Mechanistic machine learning for prediction of prime editing outcomes

Area of Science:

• Genomic medicine
• Computational biology
• Molecular biology

Background:

• Prime editing (PE) enables precise genomic DNA modifications but requires extensive optimization of PE guide RNA (pegRNA) sequences for efficient application.
• Understanding the mechanisms governing PE efficiency, including DNA repair pathways like mammalian mismatch repair (MMR), is crucial for improving its utility.

Purpose of the Study:

• To develop OptiPrime, a machine learning model predicting PE efficiency based on mechanistic insights.
• To enable accurate prediction of outcomes for different PE formats (PE3, twinPE) and identify strategies to enhance editing efficiency.
• To demonstrate the therapeutic potential of OptiPrime in preclinical models.

Main Methods:

• Developed OptiPrime, a machine learning model trained on PE efficiency data.
• Validated OptiPrime's ability to predict PE efficiency and identify determinants of mammalian mismatch repair (MMR).
• Applied OptiPrime to nominate MMR-evasive silent edits for improved PE efficiency in human and mouse cells.
• Utilized OptiPrime for in vivo correction of a pathogenic mutation in a mouse model.

Main Results:

• OptiPrime achieved state-of-the-art accuracy in predicting PE efficiency across various formats (PE, PE3, twinPE).
• The model successfully learned the influence of MMR on PE outcomes, enabling the design of MMR-evasive edits.
• OptiPrime facilitated efficient in vivo correction of a pathogenic mutation in a mouse model of KIF1A-associated neurological disorder.
• Demonstrated successful application in prospective therapeutic contexts in primary human and mouse cells.

Conclusions:

• OptiPrime is a powerful machine learning tool for predicting and optimizing prime editing efficiency.
• The model aids in designing more effective gene editing strategies by accounting for DNA repair mechanisms.
• OptiPrime shows significant promise for advancing therapeutic applications of prime editing, including in vivo correction of genetic diseases.