CnnCrispr

CnnCrispr predicts off-target propensity of CRISPR/Cas9 single-guide RNAs (sgRNAs) using deep learning models that analyze sgRNA–DNA sequence interactions.

Key Features:

• Hybrid Deep Learning Architecture: Combines convolutional neural networks (CNNs) and bidirectional Long Short-Term Memory (biLSTM) networks across multiple hidden layers to model sequence dependencies and spatial patterns in sgRNA–DNA interactions.
• Automated Sequence Feature Representation: Uses the Global Vectors for Word Representation (GloVe) model to generate vector representations of sgRNA–DNA sequence pairs for automatic feature extraction.
• Off-Target Activity Prediction: Performs both classification and regression analyses to estimate the likelihood and magnitude of sgRNA off-target effects.

Scientific Applications:

• CRISPR/Cas9 Genome Editing Design: Evaluates potential off-target cleavage sites to support the selection of sgRNAs with improved specificity.
• Gene Editing Safety Assessment: Assesses off-target risks associated with CRISPR/Cas9 experiments in functional genomics and therapeutic genome editing studies.

Methodology:

CnnCrispr converts sgRNA–DNA sequences into vector representations using the GloVe model and applies a deep learning architecture combining CNN and bidirectional LSTM layers to learn patterns associated with off-target activity.