Gene therapies deliver healthy genetic material to patients’ cells to replace a mutated, disease-causing variant. Dyno Therapeutics aims to create better delivery vehicles for those genes by using machine-learning tools to engineer new types of harmless viruses that are more effective and simpler to manufacture.
The Cambridge, MA-based biotech emerged from stealth Monday having signed deals with two drug makers who want to apply its technology to their gene therapy efforts—agreements its founders say are lucrative enough to fund the company for years to come. Dyno spun out of George Church’s lab at Harvard Medical School in late 2018 with $9 million in seed funding from Polaris Partners and CRV.
 |
Want more AI, machine learning, and drug discovery content? Network and engage with industry leaders online this June. Learn more. |
Now Dyno has signed collaboration agreements with Novartis (NYSE: [[ticker:NVS]]) and Sarepta Therapeutics (NASDAQ: [[ticker:SRPT]]) to design superior versions of the adeno-associated viruses, or AAVs, commonly used in gene therapies, and the company says the deals may preclude it from needing to raise outside funds again. That’s a rarity for a biotech that is already considering advancing its own product candidates in addition to striking R&D arrangements with other biotechs and pharmaceutical companies.
The first gene therapy in the US was approved in 2017. Finding better AAVs in which to insert genetic material has proven challenging because tweaking the complex protein shells to improve one property, such as targeted delivery, can impede others, such as their ability to evade the immune system.
“With small molecules and antibodies, the drug is the small molecule or the antibody, and sometimes you’ll require a delivery system but … that’s not the important part,” Polaris partner Alan Crane, one of Dyno’s founders, said in an interview. “It’s almost flipped in gene therapy, because everyone knows what gene to deliver, generally, for a particular disease, but they don’t know how to deliver it, and we really need these better vectors.”
Dyno will work with Novartis to use its CapsidMap platform to design superior AAV vectors for gene therapies for eye diseases; Sarepta’s area of interest is muscle disease.
Under the terms of the deals, the larger companies will be responsible for taking any gene therapy product candidates created within the collaborations through preclinical and clinical testing and commercialization. In the Novartis agreement, Dyno gets an undisclosed amount of money up front plus research funding and license fees. If any products advance, it will be eligible for payments tied to clinical, regulatory, and sales milestones, plus royalties on sales.
In the Sarepta agreement, during the research phase of the collaboration, Dyno will be eligible for more than $40 million in payments. If candidates are developed, it will be eligible for additional payments tied to development milestones, plus royalties. The company wouldn’t disclose additional financial details, but said if everything goes according to plan, the deals it has struck to date could collectively bring the startup more than $2 billion.
CEO Eric Kelsic said that when he joined Church’s lab, he aimed to use his experimental and computational biology background to combine the latest technologies in high-throughput biology, advanced machine learning, and protein engineering. He was exposed to gene therapy through Church’s work with the gene editing technology CRISPR-Cas9, one of the tools being deployed to address genetic diseases.
“It just seemed like the perfect application of the technologies we had been working on, which are certainly going to transform all of protein engineering,” Kelsic said in an interview.
The company takes available data, plus more it generates itself using high-throughput measurement technologies, and uses it to build machine-learning models that suss out the most optimal synthetic capsids. Using machine learning allows Dyno to take into account a number of important properties—delivery, immunity, packaging size, and manufacturing, for example—while weighting those most important to the particular disease indication. That’s compared to today’s efforts, which generally can only tweak one property at a time.
“It’s been the case in the past that when you only select for one property, for example the efficiency of delivery, that might be improved, but the manufacturability might actually get more challenging,” Kelsic said. “That’s a tradeoff which has really limited engineering efforts in this space, but now we can overcome that using our platform, optimizing for both the efficiency and the manufacturability and so on, across all the different properties that are important.”
Machine-learning tools also allow each experiment to build upon the findings of previous iterations.
When Crane and Kelsic first met, in June 2018, the scientist presented the investor with a spreadsheet of companies that were interested in what the startup had developed. That was before Dyno had published any of its results.
Before joining Polaris in 2002, Crane headed corporate development at Millennium Pharmaceuticals. (Takeda Pharmaceutical (NYSE: [[ticker:TAK]]) acquired Millennium in 2008 for $8.8 billion). Since joining the VC firm in 2002, Crane has been founder, chairman, or CEO of seven of the companies it has started.
“We and a lot of folks in the VC world have been looking at AI applications to healthcare, and increasingly to life sciences and biology, but this was by far the best application to biology that I have ever seen,” Crane said. “I’ve seen a lot of business development, and I’ve never seen such robust interest in a young company platform.”
Crane says Dyno is currently in talks with other companies that will likely lead to one or two more partnerships like those it has struck with Novartis and Sarepta. The company says it won’t allow any company to use its tech to explore a certain disease area exclusively so that it can potentially be used across the industry. In addition to eye and muscle diseases, company is also looking at how to design better vectors for disorders of the central nervous system and the liver.
Along with Kelsic, Church, and Crane, Dyno’s co-founders are Sam Sinai, its lead machine learning scientist, Adrian Veres, and Tomas Bjorklund.
