Accurate Prediction of CRISPR/Cas13a Guide Activity Using Feature Selection and Deep Learning

Area of Science:

• Molecular Biology
• Bioinformatics
• Genetics

Background:

• CRISPR/Cas13a is crucial for nucleic acid detection.
• Accurate prediction of CRISPR/Cas13a activity is vital for sensitive diagnostics.

Purpose of the Study:

• To develop a novel dual-branch neural network model for predicting CRISPR/Cas13a guide activity.
• To enhance the accuracy and reliability of CRISPR/Cas13a-based diagnostic tools.

Main Methods:

• Developed a dual-branch neural network integrating sequence encoding and statistical descriptive features.
• Utilized Shapley Additive Explanations and Integrated Gradients for feature importance analysis.
• Validated the model on two independent CRISPR/Cas13a datasets.

Main Results:

• Achieved high prediction accuracy and classification performance, outperforming existing models.
• Identified 99 key descriptive features influencing guide-target interactions and Cas13a activity.
• Highlighted sequence composition, mismatch characteristics, and protospacer flanking sites as primary predictive features.

Conclusions:

• The novel model provides a reliable and efficient method for predicting CRISPR/Cas13a guide activity.
• Integrating descriptive features alongside sequence information improves deep learning model performance.
• Findings offer insights into guide-target interactions, supporting rational design of CRISPR diagnostics.