Deep Genomics founder and CEO Brendan Frey, PhD, FRSC

A Toronto developer of treatments based on artificial intelligence said today it has successfully used its own AI drug discovery platform to identify its first therapeutic candidate, a therapy for Wilson disease that the company plans to advance into clinical studies in 2021.

Deep Genomics says that its initial candidate, DG12P1, is an oligonucleotide therapy designed to treat Wilson disease in patients who possess the mutation of Met645Arg—a mutation shown to lead to the loss of function of the ATP7B copper-binding protein. The mutation is among an unknown number of mutations linked to Wilson disease, which causes copper to accumulate in the liver, brain, and other vital organs. The rare and potentially life-threatening inherited disorder affects approximately one in every 30,000 people worldwide, as noted in a 2015 review article.

In a study preprint published September 17 in bioRxiv, Deep Genomics founder and CEO Brendan Frey, PhD, FRSC, and colleagues, reported that they used a minigene system and gene-edited HepG2 cells to show that the variant—also known as NM_000053.3:c.1934T>G, and highly prevalent among Wilson Disease patients of Spanish descent—caused approximately 70% skipping of exon 6. That exon skipping, in turn, resulted in frameshift and stop gain, which is expected to cause loss of ATP7B function.

Speaking with GEN, Frey said DG12P1 was the first-ever AI-discovered therapeutic candidate, and has emerged 18 months after Deep Genomics initiated its target discovery effort—compared with timeframes of 3 to 6 years for traditional biopharma drug discovery plus preclinical development, according to a 2015 FDA presentation.

The mechanism behind Met645Arg emerged from a target discovery effort that was intended to identify not only a genetic cause for Wilson disease, but the chemical properties needed in a molecule targeting the mutation, as well as a compound to treat the mutation that could be studied further.

Three-hour search

“That mutation had been studied by human researchers for over 20 years, and it stumped them. They hadn’t been able to figure out how that mutation works. It’s amazing that our AI system figured it out in three hours,” Frey said. “This mutation is an example of a whole new class of mutations that humans really just aren’t good at understanding. Those are mutations that change the regulatory code. Those are mutations that cause the instructions that are embedded in the genome, that tell the cell how to process the gene.

“They are really hard to find, because they require understanding how the genome works, not just understanding, here’s the protein and here’s the amino acid. Humans aren’t good at figuring out those regulatory mutations, whereas the AI system is able to figure those out,” Frey elaborated.

To discover DG12P1, Deep Genomics used its AI Workbench drug discovery platform, designed to identify potential oligonucleotide therapies that target genetic determinants of disease at the level of RNA or DNA. For every compound identified—from therapeutic candidates to thousands of novel exploratory compounds—the platform can generate on-target and genome-wide off-target effect data, cell viability data, and animal toxicity data. Also collected by the platform are data relevant to biomarkers to be used as endpoints in the company’s clinical trials.

The platform narrowed down candidate selection to two dozen, then a dozen candidates before Deep Genomics zeroed in on DG12P1. Deep Genomics’ AI Workbench has enjoyed a 70% success rate of going from novel target through to candidate declaration, Frey said, with discovery timelines for other candidates under consideration in the range of 18 months as well.

“It feels like the sky is the limit, now that the system is working so well, and we’ve got our first really good story for a declared candidate,” Frey said. “In the next year or so we expect to declare another five candidates, then another seven candidates the year after that.”

Beyond rare disease

As that occurs, Frey added, Deep Genomics will broaden its therapeutic focus beyond rare diseases, to more common metabolic disorders, as well as ophthalmology and neurodegenerative diseases. The company is not saying which disease indications it will pursue within those areas.

Frey did say that in applying its AI platform, Deep Genomics studied billions of data points to train its over 22 machine learning systems.

“The AI system is exposed to genome sequences, RNAi-seq datasets, the chromatin modification datasets, datasets revealing protein-DNA interaction, protein-RNA interactions, datasets having to do with allele frequencies, so population genetics kind of data,” Frey said. “Also, datasets that have to do with the mode of inheritance of different diseases: Is this disease autosomal? Recessive? Dominant? What is the way in which the disease is inherited? Dozens and dozens of different types of datasets. And for each type of dataset, dozens and dozens of different datasets.”

Spun out of the University of Toronto, Deep Genomics was launched in 2015 and has grown since then to more than 40 employees with advanced degrees and industry experience in artificial intelligence, automation, cell and molecular biology, clinical development, in vitro disease models, machine learning, medicine, molecular genetics, preclinical development, organic chemistry, and software engineering.

In June, Deep Genomics bolstered its clinical trial design and regulatory strategy expertise by naming as a strategic advisor Peter Barton Hutt, senior counsel at the law firm Covington & Burling, and a former chief counsel of the FDA. And in September 2017, Deep Genomics completed a $13 million Series A financing round led by Khosla Ventures, with participation by True Ventures.

Deep Genomics honed its AI technology by successfully using its platform to evaluate over 69 billion molecules against one million targets, in silico.  The effort, dubbed “Project Saturn,” was designed to generate a library of 1,000 compounds that were experimentally verified to manipulate cell biology as intended.

“Project Saturn was really developing the technology that enabled us to assess computations broadly that had an impact on the molecular mechanism called splicing,” Frey recalled. “Saturn was kind-of our moonshot, in terms of developing this new technology. It worked out really well, and now we’re using it to make drugs.”

Previous articleFidgeting May Help Mice Complete Their Tasks
Next articleLong Reads and CRISPR Couple as Oxford Nanopore and Caribou Strike Deal