AI and R&D: From the Eyes of an Entrepreneur

Related stories

Avak Kahvejian, PhD, is a general partner at Flagship Pioneering, where the hybrid scientist and entrepreneur has co-founded more than a dozen biotech companies, including most recently, Etiome, an artificial intelligence (AI)-based biotech focused on detecting and preempting disease progression.

In the below Q&A with Pharmaceutical Executive, Kahvejian shares front-line insights on how AI is reshaping drug discovery and development. He discusses the various strides that are occurring in the space on a practical level—from designing RNA molecules and mining electronic health records, to rethinking clinical trial strategies—and why companies that fail to embed AI into their workflows risk being left behind.

Pharm Exec: From last year until now, what advances in AI do you believe have impacted the biopharma and healthcare enterprise the most?

Kahvejian: As a general sentiment, we’re super bullish on the impact of AI in biopharma and healthcare, and we think the impacts are only beginning. We expect to see more tangible results in the market within the next few years. But because we have an inside view and we're practitioners applying it on a day-to-day basis, we are firm believers that it's already working. I know there's a lot of skepticism out there—many articles even start with the word “hype” in the headline. In our view, the debate is over: AI’s impact is real and here to stay.

In the past year, a number of our companies have been applying AI directly to optimize their modalities. The clearest example is in nucleic acid technologies. For instance, in RNA, the modules we use to impact stability of the RNA molecule, to impact its efficiency in translating protein, is normally done using human intuition and publications to say, “What are the pieces I should put together? Let's read a couple of papers and figure out a few options that could help us get the maximal output from a piece of messenger(m)RNA.” Of course, you get a range of performance, and you try to take the best ones and move forward with them.

Now, with AI plus high throughput data generation, automation, and miniaturization, we can test a far broader range of options, some of them hypothesis-driven and human-driven; others with a degree of randomness incorporated. And then take that dataset, feed it to an algorithm, and ask the algorithm to suggest the next best sets of designs based on what it learned.

These are the sequences that we put in and the performance we got out. Through that, we've started to create very nice models of generative AI guiding the design of these molecules. This is transformative: we’re no longer limited by papers or human intuition alone to design molecules.

Pharm Exec: Does predictive analytics enter into that equation prominently as well, such as looking at past modeling and simulation approaches?

Kahvejian: It is, in some ways, a predictive algorithm in itself —because it takes in large datasets and, since it’s a multi-parameter optimization problem involving thousands of nucleotides, it determines: “what sequences should be recommended to improve expression based on the data distribution provided?” So, in essence, it becomes highly predictive. It can then predict whether the next sequence will be a strong or weak performer. It's a really interesting virtuous cycle.

The other area I'll mention, distinct from this example but also AI-driven, is applying AI to electronic health records (EHRs) to identify people at risk of certain diseases. In our company, Etiome, where we're trying to preempt severe disease, we've taken this approach of analyzing EHRs with healthcare partners using anonymized, HIPAA-compliant data to identify individuals potentially at risk. That's really compelling. With permission, we reached out to some of these individuals, verified diagnostically whether they were showing early signs of disease, and the results confirmed the predictions.

This approach will be tremendously impactful in, not just in the realm of diagnostics in the future and early warning, but also even in the design of clinical trials and advancing drug discovery.

Pharm Exec: Do you, then, track these individuals over a certain amount of time to see if their predisposition manifests further?

Kahvejian: That’s the beauty of this. Instead of having to track people over time, which is the normal intuition, you would be able to say, "can I see whether a person is getting sick over time?” That’s the traditional longitudinal study model, which is valuable but extremely costly. Tracking thousands of people over hundreds of time points or dozens of time points over a prolonged period of time is cost-prohibitive.

Instead, we flipped that idea on its head. By taking a snapshot of a population, you will already have within that population, people who are not sick, obviously, and not on that disease journey. And others who maybe started on the journey and others who are later into that journey. With the snapshot, and using computation, you can align them along a continuum—as if they were one “meta-person”—and then sample across that continuum to study the molecular underpinnings at different stages.

