C-RNNCrispr

C-RNNCrispr predicts CRISPR/Cas9 single guide RNA (sgRNA) on-target activity using a hybrid deep learning framework integrating convolutional neural networks and bidirectional gated recurrent unit networks.

Key Features:

• Hybrid Neural Network Architecture: Combines convolutional neural networks (CNNs) and bidirectional gated recurrent unit networks (BGRUs) to model sgRNA sequence patterns and contextual dependencies.
• Dual-Branch Model Design: Integrates an sgRNA sequence branch and an epigenetic feature branch to incorporate both sequence information and epigenetic context.
• Encoded Input Representation: Uses encoded binary matrices representing sgRNA sequences together with four epigenetic features as model inputs.
• Transfer Learning Strategy: Applies transfer learning by fine-tuning models pre-trained on benchmark datasets using smaller datasets to improve predictive accuracy.
• Quantitative Activity Prediction: Produces regression scores representing predicted on-target activity for candidate sgRNAs.

Scientific Applications:

• CRISPR sgRNA Design: Predicts on-target editing efficiency of sgRNAs for CRISPR/Cas9 genome editing experiments.
• Genome Editing Optimization: Supports selection of highly efficient sgRNAs to improve CRISPR-based gene modification experiments.

Methodology:

C-RNNCrispr trains a hybrid deep learning model combining CNNs and bidirectional GRUs on encoded sgRNA sequences and epigenetic features to predict sgRNA on-target activity scores.