Deep Genomics debuts AI model for debugging the body's RNA code and finding paths to potential therapies

Deep Genomics is taking two emerging technologies—which in recent years have captured the public’s attention for different reasons—and is finding they can work great together.

The company has put forward an artificial-intelligence-powered foundation model designed to explore RNA biology, and how the small pieces of genetic material can contribute to various diseases or provide avenues for new therapies. 

AI foundation models, as a category, describe systems trained on massive amounts of information that can then be tailored to produce a variety of different outputs—this family includes the recently popular text and image generators such as ChatGPT and other programs. RNA research, meanwhile, famously demonstrated its potential in the COVID-19 vaccines developed by Pfizer, BioNTech and Moderna, and is currently being explored for a range of diseases.

Deep Genomics’ AI, dubbed BigRNA, is a transformer neural network designed to predict the biological mechanisms that regulate RNA expression tissue-by-tissue, to help better understand how specific variants in genes ultimately give rise to different diseases.

At the same time, the program can identify potential binding sites for proteins and microRNAs, which could possibly be exploited for the discovery and development of new drug compounds.

“Building machine learning models that can predict gene expression from DNA sequence has been a long-standing research goal, and one that has seen significant strides owing to recent advancements in deep learning,” Brendan Frey, Deep Genomics’ founder and chief innovation officer, said in a statement.

After being given a DNA or RNA sequence, the AI can examine the effects of non-coding genes, as well as mutations that may or may not alter the ultimate function of a particular protein, and then help design matching RNA-based therapeutics, Frey said.

“In a comprehensive study of 14 genes, BigRNA consistently designed highly effective steric blocking oligonucleotides that act in a tissue-specific manner, including for genes involved in Wilson disease and spinal muscular atrophy,” Frey added. 

By intervening in how cells process RNA, these oligonucleotides could potentially help reverse the effects of a disease-causing genetic variant, by blocking the overproduction of a troublesome protein or increasing the production of a deficient one.

BigRNA was built out of 1 trillion genomic signals captured from thousands of data sets, and includes 1.8 billion tunable parameters, the company said. The details of the model were published on the journal preprint site BioRxiv. The company said it also plans to present its data at future scientific meetings. 

The computer model looks to deliver on Deep Genomics’ long promise that AI can help deliver gene-based therapies that can be programmed for different objectives. The Toronto-based company raised $180 million in a 2021 series C venture capital round with a pitch to advance research on at least 10 computer-designed drugs. 

Currently, its pipeline lists a handful of undefined preclinical programs in metabolic and central nervous system diseases, though it previously identified Wilson disease as a target—a rare genetic condition where excess copper is stored within the body’s tissues.

Earlier this month the company made a change to its leadership. It appointed Brian O’Callaghan to serve as CEO, while Frey transitioned from the position to chief innovation officer.

“I’m proud of the team, the progress we have made, and the molecules we are advancing through animal studies, and I look forward to sharing updates on recent achievements in the months ahead,” Frey said in a September 15 release.

O’Callaghan most recently held the top spot at ObsEva, the women’s-health biotech that restructured its C-suite earlier this year in a bid to save cash, ahead of its delisting from the Nasdaq in March. He has also served as CEO of Petra Pharma, Acucela, Sangart and BioPartners.