An interpretable deep learning framework uncovers features governing CRISPR-Cas9 genome-editing efficiency.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• CRISPR-Cas9 genome editing efficiency depends on single-guide RNA (sgRNA) sequence and position.
• Existing deep learning models for predicting Cas9 efficiency are often black boxes, lacking interpretability.
• Previous research has not fully explored the impact of positional context on sgRNA sequence motifs influencing Cas9 activity.

Purpose of the Study:

• To develop an interpretable machine learning framework, DeepCC9, for predicting CRISPR-Cas9 genome-editing efficiency.
• To identify sequence composition and positional motifs within sgRNAs that govern Cas9 activity.
• To provide mechanistic insights into the sequence determinants of guide RNA performance.

Main Methods:

• Developed DeepCC9, an interpretable machine learning framework combining explicit sequence feature extraction with a deep architecture.
• Utilized a residual block-based deep architecture for enhanced interpretability.
• Applied the framework to multiple Cas9 variant datasets for analysis.

Main Results:

• DeepCC9 achieved superior predictive performance compared to existing methods for Cas9 efficiency.
• The framework enabled direct interpretation of sequence motifs and their positional effects on sgRNA function.
• Identified 74 sequence motifs significantly associated with Cas9 efficiency at specific positions within sgRNAs.

Conclusions:

• DeepCC9 offers a generalizable and interpretable approach to modeling sequence-function relationships in genome editing.
• The framework advances the understanding of sequence determinants underlying CRISPR-Cas9 genome editing efficiency.
• Provides mechanistic insights into how sgRNA sequence and positional context influence Cas9 binding and cleavage.