Deep learning models simultaneously trained on multiple datasets improve base-editing activity prediction.

Area of Science:

• Molecular Biology
• Genetics
• Bioinformatics

Background:

• CRISPR-derived base editors (BE) offer precise single nucleotide substitutions without DNA double-strand breaks.
• Base editing efficiency is influenced by guide RNA (gRNA) efficacy and the target DNA site.

Purpose of the Study:

• To enhance the accuracy of base editor guide RNA (gRNA) design.
• To develop deep learning models for predicting gRNA efficiency in base editing.

Main Methods:

• Generated approximately 20,000 gRNAs for A•T to G•C and C•G to T•A conversions.
• Trained deep neural networks on multiple datasets to predict gRNA efficiency.
• Developed dataset-aware prediction capabilities for base editing models.

Main Results:

• Demonstrated significant improvement in BE gRNA design accuracy using deep learning.
• The developed models enable dataset-aware predictions for base editing applications.
• Created models that can predict gRNA efficiency for specific base editing tasks.

Conclusions:

• Deep learning models trained on extensive datasets substantially improve CRISPR base editor gRNA design.
• The developed computational tools offer enhanced accuracy and flexibility for base editing experiments.
• These advancements facilitate more efficient and precise genome editing strategies.