Leveraging uncertainty quantification to optimize CRISPR guide RNA selection.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• CRISPR-Cas9 genome editing utilizes guide RNA (gRNA) to target specific DNA sequences.
• Predicting gRNA efficiency is crucial for successful gene editing outcomes.
• Machine learning models are increasingly used for gRNA efficiency prediction due to growing experimental data.

Purpose of the Study:

• To investigate the utility of uncertainty quantification in predicting CRISPR gRNA efficiency.
• To develop and evaluate novel guide selection strategies incorporating prediction uncertainty.
• To assess the performance of deep ensemble models compared to single models for gRNA efficiency prediction.

Main Methods:

• Employed a deep ensemble approach for gRNA efficiency prediction.
• Integrated uncertainty quantification into the prediction framework.
• Developed and tested new guide selection strategies based on predicted efficiency and uncertainty.
• Evaluated performance using precision and gene coverage metrics.

Main Results:

• Deep ensemble models outperformed single models in gRNA efficiency prediction.
• The proposed method provides reliable uncertainty quantification for predictions.
• Novel strategies incorporating uncertainty achieved over 90% precision in guide selection.
• Suitable guides were identified for more than 93% of genes in the mouse genome.

Conclusions:

• Uncertainty quantification is a valuable tool for enhancing CRISPR gRNA selection strategies.
• Deep ensemble methods offer improved accuracy and uncertainty estimates for gRNA efficiency prediction.
• The developed strategies represent a significant advancement in designing more effective genome editing experiments.