DeepSV

DeepSV applies a deep convolutional neural network to detect long deletions (structural variations) from high-throughput sequencing reads.

Key Features:

• Novel Visualization Method: Transforms sequence reads into image-like representations that encode multiple sources of information relevant to long deletions.
• Handling Noisy Data: Incorporates techniques to manage and mitigate noise in training data during model training to improve robustness of deletion calls.
• Deep Learning Model Training: Trains a deep convolutional neural network on the visualized sequence-read images specifically tailored to identify long deletions.
• Performance Superiority: Validated against existing methods using data from the 1000 Genomes Project, demonstrating superior accuracy and efficiency in calling deletions.

Scientific Applications:

• Structural variation detection: Accurate calling of long deletions to characterize structural variation in genomes.
• Genomic architecture and diversity: Enabling analyses of genomic architecture and genetic diversity from high-throughput sequencing data.
• Disease and phenotype studies: Supporting investigation of the molecular underpinnings of phenotypic traits and disease associated with long deletions.

Methodology:

Sequence reads are transformed into image-like visualizations; a deep convolutional neural network is trained on these visualizations with noise-mitigation techniques; the trained model is used to call long deletions from new sequencing datasets.