A fusion framework of deep learning and machine learning for predicting sgRNA cleavage efficiency

Area of Science:

• Genetics and Genomics
• Bioinformatics
• Molecular Biology

Background:

• The CRISPR/Cas9 system is a key genome editing technology.
• Single-guide RNA (sgRNA) design critically impacts editing efficiency.
• Current predictive models for sgRNA cleavage efficiency lack sufficient accuracy.

Purpose of the Study:

• To develop a more accurate method for predicting sgRNA cleavage efficiency.
• To improve the design of highly efficient sgRNAs for CRISPR/Cas9 applications.
• To provide a user-friendly tool for evaluating sgRNA effectiveness.

Main Methods:

• A fusion framework combining deep learning (CNN and RNN) and machine learning (LGBM).
• Utilized primary sequence and secondary structure features of sgRNAs.
• Trained a machine learning model using features extracted by a deep neural network.

Main Results:

• Achieved a Spearman's correlation coefficient of 0.917, a >5% improvement over existing methods.
• Reduced mean square error from 7.89 × 10^-3 to 4.75 × 10^-3.
• Developed an online tool, CRISep, for sgRNA evaluation.

Conclusions:

• The proposed fusion framework significantly enhances the prediction accuracy of sgRNA cleavage efficiency.
• This approach offers a superior method for designing effective sgRNAs compared to previous tools.
• The CRISep tool provides a valuable resource for researchers utilizing CRISPR/Cas9 technology.