DTMP-prime: A deep transformer-based model for predicting prime editing efficiency and PegRNA activity.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Prime editing (PE) is a powerful CRISPR-based genome engineering technology for correcting mutations.
• Optimizing prime editing guide RNA (PegRNA) is crucial for achieving high editing efficiency.
• Accurate prediction of PE efficiency and off-target effects is essential for clinical applications.

Purpose of the Study:

• To develop a deep transformer-based model, DTMP-Prime, for predicting prime editing efficiency.
• To facilitate the rational design of PegRNAs and ngRNAs for enhanced prime editing outcomes.
• To improve the prediction accuracy of off-target sites in CRISPR-based genome editing experiments.

Main Methods:

• A transformer-based deep learning model was constructed using extensive prime editing data.
• Features of PegRNAs and target DNA sequences were extracted and encoded.
• DNABERT-based embedding and a multi-head attention framework were integrated to enhance predictive capabilities.
• Model performance was evaluated using Pearson and Spearman correlation coefficients.

Main Results:

• DTMP-Prime demonstrated high accuracy in predicting prime editing efficiency and outcomes.
• The model showed improved predictive capabilities for off-target sites compared to existing methods.
• DTMP-Prime exhibited strong generalizability across different PE models and cell lines.
• Evaluation confirmed DTMP-Prime outperforms state-of-the-art models in predicting PE efficiency.

Conclusions:

• DTMP-Prime is an effective tool for predicting prime editing efficiency and PegRNA activity.
• The developed model aids in the design of precise and efficient prime editing strategies.
• DTMP-Prime offers a promising approach for minimizing off-target mutations in genome engineering.