Inside Perspectives: Taking the ‘Trial-and-Error’ Out of Drug Discovery Via Artificial Intelligence

Innovation That Matters

WuXi NextCODE CEO Hannes Smarason says artificial intelligence has the power to take much of the ‘trial-and-error’ out of drug discovery and development, achieving a trio of historic milestones – lower cost, increased speed, and better outcomes.

“AI will dramatically change the paradigm,” Smarason says. “For example, you’ll find a (biological) target, you’ll design the compound, you can simulate the clinical outcome before you even run a trial in patients, and so you’ll basically have a product that is virtually driven.”

Under Smarason’s leadership, WuXi NextCODE is applying AI to help partners mine and understand massive amounts of genomic, medical and biological data for drug discovery, drug design and clinical research. The company represents a merger of the WuXi Genome Center with NextCODE Health, a spinout of Iceland-based deCODE genetics.

Smarason says the main benefit of AI is that “it ends up being a much more data-driven approach as opposed to a hypothesis-driven approach. That’s the fundamental value, so that gives you a much greater degree of freedom to identify patterns and novel insights that have previously been overlooked using conventional methods.”

Smarason was CEO of NextCODE when it was acquired by WuXi AppTec in 2015. After overseeing the merger with the WuXi Genome Center, Smarason was named COO of WuXi NextCODE and became CEO in January 2017.

At deCODE genetics, Smarason served as chief financial officer and led the company’s initial public offering (IPO) in 2000, the first IPO by an Icelandic company in the U.S. He subsequently served as an adviser to deCODE and its venture investors, Polaris Venture Partners and ARCH Venture Partners, assisting in the $415 million sale of deCODE to Amgen.

During Smarason’s tenure at deCODE, the company achieved worldwide recognition as the global leader in analyzing and understanding the human genome. Following deCODE’s acquisition by Amgen in 2012, NextCODE Health was formed in 2013 to apply the systems and reference databases of deCODE to discover links between genetic variants and diseases. On this foundation, and serving leading precision medicine efforts on three continents, WuXi NextCODE has built a global standard platform for organizing, mining and sharing genome sequence data, which aims to turn that data into insights to benefit people and patients around the world.

Smarason received his MBA from the Massachusetts Institute of Technology, where he also earned his bachelor’s degrees in mechanical engineering and management.

WuXi AppTec Communications’ interview with Smarason is part of an exclusive series spotlighting the inside perspectives of thought leaders on topics shaping the future of new medicines.

WuXi: How are you applying artificial intelligence in drug discovery and clinical research and how does your approach differ from other companies?

Hannes Smarason: We are trying to convert together three different things around what we call our winning strategy for AI. It’s bringing together cutting edge algorithms, domain expertise and large data sets. We’re looking at human biology, where there are hundreds of thousands of variables and data sets, which in AI are called features.

We’re looking across all the samples we’ve gathered together, so we have a very sophisticated and robust way of mining that data to help identify and discover novel treatment pathways for clinical development and for possibly re-purposing drugs. In order to do that, we are discovering novel algorithms as well as ways in which we can distinguish between those people who respond and those who do not respond within a particular treatment area.

I would say that the differentiating feature of our AI effort is the ability to converge these three necessary elements: cutting-edge algorithms, domain expertise and large data sets.

WuXi: What are the major benefits of using AI and machine learning in drug development?

Hannes Smarason: The main benefit is that when you’re using AI, it ends up being a much more data-driven approach as opposed to a hypothesis-driven approach. That’s the fundamental value, so that gives you a much greater degree of freedom to identify patterns and novel insights that have previously been overlooked using conventional methods. So one of the things that we’ve seen come out of our work in deep learning has been this ability to identify the true causality in relationships within the biology in question. We can find the causal genes or pathways that drive disease, and the hope is that that will allow us to have a much more powerful starting point for the development of treatments.

WuXi: How are you using AI to identify what is happening at the genomic level in biology?

Hannes Smarason: I’ll give you an example. We did some work with the Yale Department of Medicine looking at some research related to the development of the vascular system. The development of the vascular system is very important for the cardiovascular system and for cancers, which need their own vasculature in order to grow.

Our deep learning algorithms made a prediction that a particular mechanism was a key driver in the pathway for the development of the vascular system. That was a mechanism that had not previously been described anywhere. Yale biologists then validated that discovery in an animal model. We were thus able to prove that our AI method had accurately predicted the role of this particular pathway for vascular development.

So here all of a sudden we’ve opened up a whole new druggable pathway, if you will, that can have a wide range of therapeutic applications both in the cardiovascular space and in oncology. That’s a great example of how we use a hypothesis-independent, data-driven approach to find and describe novel mechanisms and validate them using biology. It’s true discovery using data.

