Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

AI Designs Better Gene Therapy Delivery Viruses

Adeno-associated virus

Adeno-associated virus (Jazzlw, wikimedia Commons)

12 Feb. 2021. An academic-industry team used machine learning to design more effective variations of common benign viruses that deliver gene therapies. Researchers from Dyno Therapeutics, Google Research, and the Wyss Institute at Harvard University describe their techniques in yesterday’s issue of the journal Nature Biotechnology (paid subscription required).

Gene therapies are a promising new type of treatment for a growing number of inherited diseases, where healthy genes replace inherited mutations responsible for disease. Many current gene therapies use adeno-associated virusesbenign and naturally occurring microbes that infect cells, but do not integrate with the cell’s genome or cause disease, other than at most mild reactions.

In their natural state, however, adeno-associated viruses, or AAVs, an imperfect and inefficient delivery vehicles, interrupted in some cases by immune reactions by patients receiving gene therapies. Also, the capsids that make up the outer protein shell of AAVs offer only a few ways of targeting cells and tissue in the body. The researchers in this project sought ways of addressing these limitations in AAVs for gene therapies using machine learning in neural networks, a form of artificial intelligence.

The team led by Harvard and Wyss Institute genetics professor George Church, Google’s Lucy Colwell, and Dyno Therapeutics CEO Eric Kelsic used machine learning to investigate more than 201,000 possible variations in amino acid sequences in natural-state capsid proteins. From these variations, the researchers designed nearly 111,000 engineered candidate capsid shells, with more than 57,000 of these designed capsids surpassing the diversity of capsids in their natural state. Each of the more diverse capsids expressed 12 to 29 mutations in their amino acid sequences.

Algorithms design optimized capsids

Church, who heads Harvard’s synthetic biology platform where the research began says in a Wyss Institute statement, “It shows that neural networks combined with the high-throughput synthetic testing developed in our lab is changing the way we design gene delivery vehicles and protein drugs.”

Dyno Therapeutics is a spin-off company from Church’s lab founded by Kelsic, Church, and co-author Sam Sinai in 2018. The company extends the academic lab’s work in a technology called CapsidMap that uses machine learning algorithms to design optimized AAV capsids. The algorithms find millions of optimal combinations of targeting ability, payload size, immune evasion, and manufacturing capability, then give each variation a unique DNA identifier. The optimized capsids are then assembled to meet specific therapeutic needs, with each design adding to and refining the algorithms’ experience.

“Our approach achieves the highest functional diversity of any capsid library thus far,” notes Kelsic. “It unlocks vast areas of functional but previously unreachable sequence space, with many potential applications for generating improved viral vectors, like AAVs with much reduced immunogenicity and much improved target tissue selectivity, and also for highly efficient gene therapies.”

Science & Enterprise reported several times on Dyno Therapeutics, most recently on licensing agreements between the company and drug makers Novartis and Sarepta Therapeutics in May and Roche Group in October 2020. The two deals could bring Dyno Therapeutics as much as $3.5 billion if all terms are met.

More from Science & Enterprise:

*     *     *

4 comments to AI Designs Better Gene Therapy Delivery Viruses