Given their ability to transfer DNA into target cells, and non-pathogenic nature, adeno-associated viruses (AAVs)—the spherical protein structures enveloping the virus’ DNA genome—are the current delivery method of choice for gene therapy. However, there are limitations with AAVs that need to be overcome in order for gene therapies to become a reality for more patients.

First, currently used AAV capsids are limited in their ability to specifically hone in on the tissue affected by a disease. In addition, patients’ immune systems, after having been exposed to a similar AAV virus, can produce neutralizing antibodies that, even at low levels, can destroy AAVs upon re-exposure (neutralization), blocking the delivery of their therapeutic DNA cargo.

Indeed, it is estimated that up to 50–70% of the human population have pre-existing immunity to natural forms of the AAV vectors currently being using to deliver gene therapies. This immunity renders a large portion of patients ineligible to receive gene therapies. Overcoming the challenge of pre-existing immunity to AAV vectors is therefore a major goal for the gene therapy field.

Now, a new study initiated by George Church, PhD, a Wyss Core Faculty member, in collaboration with Google Research and Dyno Therapeutics, has applied a computational deep learning approach to design highly diverse capsid variants from an AAV virus. The team focused on DNA sequences encoding a key protein segment that plays a role in immune-recognition as well as infection of target tissues.

They demonstrated the use of AI to generate an unprecedented diversity of AAV capsids in order to identify functional variants capable of evading the immune system. Starting from a relatively small collection of capsid data, the team generated 200,000 virus variants, and thus the most diverse collection of any virus capsid protein. The approach could lead to much improved and more efficient gene therapies.

The research is published in Nature Biotechnology in the paper, “Deep diversification of an AAV capsid protein by machine learning.

The approach described in the paper “opens a radically new frontier in capsid design,” said Eric Kelsic, PhD, Dyno’s CEO and co-founder. “Our study clearly demonstrates the potential of machine learning to guide the design of diverse and functional sequence variants, far beyond what exists in nature,” he added. “We continue to expand and apply the power of artificial intelligence to design vectors that can not only overcome the problem of pre-existing immunity but also address the need for more effective and selective tissue targeting. At Dyno, we are making rapid progress to design novel AAV vectors that overcome the limitations of current vectors, improving treatments for more patients and expanding the number of diseases treatable with gene therapies.”

The work describes the rapid production of a large library of distinct AAV capsid variants designed by machine learning models. Nearly 60% of the variants produced were determined to be viable, a significant increase over the typical yield of <1% using random mutagenesis, a standard method of generating diversity.

This research builds upon previous work in which a complete landscape of single mutations around the AAV2 capsid was generated followed by evaluation of the functional properties important for in vivo delivery.

“The more we change the AAV vector from how it looks naturally, the more likely we are to overcome the problem of pre-existing immunity,” added Sam Sinai, PhD, Dyno co-founder and machine learning team lead. “Key to solving this problem, however, is also ensuring that capsid variants remain viable for packaging the DNA payload. With conventional methods, this diversification is time- and resource-intensive, and results in a very low yield of viable capsids. In contrast, our approach allows us to rapidly unlock the full potential diversity of AAV capsids to develop improved gene therapies for a much larger number of patients.