Machine learning model to predict the fitness of AAV capsids for gene therapy

A new study in Human Gene Therapy describes a machine learning (ML) model that can be used as a surrogate for laborious in vitro experiments. This in silico approach aims to increase the fitness of clinical adeno-associated virus (AAV) capsids to make gene therapies more economically viable for patients.

Developing AAV capsids with improved yield, or fitness, is a key strategy for reducing manufacturing costs in order to make gene therapies more affordable.

Christian Mueller and co-authors from Sanofi describe a state-of-the-art ML model that predicts the fitness of AAV2 capsid mutants based on the amino acid sequence of the capsid monomer.

“By combining a protein language model (PLM) and classical ML techniques, our model achieved a significantly high prediction accuracy (Pearson correlation = 0.818) for capsid fitness,” stated the investigators. “Importantly, tests on completely independent datasets showed robustness and generalizability of our model, even for multi-mutant AAV capsids.”

“The emergence of artificial intelligence (AI)-based approaches is an exciting development in capsid engineering that has the potential to be more systematic, comprehensive, and cost-effective than traditional directed evolution and rational design-based strategies. The study by Wu et al. is a great step forward in developing AI tools for the gene therapy field,” says Managing Editor of Human Gene Therapy Thomas Gallagher, Ph.D., from the University of Massachusetts Chan Medical School.