Determinants of Base Editing Outcomes from Target Library Analysis and Machine Learning

Area of Science:

• Genetics and Genomics
• Biotechnology
• Bioinformatics

Background:

• Base editors are crucial for targeted point mutations but their editing determinants remain unclear.
• Understanding sequence-activity relationships is key to optimizing base editing tools.
• Mammalian cell systems provide a robust platform for characterizing gene editing outcomes.

Purpose of the Study:

• To characterize sequence-activity relationships for cytosine and adenine base editors (CBEs and ABEs).
• To develop a predictive machine learning model (BE-Hive) for base editing outcomes and efficiency.
• To engineer novel base editor variants with enhanced capabilities.

Main Methods:

• Genome-wide characterization of 11 base editors across 38,538 integrated targets in mammalian cells.
• Training a machine learning model (BE-Hive) using experimental base editing data.
• In silico prediction and experimental validation of editing outcomes, including bystander edits.

Main Results:

• BE-Hive accurately predicts base editing genotypic outcomes (R ≈ 0.9) and efficiency (R ≈ 0.7).
• Successfully corrected 3,388 disease-associated single nucleotide variants (SNVs) with high precision.
• Identified novel determinants for C-to-G and C-to-A editing, enabling correction of 174 pathogenic transversion SNVs.
• Engineered new CBE variants with modulated editing capabilities.

Conclusions:

• BE-Hive significantly advances the predictability and precision of base editing.
• The study expands the scope of base editing to previously intractable targets, including transversions.
• Novel base editor variants offer improved editing performance and control, paving the way for therapeutic applications.