CRISPR-DIPOFF: an interpretable deep learning approach for CRISPR Cas-9 off-target prediction

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Biotechnology

Background:

• CRISPR-Cas9 technology enables precise DNA editing but poses risks due to potential off-target modifications.
• Accurate prediction of off-target effects is crucial for safe and effective application of CRISPR-Cas9.
• Existing deep learning models for off-target prediction face challenges with the precision-recall trade-off and lack interpretability.

Purpose of the Study:

• To develop a highly accurate and interpretable deep learning model for predicting CRISPR-Cas9 off-target effects.
• To address the precision-recall trade-off limitations in current off-target prediction methods.
• To provide insights into the biological factors influencing off-target modifications.

Main Methods:

• Exploration of recurrent neural network (RNN) based deep learning models for sequence data analysis.
• Application of genetic algorithms for hyperparameter tuning to optimize model performance.
• Utilization of the integrated gradient method for model interpretability.

Main Results:

• Demonstrated significant performance improvements in off-target prediction compared to state-of-the-art methods.
• Achieved a desirable balance between precision and recall, overcoming previous limitations.
• Identified specific sub-regions within the single guide RNA seed region as key contributors to off-target effects.

Conclusions:

• The developed deep learning model offers enhanced efficacy and interpretability for CRISPR-Cas9 off-target prediction.
• The findings provide a deeper understanding of the mechanisms underlying off-target modifications.
• This work represents a significant step towards safer and more reliable genome-editing applications.