Prediction of base editor off-targets by deep learning.

Area of Science:

• Genetics and Genomics
• Molecular Biology
• Bioinformatics

Background:

• Base editors are powerful tools for precise genome editing.
• Off-target mutations, caused by guide RNA (gRNA) sequence mismatches, limit the safety and efficacy of base editing.
• Predicting and minimizing these off-target effects is crucial for clinical applications.

Purpose of the Study:

• To develop predictive models for Cas9-dependent off-target mutations induced by adenine base editors (ABEs) and cytosine base editors (CBEs).
• To create a user-friendly web server for predicting base editing off-target sites.

Main Methods:

• Designed and integrated gRNA-off-target pairs for ABEs and CBEs into human cells.
• Generated large-scale datasets of off-target mutation efficiencies (54,663 for ABEs, 55,727 for CBEs).
• Trained deep learning models (ABEdeepoff and CBEdeepoff) using the generated datasets.

Main Results:

• Developed ABEdeepoff and CBEdeepoff models capable of predicting off-target sites with high accuracy.
• Achieved high prediction performance for endogenous loci, with Spearman correlation values ranging from 0.710 to 0.859.
• Integrated the predictive models into an accessible online web server (http://www.deephf.com/#/bedeep/bedeepoff).

Conclusions:

• The developed deep learning models and web server effectively predict base editor off-target mutations.
• These tools can significantly aid researchers in minimizing off-target effects, enhancing the safety of base editing technologies.
• Facilitates the broader application of base editing in research and potentially therapeutic settings.