Applying AI to Molecular Structure (BIO 2024)

Raphael Townshend, founder and CEO of Atomic AI, a US-based emerging biotechnology company specializing in applying the fusion of artificial intelligence (AI) and structural biology for RNA-based drug discovery, discusses the use of AI as a tool for characterizing molecular structure and, in doing so, advance molecule engineering.

“At a high level, why we care about molecular structure is because the structures of molecules really determine what they do: structure determines function, so to speak. What we found in the last few years is that AI is a very powerful tool for understanding the structures of these molecules,” Townshend says. He notes that, traditionally, methods for understanding structure have names, such as X-ray crystallography, or nuclear magnetic resonance, or cryo-electron microscopy. “The issue is that they can oftentimes be very expensive to use and slow to use; it's a very bespoke process. Every single structure that is solved can be a very manual and involved undertaking. And what we've seen is that we can take all this work that's been performed in the field of structural biology, of understanding structures, and feed that into these AI algorithms that are now able to predict those structures in minutes or seconds instead. We've seen these orders of magnitude speed up our ability to model these structures and understand what they look like in their native environments,” he explains, emphasizing that AI enables a “rational drug design”.

“We're already starting to see that and bear fruit today. The use of [AI] … speed[s] up the process of … the whole drug discovery space, of doing the whole drug discovery process, so to speak,” he observes. He further explains that, as an additional layer on top of the traditional drug discovery process, AI can enable rapid screening, or a rapid search, through many possible drug candidates. “[T]he idea is that we're using these kinds of approaches to get to better, faster drugs, and getting things faster to the clinic and running those through a better rate.”

“As I like to say, AI is a cool thing of the day. It's a powerful tool, but it's a tool nonetheless,” Townshend adds. “It's not going to be a panacea; it's not going to solve every single problem under the sun. In particular, the areas where AI works well are those where there's [a significant amount of] data associated with it … and, so, in areas where that data [are] more limited … that can be an obstacle to the development of the correct AI applications in this space,” he notes when discussing some of the hurdles in implementing AI in the drug discovery space. Townshend says that, to address these limitations, the industry would need to collect additional datasets to push progress forward.

“Some areas to really think about, for example, are in-vivo studies where the amount of animal data that we might have might be relatively limited as compared to high throughput screening stages where you have [much] data, comparatively speaking. [T]hese are resolvable obstacles, but, today, the amount of data out there does not allow for the use of AI in its most advanced form across the entire drug discovery landscape. And that's not even speaking about the development cycle, [such as] clinical trials. There's even more complexity and even more limitations interpret the data. So, while there was this dream of one day using AI across the entire process, right now, there are some key areas where it's providing significant leverage, and we're working towards shoring up those additional gaps,” Townshend says.

View the video above for the full interview with Townshend. Townshend spoke at the New Tools, New Times: Chemical and Structural AI for Drug Discovery panel at BIO 2024 in San Diego, Calif., which is being held June 3–6.