are also interfacing with leading institutions, and pioneers to co-develop and innovate new methodologies that has not been around, which is brand new. And then we can apply it to standard problems, as well as new problems. 

MICHAEL KRIGSMAN: Bülent, so OK. So you’re dealing with a group of very, very bright people, in many cases industry leading folks. Whether it’s your partner academic institutions, or the people from different backgrounds working on your team. So my question is a really basic one, which is, how do you get them from killing each other? 

BÜLENT KIZILTAN: That’s a good question. Attracting talent with diverse backgrounds, definitely has its ups and downs. And certainly gathering that diverse background and insight into a given problem, is very, very valuable. But also, it brings in different characters with different cultural backgrounds. 

How they process information, how they communicate can be very domain specific. And so that brings a challenge to the leadership, where they have to really use their leadership skills to build an effective team that can operate as one team, and that can execute as one team and that is certainly a challenge. One of the two features that we see in global studies, that is a determining factor in success or failure for data science operations, is actually not the technical capability. 

One of them is the type of leadership, and the second one is the culture. And both are very related in terms of determining whether a data science operation, can be and will be successful in the mid-term and long term. And certainly, what I have been advocating and telling to students when I give public talks is, in addition to the technical skill sets, definitely invest into your soft skills. 

Because it will be very important not only with your internal teams, but also when you are interfacing with domain experts that will be from a different discipline. Because everybody communicates very differently, and opening up an effective communication channel is number one in order to understand what the problem is, and then we’ll build the solution on top of that. 

MICHAEL KRIGSMAN: I would have to assume that aside from communication that you described so eloquently, being very clear about the outcomes must be top of mind, as well as a means to align the team and get everybody working in the same direction, to achieve the same or a common goal. 

BÜLENT KIZILTAN: Absolutely, being outcome-focused is very important, but also there is an exploratory aspect to data science. And sometimes it can be very limiting to the whole process. It can make things not very clear. 

So it is up to the leader to clarify that process, and build a roadmap that accounts for the exploratory aspect of the whole process. So certainly, we need to explore. We cannot always identify certain outcomes from the get go. We need to be able to adapt to information that’s coming in, and build the risk accordingly. 

MICHAEL KRIGSMAN: That’s hard. Managing that kind of R&D is really hard, because you’re dealing with so many unknowns. 

BÜLENT KIZILTAN: That’s correct. So in data science operations, what we see in the domain is, primarily it’s outcome-driven by business units. This is not where essentially innovation happens, but then there’s the other end of the spectrum where most of the data science operations is R&D driven, where a lot of exploration goes on. 

Our perspective and approach to innovation, is to find a strategic balance between R&D and real-world execution. And certainly, it will become a kind of a balance that is driven by the priorities and resources of companies. 

MICHAEL KRIGSMAN: Bülent, given all of this, what advice do you have for AI leaders or business leaders in general, for managing AI teams and becoming more effective inside their own organizations, at dealing with large amounts of data or even dealing with sparse data problems as you were describing earlier? 

BÜLENT KIZILTAN: The value proposition of data science and AI is not only for long-term implementation. And this is typically the perceived approach of R&D driven institutions, which have the resources to invest only for the long term. But in many of the industry partners and companies, the resources are not there or the priorities are not in place to invest only for the long term. 

Then the strategy goes into opposite direction, where data science and AI is used only for short-term outcomes and use cases. And there what we see, is that data science operations will become less effective. And the value proposition will decline over time, because there are not enough investments for mid-term and long-term capability building. 

So in order to sustain the value creation that comes from data science and AI, I think a smart strategic roadmap will be to invest both into R&D driven exploratory core capability building strategies, as well as real-world impact that can be executed immediately. 

So a balance between the two, I think is an ideal place to be for any data science operations. And where that needle is in terms of the percentages. How much do you want to invest into R&D exploratory core capability building, as well as execution will depend on the priorities, the timelines and the alignment of multiple partners internally in any company. 

MICHAEL KRIGSMAN: Great advice. everybody should have clarity around where they strike that balance for their particular organization at this particular time, or where they’re projecting into the future. And speaking of the future, can we finish up– can you give us some glimpse as to where data-driven drug discovery and development are headed? 

BÜLENT KIZILTAN: I am very optimistic on the impact that data science will have. Today we’ve seen many companies emerging in that specific field. Larger companies are investing a lot in using data science and AI methods in drug discovery, but we’re just at the beginning of that journey in my opinion. The methodologies that are coming out of academia and R&D driven institutions, are very promising. Mature executions in that whole pipeline has not happened yet. 

So it’s an ongoing process. It will take several years before we see that impact really come about, but I’m very optimistic for the future. And I’m hoping that AI and data science will become a critical component of the whole drug discovery and development process. 

MICHAEL KRIGSMAN: Bülent, where do you see data-driven drug discovery and development headed? 

BÜLENT KIZILTAN: We are seeing an impact today and value creation, that comes from AI and data science implementation in the process of drug discovery and development. But I think we’re just at the beginning of that journey, mainly because the whole pipeline evolution is rather slow in the domain of biotechnology and pharma. 

And also, when I look at the methodologies that are being developed by leading institutions and pioneers in the field of data science, that can be implemented in the drug discovery, are just making their ways into that process. So I’m very optimistic that the impact and footprint of AI and data science in the whole drug discovery process, will increase maybe exponentially in the near future. 

MICHAEL KRIGSMAN: OK, what a fascinating conversation. Bülent, thank you so much for taking the time to speak with us. I really, really appreciate it. 

BÜLENT KIZILTAN: Thank you for the invite. It’s always great talking to you, Michael. We’ve been talking with Bülent Kiziltan. He is Head of Causal and Predictive Analytics at the AI Innovation Lab at Novartis. Thank you so much to Redis for making this conversation possible.