cnnLSV

cnnLSV detects and refines structural variant calls from long-read sequencing data by applying convolutional neural networks to filter false-positive predictions.

Key Features:

• Alignment-to-Image Encoding: Converts long-read alignment information surrounding structural variant regions into image-like representations for neural network analysis.
• CNN-Based Variant Filtering: Uses a convolutional neural network model to distinguish true structural variants from false-positive calls.
• Merged Callset Refinement: Processes combined structural variant callsets generated by existing detection tools to improve precision.
• Sample Label Correction: Applies principal component analysis (PCA) and k-means clustering during training to identify and remove mislabeled samples.

Scientific Applications:

• Structural Variant Detection: Identifies genomic structural variants including insertions, deletions, inversions, and duplications from long-read sequencing data.
• Genomic Variation Analysis: Supports studies of genetic diversity, disease-associated variants, and genome structural variation.

Methodology:

cnnLSV encodes long-read alignment signals around structural variant regions into image representations, trains a convolutional neural network to classify true variants, and applies the trained model to merged callsets while correcting mislabeled training samples using principal component analysis and k-means clustering.