WuXi: Do you see any limitations in applying AI?

Hannes Smarason: Yes, of course. We view AI as a force-multiplier as opposed to a replacer. It’s a tool that needs to be used by scientists who look at it as a tool kit as opposed to something that’s going to replace independent thinking. This is also one of the reasons that having domain expertise in genetics and biology within our AI effort is absolutely key.

The second thing is that there are limitations which are dictated by the fact you have to have access to a wealth of information, including cohort studies as well as both genomic and phenotypic data. So that’s one of the limitations and that’s why we believe we have some unique capabilities, addressing the challenges with the large global data sets we have access to and with the necessary domain expertise.

WuXi: How do you see AI evolving over the next five-to-10 years?

Hannes Smarason: I would say we look at AI as another tool in the toolbox, meaning that it’s going to be a useful set of tools you can apply to these very complicated and sophisticated data sets. Much like any major technological development it’s going to start up slowly and then gather momentum.

We do believe that because of AI’s unique ability to take all of these very complicated data sets together, and identify patterns within them, that in the end it’s going to have an exponential impact in advancing and applying precision medicine. The result is going to be game-changing benefits to patients around the globe in terms of better diagnostics and better targeted drugs. It holds great promise. I think the challenge will be to take the necessary steps to show that it really can deliver, and the way to get it off the ground is to make sure our scientists and key decision-makers are better equipped to make the right decisions.

WuXi: What will be the next technological advance in applying AI or do you already have everything you need?

Hannes Smarason: No we don’t have everything we need. One of the most rate-limiting steps in our application of AI is the vastness of the data systems that we are working with and the resulting complexity of feature selection. Remember that the features we have are so much more complicated than facial recognition – we are dealing with hundreds of thousands of features for every single gene and for every single sample that we are processing through our AI. We have this massive data matrix that we are trying to train our deep learning algorithms on, so that’s computationally very intensive.

One of the big advances we’re looking forward to is the advent of quantum computing. Quantum computing will allow us to actually continue to train these computationally intensive deep learning algorithms without having to manually define and select our features, thus taking full advantage of the complicated mathematics and the dimensionality of the data. That’s going to allow us to get to an answer more quickly and more accurately. So a processing step that today may take two or three days will be done in minutes and ultimately seconds. That’s going to revolutionize both the power and penetration of AI across all of life sciences. That’s what I would see as the major game changer – quantum computing.

WuXi: When do you see AI becoming the norm in biotech and pharmaceutical companies? Will it lead to more efficient R&D?

Hannes Smarason: Yes, I think it’s going to be one of those things that will differentiate enterprises into the haves and have-nots. I don’t know whether there is going to be a place for a biotech or pharmaceutical company that does not use AI. This is one of those technologies that is so comprehensive and ubiquitous that I think it’s going to be very hard to compete without having fully understood and embraced the technology. Going forward, companies will have either to develop it themselves or gain access to it through partnerships with WuXi NextCODE or others who have the necessary technology. It’s going to become the norm and not the exception.

WuXi: Do you think AI will speed up the overall R&D process?

Hannes Smarason: Yes, I think it’s going to impact all dimensions. It’s going to impact the cost, the quality and the time. We’re going to be able to do our work based on the data and make better predictions about the future. We’re going to be able to do it more accurately and it’s going to happen faster and faster. So it’s going to address all of the dimensions that you want.

Let me give you one more example that will highlight this. What I’m going to describe, you could do on a hundred- or thousand-fold scale with AI.

One of the things that we were able to do just by using conventional data mining methods – this is before AI – was to actually help identify and discover a novel target, the ASGR1 gene, linked to risk of heart disease. This is a deCODE project, leveraging a vast amount of data on a whole population, and which Amgen now is pursuing for drug development.

This was discovered using our database architecture and discovery tools. We could then actually begin to answer the question: What happens if you look across a population and knock down this gene, what’s going to happen to their overall health? So we could go in and interrogate the database and look at what has happened to these people when it comes to longevity or other diseases. Are there complications in their life by having this particular gene turned off? The answer to that turned out to be no.

So even before you start the drug discovery process, not only did you have a novel target, you also knew that if you drugged the target and turned off the gene there would be no important side effects. So at your entry point, before you developed any compound or did any of the discovery work, you had all that know-how.

