Optimized CRISPR guide RNA design for two high-fidelity Cas9 variants by deep learning

Area of Science:

• * Molecular Biology
• * Genomics
• * Bioinformatics

Background:

• * Highly specific Cas9 nucleases are crucial for precise genome editing.
• * Understanding guide RNA (gRNA) activity is essential for maximizing their application.
• * Current knowledge gaps limit the broad utility of SpCas9 variants.

Purpose of the Study:

• * To comprehensively measure gRNA activity for SpCas9 variants (eSpCas9(1.1), SpCas9-HF1) and wild-type SpCas9 (WT-SpCas9).
• * To identify key features influencing gRNA activity and develop predictive models.
• * To create an accessible online tool for gRNA design.

Main Methods:

• * Genome-scale screening of over 50,000 gRNAs across ~20,000 genes in human cells.
• * Evaluation of 1,031 biological and sequence-based features impacting gRNA activity.
• * Development and benchmarking of predictive models, including Recurrent Neural Networks (RNNs).

Main Results:

• * Indel rates were quantified for thousands of gRNAs with three Cas9 nucleases.
• * A combined RNN and feature-based model demonstrated superior gRNA activity prediction accuracy.
• * The developed model outperformed popular existing gRNA design tools.

Conclusions:

• * Accurate prediction of gRNA activity is achievable using integrated biological features and advanced modeling.
• * The DeepHF tool provides a valuable resource for designing effective gRNAs for specific Cas9 nucleases.
• * This work enhances the precision and applicability of genome editing technologies.