This creates a continuous trace, as if you had followed one person. But instead, you have taken a snapshot of thousands of people who represent the entire continuum. That allows you to draw a trace of how the disease is progressing across a large group of people; essentially, it’s like tracking one person over time, but instead it’s done almost instantaneously.

Pharm Exec: Are we nearing the moment that pharma companies—through practical application—fully embrace AI and data analytics as a value differentiator?

Kahvejian: I'm sure everyone is thinking about how they apply it. In pharma, it is mostly being applied to existing systems. Companies are likely sitting on disjointed silos of data and thinking of running algorithms on those data to see if they can gain insights and take advantage of off-the-shelf tools to improve their existing processes. So, yes, I think pharma is embracing this.

But what’s unique in the startup world is the ability to design processes from scratch with AI and data in mind. That gives us a distinct advantage, even if we start data-poor compared to data-rich pharma. But it allows us to architect our data workflow and the generation of the data in the right way so that it is amenable to training, but is also amenable to being translatable to humans, and to ensuring that the organization and the R&D are structured.

Both aspects will occur in parallel. We’ll see AI tools applied to existing paradigms to find efficiencies and generate new insights. We can also expect–from what we’re seeing in the startup world–building from scratch with AI in mind to generate the right kind of data, the right amount of data, and to build the right infrastructure to capture those data so we can have a virtuous cycle of prediction and generative AI helping us design the drugs. I think it's going to be exciting in both regards.

Pharm Exec: Do you see pharma C-suites—across industry—incorporating AI into their daily organizational practices and functions?

Kahvejian: It's got to be on everybody's mind and on everybody's tongue. “AI” is such a general term, but it’s already being applied to business processes such as reporting, knowledge management, and even accelerating employee learning and onboarding. What we're excited about most is arguably rearchitecting our platforms or building our platforms with a rearchitected mindset to maximize AI’s impact.

I think a lot of people fall into the trap of saying, “it’s all hype until an AI-generated drug is approved.” I don't ascribe to that mentality. I think drugs can fail for thousands of reasons. Trials can fail due to clinical design, manufacturing challenges, or countless other external factors. And AI won't be the faulty party. AI will be the one that helps us get to a certain insight and unique molecule, and the failure will have nothing to do with some of those idiosyncrasies that are outside of AI’s role.

I think it’s a false premise that until an AI-driven drug is approved, this is all hype. That mindset might lead many to be skeptical for a while and sit on the sidelines, which I believe would be a costly mistake.

Pharm Exec: AI and biotech integration were often cited as a top driver for investment and funding recovery in the sector, and recent dealmaking and partnerships, even out of Big Pharma, seem to reflect that. Do you see AI remaining a pacesetter of sorts for future growth pursuits?

Kahvejian: Put simply, companies that don't incorporate AI will be at a disadvantage and less competitive. In some ways, I don't know if AI is going to drive more investment; I think the investment dollars, especially in our field, are going to be looking for: how are you using AI? Are you using it wisely? Are you just putting it in your headline or tag line just to be part of the vernacular and the hype cycle? Or are you truly implementing it within your mindset, philosophy, and workflow?

I think the biotech space will continue to innovate and there's going be a continued investment in this arena. At Flagship, we focus on inventing and building internally, and we're amplifying those efforts.

We are seeing the tech world take interest in biotech, and the word “techbio” has been thrown around for a few years but is now gaining prominence. I do think we’ll see attempts at convergence by companies like OpenAI, Nvidia, Meta, and Google. There's going to be a convergence of these kinds of things in different philosophies with respect to: does software come first and then a wet lab comes second? Do we need any wet lab at all, in some cases?

Drugs are very complex and, at the end of the day, have to be validated, obviously, in humans. So, there will always be a necessity for wet lab validation along the way.

But we’ll continue to see more and more attempts at the convergence of putting more raw-compute power and algorithmic power toward biology—whether in small molecule design, predicting clinical trial results, patient recruitment, and beyond.

AI and R&D: From the Eyes of an Entrepreneur

Key Takeaways

Newsletter