Nat. Biotechnol. | 面向先导编辑结果预测的机制机器学习

TL;DR - OptiPrime embeds prime-editing mechanisms into a machine-learning model to predict pegRNA efficiency more accurately and transfer across editing systems. It accelerated therapeutic design and enabled over 40% average Kif1a correction in mouse cortex.

• Trained on 297,962 measurements across 40 experimental settings, OptiPrime outperformed DeepPrime-FT and PRIDICT2.0 on independent tests.
• The model represents Cas9 activity, reverse transcription, flap formation, mismatch repair, and other DNA-repair steps as interpretable pseudo-rates.
• Mechanistic modules generalized to PE3 and twinPE despite not being directly trained on those systems.
• In vivo AAV9 delivery achieved over 70% correction in transduced mouse brain cells after a four-week optimization process.