BreakNet

BreakNet detects deletions in genomic sequences from long-read alignment data using deep learning to improve structural variant calling accuracy.

Key Features:

• Feature Extraction: Extracts feature matrices from aligned long-read sequencing data that capture signals for potential deletion regions.
• Time-distributed CNN processing: Uses a time-distributed convolutional neural network to transform feature matrices into continuous feature vectors and capture spatial hierarchies.
• Bidirectional LSTM analysis: Applies a bidirectional long short-term memory model to process continuous feature vectors in both forward and backward directions to capture temporal dependencies and contextual information.
• Classification module: Classifies genomic regions as deletions or non-deletions based on the outputs of the preceding models.
• Performance benchmarking: Achieves higher F1 scores on real long-read sequencing datasets compared to Sniffles, SVIM, and cuteSV.

Scientific Applications:

• Deletion detection: Accurate calling of deletion structural variants from long-read sequencing data.
• Structural variant analysis: Identification and characterization of deletions within broader structural variation studies.
• Disease-associated deletion discovery: Detection of deletions implicated in genetic diseases.
• Benchmarking SV callers: Comparative evaluation of structural variant callers using F1 score metrics.

Methodology:

BreakNet extracts feature matrices from aligned long-read sequencing data, processes them with a time-distributed convolutional neural network into continuous feature vectors, analyzes these vectors with a bidirectional long short-term memory model, and classifies regions as deletions or non-deletions.