TransCrispr: Transformer Based Hybrid Model for Predicting CRISPR/Cas9 Single Guide RNA Cleavage Efficiency

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• CRISPR/Cas9 is a powerful genome editing technology for targeted DNA modification.
• Accurate prediction of single guide RNA (sgRNA) on- and off-target effects remains a challenge.
• Computational methods are crucial for optimizing sgRNA design with high cell-specific sensitivity and specificity.

Purpose of the Study:

• To develop a novel computational approach for predicting sgRNA knockout efficacy.
• To improve the accuracy and reliability of sgRNA design for CRISPR/Cas9 applications.
• To investigate the influence of biological features on sgRNA on-target activity.

Main Methods:

• Introduced TransCrispr, a hybrid architecture combining Transformer and Convolutional Neural Network (CNN).
• Encoded sgRNA sequence data, positional information, and biological features as network input.
• Utilized CNN for feature representation learning and Transformer for self-attention mechanisms.

Main Results:

• TransCrispr demonstrated superior prediction accuracy compared to existing state-of-the-art methods.
• The model showed enhanced generalization ability across diverse datasets.
• Experimental validation on seven public datasets confirmed the model's performance.

Conclusions:

• TransCrispr offers a significant advancement in predicting sgRNA knockout efficiency.
• The integration of sequence and biological features improves prediction performance.
• This tool facilitates more precise and effective sgRNA design for genome editing.