AI Model Designs Proteins for Gene Therapy Delivery

Gene therapy is a promising technique that can treat various diseases by introducing genetic material into the cells of a patient. However, one of the major challenges of gene therapy is how to deliver the genes to the target cells without triggering an immune response or causing side effects.

One of the most common methods of gene delivery is using viral vectors, which are modified viruses that can carry genes into the cells. Among the different types of viral vectors, adenoviruses are widely used because they can infect a broad range of cell types and carry large DNA payloads.

However, adenoviruses also have a major drawback: they are recognized by the human immune system, which can neutralize them and cause inflammation. This reduces the efficacy of gene therapy and increases the risk of adverse reactions.

To overcome this problem, researchers at the University of Toronto have used an artificial intelligence (AI) framework to redesign a crucial protein involved in the delivery of gene therapy. The protein is called hexon, and it forms the outer shell of the adenovirus. Hexon is responsible for binding to cell receptors and mediating entry into the cells, but it is also the main target of antibodies that can block or destroy the virus.

The researchers wanted to design new variants of hexon that are distinct from natural sequences and are not recognized by the immune system. They used an AI-based approach called ProteinVAE, which can generate novel protein sequences based on a limited amount of data.

ProteinVAE is a type of generative model that can learn the characteristics of a protein family and produce new members that are similar but not identical to the existing ones. The model uses a pre-trained language model to encode the input sequences into a latent space, where each point represents a possible protein sequence. Then, it uses a convolutional neural network and self-attention mechanisms to decode the latent points into output sequences.

The researchers trained ProteinVAE on 1,200 natural hexon sequences from different adenovirus serotypes. They then used the model to generate 10,000 new hexon sequences that were different from any known human or animal variants. They selected 20 candidates based on their diversity and stability and tested them in vitro for their ability to bind to cell receptors and evade antibodies.

The results showed that some of the designed hexons had improved properties compared to natural ones. They had higher affinity for cell receptors, lower immunogenicity and higher stability. The researchers also showed that some of the designed hexons could be incorporated into adenovirus vectors and enhance their gene delivery efficiency.

The study, published in Nature Machine Intelligence , demonstrates how AI can be used to design new proteins for gene therapy and other biomedical applications. The researchers believe that their approach can be extended to other viral proteins and other types of vectors, such as lentiviruses or adeno-associated viruses.