Now you can imagine, super-imposing AI on to that and all the details we have, we can get to a world where we can do so much modeling a priori, that the experiment and the clinical trials and the things that we do in the physical world will be confirmatory much more than discovery-oriented. For example, you’ll find a target; you’ll design the compound; you can simulate the clinical outcome; and you’ll basically have a product that is virtually driven. But before you start administering it to humans, you’ll do a clinical trial to validate whether your in silico observations were accurate, and if it turns out well, you’ll go for approval. I see a massive ability to compress the time and the cost and increase the successes.

WuXi: Will AI reduce the ‘trial-and-error’ aspect of drug development and clinical research?

Hannes Smarason: I think that’s a great way to put it. AI will dramatically change the paradigm. We’re going to be able to do so much more. You’ll have so much more information about what you’re trying to do before you even do it.

Think about it today: When I got in my car this morning, my iPhone automatically told me how long it was going to take me to drive into work. I didn’t actually do anything. I never told my phone I was going to work. I never even told my phone I was sitting down in my car. But my phone knew I was in my car and that I was about to drive to work. Now obviously this is happening in the background. I have no idea the amount of data mining that happens behind the scenes before that was something that popped up on my screen. But it certainly gave me intel, which was actionable and relevant to what I was just about to do without me having to do anything.

So you can imagine doing the same thing for drug discovery. Imagine the power that’s going to give you. A scientist is about to run an experiment and before he knows it, the knowledge system pings something and says, ‘Here are the three things you should do;’ or, ‘Why don’t you look at these four results? I already did the experiments for you.’ The power of this is so immense that if properly applied, it can truly transform the drug discovery process.

WuXi: Are there any barriers to applying AI to drug discovery? For example, are there regulatory challenges that still have to be sorted out?

Hannes Smarason: I think there are. Obviously go back to that example I was using before (concerning drug discovery and clinical development in silico). At the end of the day you’re always going to be left with having to do the last and final step of validation, which proves this isn’t some artifact that you are uncovering in silico; but that it plays out in the right way in a human. So it’s going to be a long time before we do away with that step. That’s going to be the necessary final step to get the regulators comfortable.

The other thing that’s going to be a challenge is for the regulators to have the sophistication to take into account the data files that show results generated by AI, or something that is AI-driven, or AI-created. It’s unclear how that’s going to play out.

But as with any new technology that’s going to be a challenge. Still, I have no doubt that the regulators are going to embrace AI, just like they are beginning to do with genomics right now. For example, they have put in place something called precisionFDA, to try to facilitate the inclusion of genomic data in the filings that companies are making with the FDA. This aims to make it easier to submit the right data; standardize the information, standardize its capture; and impose certain quality standards as well. All of those things are going to need to happen and it’s going to be an important part of the widespread use of AI.

But if the goal of AI is to facilitate learning and get you to the target quicker, and it helps you develop the novel compound faster, that’s very much going to happen. It’s going to fit right into the conventional regulatory framework that we have.

WuXi: What kind of partnerships will be important in moving AI technology forward and integrating it into the R&D and clinical research process?

Hannes Smarason: In our case, this is all about partnerships. There is so much happening in this field right now you have to function as an ecosystem. You have to find and put together the right relationships and networks of people that you can work with to allow you to advance the research in the work you are doing. I think that all the members of that ecosystem are important: the data generators, national health systems, research hospitals, diagnostic companies, pharma companies; having the people developing the right algorithms, having access to the large data sets.

The evolution of AI is going to be ultimately a network activity. It’s not going to happen in any other way. I don’t think any one company is going to be in position to do this on its own. You can develop components and part of what you need, but at the end of the day, it’s really going to be a network solution. The winners are going to be those who can work effectively in that type of a complex, adaptive system, where you have to be both a contributor and a recipient. It’s going to be very difficult to be a silent beneficiary without also being a contributor. There’s going to be a lot of give and go here, and if you’re comfortable in that type of environment you have a chance of surviving and thriving. Otherwise it’s going to be very challenging, because of the necessity to have these multi-faceted and multi-dimensional relationships to build out your infrastructure.

WuXi: How soon will patients and health systems begin to see the benefits of this new technology? Have there been any product approvals linked directly to the application of AI?

Hannes Smarason: I know of nothing that’s been approved, which has been linked directly to AI. I do know of things that are in the hopper. When it comes to predicting a definitive time, it’s always difficult. As the famous American philosopher Yogi Berra once said, “It’s very difficult to make predictions, especially about the future.”

I know of companies that we have collaborated with, that we have helped advance their discovery of projects using AI. I would think those companies are still some years away with having those drugs approved. But I would say we are definitely in the sub-10 year time frame before we’re going to see something important come out where people can directly tie it back to AI.

Subscribe to WXPress

Receive our newsletter and information on upcoming events.