Case Study
AI/ML Optimization Reduces AAV Discovery Timeline from Years to Days
Currently, gene therapy developers use empirical methods to design AAV constructs and then spend weeks to produce the vectors and test them in vitro. This approach limits developers to testing a limit number of designs possibilities, and each round of iterative optimization is consuming and labor intensive, resulting in discovery timelines of years. Here, we demonstrate how AI/ML models can be used to reduce vector discovery timelines from years to days.
We started with vector sequences that had undergone 15 rounds of empirical optimization and still had poor yield. We used FORMsightAI to explore over 100 million vector designs in silico to identify better performing combinations of genetic elements that result in reduced genome truncations and also optimized the CDS of the gene of interest for improved yield. Following AI/ML optimization, we identified 4 optimal designs that were subsequently evaluated in vitro.
After 15 rounds of emprical optimization performed in over 3 years by a publicly-traded gene therapy company, the maximum full capsid percentage achieved from a vector design was 61%. AI/ML optimization identified 4 optimal designs that were tested in parallel over approximately 4 weeks. The AI/ML optimized designs were significantly improved and achieved a maximum ull capsid percentage of 91% (Figure 1).
