With New Funding, Atomic AI Envisions RNA as the Next Frontier in Drug Discovery TechCrunch
The biotechnology industry is seeing a rush to use AI-powered tools for many aspects of the complex drug discovery process. But one that has flown under the radar, and which is increasingly thought to be key to certain diseases but is woefully under-studied, is RNA. With $35 million in new funding, AI offspring It aims to do for RNA what AlphaFold did for proteins, creating entirely new therapies in the process.
If you still remember your high school biology, you probably remember RNA as a kind of intermediate between DNA (long-term storage of information) and proteins (the mechanism of cellular life at the molecular level). But like most things in nature, it doesn’t look so simple, says Raphael Townsend, CEO and founder of Atomic AI.
There’s this central dogma that DNA goes into RNA, which goes into proteins. But it turns out in recent years that it does much more than just encode information,” he said in an interview with TechCrunch. “If you look at the human genome, about 2% turns into a protein at some point. But 80 percent become RNA. And she does… who knows what? It’s pretty much unexplored.”
Compared to DNA and proteins, little work has been done in this area. Academia has focused on other pieces of the puzzle, and pharmaceuticals have, in part as a result, pursued proteins as drug mechanisms. The result is a severe lack of knowledge and data about RNA structures.
But what Atomic AI posits is that RNA is functional and worth pursuing as a method of treatment. The secret lies in the “non-coding” regions of the RNA, which are like the header and footer of a document. They do a protein-like job but aren’t proteins — and they’re not the only example.
You can think of strands of RNA as a beaded necklace, a thread that is much more than a bead. The thread is “elastic” and that’s what its critics think: a medium. But every now and then you get really interesting knots that seem very unlikely to have formed by accident. As with proteins, if you can learn their structure, that goes a long way toward understanding what they do and how they might be affected.
“The key is finding those beads, those organized pieces. It’s high in information content, it can be targeted, it’s potentially functional as well,” Townsend said. “It’s seen in drug discovery as a major new frontier.”
An interesting idea for a graduate thesis, perhaps (and it was for Townshend), but how do you build a business around it?
First, if the field is about to become more important, building methods of study is very valuable. Then, if you build these methods, you can be the first to use them. Atomic AI does both at once.
The core of Atomic’s IP, though this is a bit of a simplification, is AlphaFold for RNA. The biology is different, and the way the models work is different, but the idea is the same: a machine-learning model trained on a finite set of a molecule type can make accurate predictions about the structure of other molecules of that type.
The catch is that Townshend’s team made a model like this, which outperforms the others by a large margin, by feeding it the characteristics of just 18 RNA molecular structures “published between 1994 and 2006.” This completely nude model is mopping the floor with others, as revealed in a front-page article published in Science in 2021.
Since then, Townshend quickly added, the company has greatly enhanced its models and styles with more raw materials, many of which it creates on its own in its wet labs. They call the updated set of tools PARSE: a platform for AI-powered RNA structure exploration.
“The scientific paper represents an initial breakthrough, but we have actually created an enormous amount of… structure-adjacent data,” he explained. “Not the whole structure itself, but data about the structure, tens of millions of data points; The same amount of data you need to train large language models. Combined with other machine learning work, we were able to significantly improve the speed and accuracy from paper.”
This means that only Atomic AI has, publicly at least, a system that can take the raw data of an RNA molecule and output a reasonably reliable estimate of its structure. This is useful for anyone doing RNA research in or out of medicine, and with gene therapies and mRNA vaccines, the field is definitely on the rise.
With this tool, you can go one of two ways: license as a “structure as a service” platform, as Townshend said, or use it yourself. Atomic has chosen the latter, and is pursuing its own drug discovery program.
This approach is markedly different from a lot of the AI discovery processes out there. The general idea is that you have a protein, say one that you want to prevent from being expressed in the human body, but what you don’t have is a chemical that binds reliably and exclusively to that protein, exactly where and when you want it (and cheaply, if possible).
AI drug discovery efforts tend to produce thousands, millions or billions of candidate molecules may be Work, rank them, and let the wet labs start working through the list as fast as they can. If you can find a drug that meets those above characteristics, you can produce a new drug or replace a more expensive drug on the market. But the main thing is that you are competing to find new bonds to a known protein.
“We’re not just finding binders, we’re finding what’s targetable in the first place. The interesting reason is that, at the end of the day, these big drugs are more interested in new biology than they are in new molecules. You’re enabling something that wasn’t possible before,” Townsend said. accept by finding this new target, rather than increasing the number of molecules available to target it.
Not only that, but some proteins have been found to be close to being ineligible for whatever reason, and to produce drug-resistant diseases. RNA could allow these same diseases to be treated by putting a stop to the problematic protein.
At present, Atomic AI has narrowed the list down to some cancers that lead to pathological overproduction of proteins (and thus good options for anticipating the mechanism), and neurodegenerative diseases that might also benefit from initial intervention.
Of course all this work is very expensive, requiring a great deal of lab work and intensive data science. Fortunately, the company has raised $35 million, led by Playground Global, with participation from 8VC, Factory HQ, Greylock, NotBoring and AME Cloud Ventures as well as angels Nat Friedman, Doug Mohr, Neal Khosla and Patrick Hsu. (The company previously raised a $7 million seed round.)
“People have picked all the low-hanging fruit from protein land,” Townsend said. “Now there is a new biology to pursue.”