PrimeNet: rational design of Prime editing pegRNAs by deep learning.

Area of Science:

• Molecular Biology
• Genomics
• Biotechnology

Background:

• Prime editing is a precise gene editing tool with potential in research and therapeutics.
• Suboptimal efficiency for large fragments and lack of accurate predictive models limit prime editing adoption.
• Existing predictive models often neglect crucial epigenetic factors influencing gene editing outcomes.

Purpose of the Study:

• To develop a novel prediction model, PrimeNet, for prime editing.
• To enhance the accuracy and generalization performance of prime editing predictions.
• To integrate significant epigenetic factors into a predictive framework for gene editing.

Main Methods:

• Developed PrimeNet, a prediction model incorporating chromatin accessibility and DNA methylation data.
• Utilized multiscale convolution and attention mechanisms within the model architecture.
• Trained and validated the model on datasets from HEK293T and K562 cell lines.

Main Results:

• PrimeNet achieved high prediction accuracy, with Spearman correlation coefficients of 0.94 and 0.82 on two distinct cell line datasets.
• The model demonstrated superior performance compared to existing gene editing prediction models.
• Integration of epigenetic factors significantly improved predictive capabilities.

Conclusions:

• PrimeNet offers a more accurate and reliable method for predicting prime editing efficiency.
• The model can guide experimental design, improving success rates and reducing costs in gene editing applications.
• This advancement holds potential for accelerating the application of gene editing in treating genetic diseases.