Case Study
Improving AAV Manufacturability with AI/ML Simulation to Reduce Genome Truncations
Partially filled AAV capsids are a major challenge in gene therapy development, causing manufacturability risks that can lead to the failure of drug programs. One way to reduce the amount of partially filled capsids in AAV production is to design vectors with a reduced propensity for creating truncated genomes during replication. Here, we evaluated whether AI/ML models could be used to simulate the performance of promoter sequences to help researchers select the promoter with the lowest risk for genomic truncation.
We used FORMsightAI to simulate the performance of genetic elements in two vector designs. In both designs, all elements, including the GOI CDS, enhancer and polyA tails, were preserved, except for the promoters (P1 and P2). The vectors were then packaged into AAV5 capsids and manufactured by a leading CDMO. Subsequently, we used PacBio long-read sequencing to sequence the AAV genomes and anlayzed the data to identify truncation products.
FORMsightAI simulation of AAV5_P1_PRO1_GOI and AAV5_P2_PRO1_GOI constructs predicted approximately similar truncation risk profiles between the two vector designs at the CDS, enhancer, polyA and other genomic regions. However, the promoter region of P1 was projected to result in increased truncation risk (2.4%), when compared to P2 (1.4%). Therefore, P1 had a ~40% higher truncation risk when compared to P2 for expression of the GOI (Table 1). Our predictions were confirmed using long-read sequencing data, which indicated that AAV5_P1_PRO1_GOI had more genome truncations in the promoter element than AAV5_P2_PRO1_GOI (Figure 1).
