Behind The Data: Active Learning-Guided Optimization Of Large Gene Insertion Effectors In Mammalian Cells

ElevateBio is applying active learning and machine learning to accelerate the engineering of large serine recombinases for targeted gene insertion. By training a regression model on measured protein activity rather than evolutionary fitness alone, researchers can predict which amino acid variants are most likely to improve performance. Iterative testing identified single mutants with substantially greater activity than wild-type protein, while pairwise screening produced double mutants that further increased potency. The resulting data also captured epistatic interactions, revealing which mutations work well together and supporting the design of higher-order variants.

Discover how this scalable approach offers a faster, more cost-effective path to optimizing gene insertion effectors, even when protein structures are unknown, and can be extended to other protein engineering programs with complex or multiple objectives.