Deep learning improves the ability of sgRNA off-target propensity prediction

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• CRISPR/Cas9 is a powerful genome editing tool for gene repair and regulation.
• Off-target mutations are a significant concern, impacting CRISPR/Cas9 specificity and safety.
• Existing prediction methods struggle to meet clinical accuracy demands due to data expansion and deep learning advancements.

Purpose of the Study:

• To develop a highly accurate prediction method for CRISPR/Cas9 off-target effects.
• To improve the prediction of sgRNA off-target propensity in specific DNA fragments.
• To provide a tool that surpasses current state-of-the-art models in classification and regression performance.

Main Methods:

• Proposed CnnCrispr, a deep learning model for predicting sgRNA off-target propensity.
• Utilized GloVe model for automatic sequence feature training of sgRNA-DNA pairs.
• Integrated biLSTM and CNN with five hidden layers to analyze sequence features.

Main Results:

• CnnCrispr demonstrated high performance in predicting off-target effects.
• Achieved an area under the receiver operating characteristic curve (auROC) of 0.957 and area under the precision-recall curve (auPRC) of 0.429 in cross-validation.
• Exhibited strong classification and regression capabilities, outperforming existing models.

Conclusions:

• CnnCrispr offers superior classification and regression performance compared to current state-of-the-art off-target prediction models.
• The developed method enhances the reliability and safety of CRISPR/Cas9 applications.
• The CnnCrispr code is publicly available for research use.