Gene therapy holds immense promise in the field of medicine, offering the potential to cure genetic diseases by delivering new genes to specific cells. However, one of the major challenges in gene therapy is the efficient and safe delivery of these new genes to target cells. Adeno-associated viruses (AAVs) are commonly used as gene-delivery vehicles, but the process of engineering these AAVs can be slow and inefficient.
In a groundbreaking study published in Nature Communications in 2024, researchers at the Broad Institute of MIT and Harvard have developed a systematic multi-trait protein optimization paradigm that promises to revolutionize AAV engineering for gene therapy. This innovative approach utilizes machine learning to design the protein shells of AAVs, known as capsids, with multiple desirable traits. Unlike traditional methods that focus on optimizing one trait at a time, this new approach allows researchers to engineer capsids with a combination of traits, such as organ-specific targeting and cross-species compatibility.
The team at the Broad Institute used their approach to design capsids for AAV9, a commonly used type of AAV, that efficiently targeted the liver and could be easily manufactured. Through their machine learning models, they were able to predict the behavior of these engineered capsids in human liver cells and in macaque monkeys, demonstrating the potential for translating gene therapies across species.
One of the key advantages of this new approach is its ability to streamline the process of AAV engineering. Traditional methods involve generating large libraries of capsid variants and testing them in multiple rounds of selection, which can be time-consuming and costly. By leveraging machine learning, researchers can design AAVs with multiple desired functions simultaneously, significantly accelerating the development of gene therapies.
The researchers at the Broad Institute created a library of capsids with multiple desired functions, including manufacturability and liver targeting, with nearly 90% of these capsids displaying all desired functions simultaneously. This success highlights the potential of machine learning in protein design and optimization for gene therapy applications.
Looking ahead, the researchers envision a future where their machine learning models can be used to create gene therapies that target specific organs or avoid certain tissues. They also hope that their approach will inspire other labs to generate their own models and libraries, ultimately leading to a machine-learning atlas that can predict the performance of AAV capsids across a wide range of traits.
In conclusion, the systematic multi-trait protein optimization paradigm developed by the researchers at the Broad Institute represents a significant advancement in the field of gene therapy. By combining machine learning with traditional wet lab experiments, this approach has the potential to revolutionize the design and optimization of gene-delivery vehicles, bringing us one step closer to realizing the full potential of gene therapy in treating genetic diseases.