HMN 2025: How Machine {learning} model can foretell the health of AAV capsids for gene remedy

A brand new study in Human Gene Therapy describes a machine {learning} (ML) model that can be utilized as a surrogate for laborious in vitro experiments. This in silico method goals to extend the health of medical adeno-associated virus (AAV) capsids to make gene therapies extra economically viable for sufferers.

Developing AAV capsids with improved yield, or health, is a key technique for lowering manufacturing prices to be able to make gene therapies extra reasonably priced.

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

“By combining a protein language model (PLM) and classical ML methods, our model achieved a considerably excessive prediction accuracy (Pearson correlation = 0.818) for capsid health,” said the investigators. “Importantly, checks on utterly unbiased datasets confirmed robustness and generalizability of our model, even for multi-mutant AAV capsids.”

“The emergence of synthetic intelligence (AI)-based approaches is an thrilling improvement in capsid engineering that has the potential to be extra systematic, complete, and cost-effective than conventional directed evolution and rational design-based methods. The study by Wu et al. is a good step ahead in growing AI instruments for the gene remedy discipline,” says Managing Editor of Human Gene Therapy Thomas Gallagher, Ph.D., from the University of Massachusetts Chan Medical School.