igRNA Prediction and Selection AI Models (igRNA-PS) for Bystander-less ABE Base Editing

Area of Science:

• Molecular Biology
• Biotechnology
• Genetics Engineering

Background:

• CRISPR base editing often results in multiple edits, hindering precise correction of single nucleotide variations (SNVs).
• Existing methods struggle with bystander edits, complicating the reversion of disease-associated SNVs.

Purpose of the Study:

• To develop a novel imperfect guide RNA (igRNA) strategy for bystander-less single-base editing.
• To create AI models for predicting igRNA performance and selecting optimal sequences for precise base editing.

Main Methods:

• Designed and tested approximately 5000 loci with systematically designed ABE igRNAs using a high-throughput method.
• Developed deep learning-based AI models (igRNA-PS) to predict editing efficiency, product purity, and identify major editing sites.
• Generated and evaluated sets of 64 igRNAs per locus to identify the best performer.

Main Results:

• The igRNA-PS models accurately identify major editing sites from bystanders with near 90% accuracy.
• A modified single-base editing efficiency (SBE) metric was developed to predict both editing efficiency and product purity.
• The system provides optimized igRNAs for specific editing loci, facilitating bystander-less base editing.

Conclusions:

• The igRNA editing strategy effectively achieves bystander-less single-base editing, overcoming a key limitation of base editors.
• The developed AI models offer a convenient and efficient approach for selecting high-performing igRNAs.
• This work advances precise gene editing for correcting disease-associated SNVs.