CRISPRLand

CRISPRLand predicts DNA repair outcomes following Cas9-induced double-strand breaks by exploiting sparsity in the Walsh-Hadamard spectral domain to enable interpretable and scalable modeling.

Key Features:

• Spectral Framework: Transforms the DNA repair landscape into the Walsh-Hadamard spectral domain and leverages its sparsity for analysis.
• Divide-and-Conquer Strategy: Applies a divide-and-conquer approach using a fast peeling algorithm to learn DNA repair models efficiently.
• Efficient Computation: Reduces computation time from an impractical 5,230 years to approximately one week and decreases sampling complexity from 10^12 to ~3 million guide RNAs while maintaining an R² of ~0.9.
• Feature Capture: Captures lower-degree features near the cut site associated with short insertions and deletions and higher-degree microhomology patterns correlated with longer deletions.

Scientific Applications:

• Genome editing outcome prediction: Predicts repair outcomes after Cas9 cuts to inform interpretation of genome editing experiments.
• Model training and scalable inference: Enables reliable training and scalable inference of DNA repair models with interpretable components.
• Gene-editing strategy and genomic stability studies: Informs design of precise gene-editing strategies and investigation of mutation processes and genomic stability.

Methodology:

Transforms repair outcomes into the Walsh-Hadamard spectral domain and applies a divide-and-conquer strategy with a fast peeling algorithm that exploits sparsity of the repair landscape.