US-based start-up Atomic AI is using artificial intelligence (AI) to design more intelligent RNA-targeted and RNA-based medicines. Diana Spencer spoke to Raphael Townshend, PhD, Founder and CEO of Atomic AI, about the advantages of combining two such exciting technologies.
Artificial Intelligence (AI) has been enthusiastically adopted by the drug discovery sector as a tool for shortening drug discovery cycle times and increasing success rates. With the capability to process vast data sets quickly, machine learning (ML) is being used to identify potential drug candidates, repurpose existing drugs and potentially reduce the need for preclinical and clinical testing.
The technology can also generate synthetic data sets, or make human trials more efficient, by identifying the most suitable patients, predicting outcomes and optimising dosages and regimens of drugs.
As tech experts call for improved regulation and a halt to the development of AI, we take a look at how one company, Atomic AI, is using it to develop RNA-targeting treatments for a range of diseases.
AI-driven RNA Structure Exploration (PARSE)
Atomic AI launched at the start of 2023 with a $35 million series a financing, having previously raised previously raised a $7 million in a seed round.
The company’s AI-driven 3D RNA structure engine generates RNA structural datasets, integrating machine learning foundation models with experimental wet-lab biology to reveal functional binders to RNA targets.
The technology can predict structured, ligandable RNA motifs at high speed and accuracy. By uniting these algorithmic advances with large-scale experimental biology, founder Raphael Townshend and colleagues can design novel RNA-targeting and RNA-based medicines to treat currently undruggable diseases.
The concepts behind the company’s platform are based on Townshend’s own PhD thesis on applying machine learning to the field of structural biology.
He explains how they overcame the current issues with RNA drug discovery: “A major barrier for the entire field is the limited RNA structural datasets that can be fed into AI models. However, Atomic AI has a unique structural engine that enables more intelligent and rational design of RNA-targeted and RNA-based medicines. This is what we mean by the ‘next generation of RNA drug discovery’. We can use our structural tools and AI-driven methodologies to create an entirely new way of identifying drug targets.”
Another key challenge has been how to target a known RNA transcript associated with an ‘undruggable disease’ with no knowledge of where to start on the transcript, or more specifically, which parts of the transcript are even targetable through various therapeutic modalities.
“By identifying which parts are well structured, we are essentially pinpointing which transcripts are druggable, thus enabling us to selectively target those regions with functional molecules,” Townshend explains. “This can potentially make ‘undruggable diseases’ become ‘druggable’.”
The limitations of AI in drug discovery
While AI is a powerful tool when used appropriately, it does have its limitations and Townsend cautions against treating AI or ML as a quick fix that can solve every problem. Realising where the technology can make dramatic improvements is vital, he emphasises, such as when you have significant amounts of data you can exploit, an AI learning algorithm can do better than any hand-designed, manual computational system.
The issue is that a huge amount of data needs to be collated, processed and ‘fed’ into the model to enable it to ‘learn’ the dataset and then process it accurately. One solution to this is to use foundation models, which are trained on a broad set of unlabeled data that can be used for different tasks. The AI can apply knowledge across different applications and situations, making it highly adaptable.
Townsend explains the advantages of this approach: “For the biotech sector, we can only expect these algorithms to be as good as the data fed into the technology. To an extent, we have the opportunity to get around this challenge by using foundation models where we can generalise relatively easily through training core algorithms, even with limited data. For Atomic AI, we can potentially achieve generalisation by understanding and exploiting the underlying structure of RNA molecules. This means we can train the physics-informed AI algorithms on a limited set of RNA then use them to analyse completely novel RNA transcripts.”
Future possibilities
With the creation of more intelligent generative AI models like ChatGPT has come increased concern about how advanced the technology could become and the lack of regulation in its current development.
However, as far as drug discovery goes, AI remains a tool that can only benefit the health of mankind in the future, by speeding up the development of drugs for currently untreatable diseases.
Townshend is optimistic about the possibilities: “We’re still scratching the surface. Much of the high-level AI research has not been focused on drug discovery or the biotech space, mainly because these areas have been inaccessible to the general AI/ML community. By bridging the gap and bringing world-leading AI researchers into the life sciences, we have the opportunity to create safer and more effective therapeutics for a broad range of diseases.”
Atomic AI plan to use their engine to create their own pipeline of small molecule drug candidates and seek partnerships for their technology. In addition, they hope to pursue RNA-targeting small molecules in some of the potential therapeutic areas of oncology, neurodegeneration, neuromuscular disease, infectious disease, rare disease and immunology in the future.
Townshend is also looking forward to seeing further developments in the RNA space, inspired by successes like the Covid-19 vaccines and the approval of the first-ever RNA-targeting small molecule, risdiplam, both of which have demonstrated the enormous potential of the platform.
“Over the next five years, there will be incredible developments in the next generation of RNA-based medicines such as mRNA vaccines and circular RNA, and RNA-targeting medicines, like small molecules and antisense oligonucleotides. We will also see improved gene therapies and advanced CRISPR guide RNA design tools. Synergising RNA with AI enables further progress on the RNA drug discovery frontier. It is excellent timing to combine these technologies.”
