DeepSpCas9

DeepSpCas9 predicts SpCas9 activity by using deep learning to estimate sgRNA-mediated indel frequencies for Streptococcus pyogenes Cas9 (SpCas9) target sites.

Key Features:

• Deep learning model: A predictive model trained using deep learning to infer SpCas9 activity from sequence-derived data.
• Training data scale: Trained on a high-throughput human cell library containing paired sgRNA-encoding constructs and target sequences covering 12,832 target sites.
• Input data types: Uses sgRNA sequences, corresponding target DNA sequences, and SpCas9-induced indel (insertion-deletion) frequency measurements as input.
• Prediction output: Produces estimates of sgRNA efficiency expressed as expected indel frequencies and likelihood of successful gene editing events.
• Generalization validation: Evaluated against independently generated datasets, including those from other research groups and proprietary datasets, demonstrating consistent high accuracy.
• High-throughput approach: Developed from comprehensive high-throughput measurements of SpCas9 activity across many target sequences.

Scientific Applications:

• CRISPR experiment design: Guides selection and optimization of sgRNAs to improve SpCas9 editing efficiency in experimental workflows.
• Functional genomics: Supports identification of effective sgRNAs for gene perturbation studies and loss-of-function screens.
• Therapeutic development: Assists preclinical sgRNA selection for genome-editing strategies relevant to therapeutic applications.
• Synthetic biology: Enables design of precise genome edits by predicting efficient SpCas9 target sites for engineered constructs.

Methodology:

The model was trained using deep learning on datasets of SpCas9-induced indel frequencies derived from a high-throughput human cell library of paired sgRNA-encoding constructs and target sequences (12,832 targets) and was evaluated on independently generated datasets from other groups and proprietary sources.