CROTON

CROTON predicts CRISPR/Cas9 editing outcomes using variant-aware deep multi-task convolutional neural networks and neural architecture search to model and interpret sequence determinants and SNV effects.

Key Features:

• Automated feature and model engineering: Uses neural architecture search (NAS) to optimize convolutional neural network architecture and automate feature extraction.
• Deep multi-task convolutional neural networks: Employs deep multi-task CNNs to model multiple CRISPR/Cas9 editing outcome types simultaneously.
• Predictive outputs: Predicts probabilities of 1 base pair insertions and deletions, frequencies of deletions, and frameshift mutation rates.
• Interpretability: Provides insights into local sequence determinants that influence diverse editing outcomes.
• Variant awareness: Assesses the impact of single nucleotide variants (SNVs) on genome editing efficacy and evaluates guide RNA applicability across genetic backgrounds, including ACE2, CCR5, CTLA4, and PDCD1.
• Performance advantage: Demonstrates superior performance compared to existing models and non-NAS CNNs.

Scientific Applications:

• Biology and Biotechnology: Predicts CRISPR/Cas9 editing outcomes to inform guide RNA design and experimental planning.
• Medicine: Evaluates how SNVs affect genome editing outcomes to inform personalized therapeutic strategies targeting genes such as ACE2, CCR5, CTLA4, and PDCD1.

Methodology:

Trained deep multi-task CNNs on a synthetic large-scale construct-based dataset; model architecture optimized using neural architecture search (NAS); tested on an independent primary T cell genomic editing dataset.