The biotech industry is experiencing a rush of AI-powered tools for many aspects of the complex drug discovery process. But one that has flown under the radar, increasingly believed to be the key to certain diseases but woefully underexposed, is RNA. With $35 million in new funding, Atomic AI wants to do for RNA what AlphaFold did for proteins, and thereby find completely new treatments.
If you can remember your high school biology, you probably remember RNA as a kind of intermediary between DNA (long-term storage of information) and proteins (the machinery of cellular life at the molecular level). But like most things in nature, it doesn’t seem to be that simple, Atomic AI CEO and founder Raphael Townshend explained.
“There’s this central dogma that DNA goes to RNA, which goes to proteins. But in recent years it’s been shown to do a lot more than just encode information,” he said in an interview with ukbusinessupdates.com. “If you look at the human genome , about 2% becomes protein at some point. But 80 percent becomes RNA. And it does… who knows what? It’s massively underdeveloped.”
Little work has been done in this area compared to DNA and proteins. Academia has focused on other pieces of the puzzle, and pharmaceuticals, in part as a result, have pursued proteins as drug delivery mechanisms. The result is a serious lack of knowledge and data on RNA structures.
But what Atomic AI argues is that RNA is functional and worth pursuing as a method of treatment. The secret is in the “non-coding” regions of RNA, which resemble the header and footer of a document. They do protein-like work, but aren’t proteins — and they’re not the only example.
You can think of RNA strands as strings of beads, much more rope than beads. The string is “floppy” and more or less what its detractors think it is: an intermediary. But every once in a while you get a really interesting knot that probably didn’t happen by accident. Just like with proteins, being able to figure out their structure goes a long way in understanding what they do and how they can be influenced.
“The key is to find those beads, those textured pieces. It contains a lot of information, it’s targetable and it’s probably functional too,” said Townshend. “It’s seen as an important new frontier in drug discovery.”
Perhaps an interesting idea for a thesis (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 out the methods of study has a lot of value. Then when you build those methods, you can be first in line to use them. Atomic AI does both simultaneously.
At the core of Atomic’s IP, while this is something of an oversimplification, is an 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 limited set of a type of molecule that can make accurate predictions about the structure of other molecules of that type.
What’s wild is that Townshend’s team created just such a model, which far outperforms others, giving it the features of just 18 RNA molecular structures “published between 1994 and 2006.” This absolutely bald model wiped the floor with others, as revealed in a front-page article published in Science in 2021.
Since then, Townshend was quick to add, the company has vastly expanded its models and methods to include more raw material, much of which it has made itself in its own wet laboratories. They call the updated set of tools PARSE: Platform for AI-driven RNA Structure Exploration.
“The Science paper represented a first breakthrough, but we actually have an enormous amount… structure-adjacent data,” he explained. “Not the entire structure itself, but data related to the structure, tens of millions of data points; the same data scale you need to train large language models. And when combined with other machine learning work, we have been able to dramatically improve both the speed and accuracy of the paper.”
That means Atomic AI is the only one that has, at least publicly, a system that can take the raw data of an RNA molecule and spit out a reasonably reliable estimate of its structure. That’s useful for anyone doing RNA research, in or out of medicine, and with gene therapies and mRNA vaccines, the field is certainly on the rise.
You can work with such a tool in two ways: license it as a “structure as a service” platform, as Townshend put it, or use it yourself. Atomic has chosen the latter and is running its own drug discovery program.
This approach is significantly different from many of the AI discovery processes out there. The general idea is that you have a protein, say one that you want to inhibit the expression of 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 to ( and inexpensive, if possible).
AI drug discovery efforts tend to produce thousands, millions, even billions of candidate molecules that power work, rank them and let the wet labs work their way through the list as quickly as possible. If you can find one that meets the above characteristics, you can produce a new drug or replace a more expensive drug on the market. But most importantly, you compete to find new binders for a known protein.
“We don’t just find binders, we find what is target in the first place. The reason that’s interesting is that these big pharmaceutical companies ultimately care more about new biology than new molecules. You’re enabling something that wasn’t possible before by finding this new target, as opposed to increasing the number of molecules available to target it,” Townshend said.
Not only that, but some proteins have been found to be nearly incurable for some reason, causing disease that is resistant to drugs. RNA could enable the treatment of the same diseases by making an end run around the problem protein.
For now, Atomic AI has narrowed the list down to certain cancers that lead to pathological protein overproduction (thus good options to prevent the mechanism), and neurodegenerative diseases that could also benefit from upstream intervention.
Of course, all this work is extremely costly, as it requires a large amount of lab work and intensive data science. Thankfully, the company raised a $35 million round led by Playground Global, with participation from 8VC, Factory HQ, Greylock, NotBoring, 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 already picked the low-hanging fruit in protein land,” Townshend said. “Now there’s new biology to go after.”