Nat. Biotechnol. | 稀疏数据驱动的自适应模型引导蛋白进化优化紧凑型真核基因组编辑器

TL;DR - A Nature Biotechnology study introduces EvoMax, a sparse-data, model-guided protein evolution framework that produced a compact Fanzor genome editor with substantially improved mammalian editing. It shows that combining limited experimental data with protein language and structural models can navigate protein fitness landscapes, while in vivo toxicity highlights the need to optimize safety alongside activity.

• EvoMax combines Gaussian process regression trained on 209 measured mutations with ESM-2 evolutionary priors and ESM-IF structural compatibility, iteratively testing only about 10–20 candidates per round.
• The optimized FanzMAX v3-hLa system reached up to 97% editing at its best endogenous site and averaged roughly 33% across 19 sites—over 2.6-fold higher than two existing compact editors.
• Model-guided mutations, engineered ωRNA, and an hLa fusion jointly improved activity, broadened TAM compatibility, and restored function in several naturally inactive Fanzor2 homologs.
• Single-AAV delivery edited mouse liver PCSK9, but the most active constructs caused toxicity and more large genomic deletions, demonstrating a critical activity–safety tradeoff.