De novo and somatic structural variant discovery with SVision-pro

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate discovery of structural variants (SVs) from long-read sequencing data is crucial for genomic research.
• Current methods for de novo and somatic SV detection face challenges, especially with complex genomic rearrangements.
• Comparative genomic analysis is often required but can be computationally intensive.

Purpose of the Study:

• To develop a novel framework for efficient and accurate structural variant discovery using long-read sequencing.
• To enable visual representation of genome-to-genome differences for intuitive SV identification.
• To improve the detection of complex SVs and low-frequency variants.

Main Methods:

• Development of SVision-pro, a neural-network-based instance segmentation framework.
• Application of the framework for comparative genomic analysis between samples.
• Evaluation of SVision-pro against state-of-the-art SV detection approaches.

Main Results:

• SVision-pro visually represents genome-to-genome sequencing differences.
• The framework enables comparative SV discovery without needing pre-trained inference models.
• SVision-pro demonstrates superior performance, particularly in resolving complex SVs.
• Achieved low Mendelian error rates and high sensitivity for low-frequency SVs.
• Reduced false-positive rates compared to traditional SV merging techniques.

Conclusions:

• SVision-pro offers a powerful and accurate approach for long-read-based structural variant discovery.
• The visual and comparative nature of SVision-pro simplifies the identification of complex genomic alterations.
• This tool advances the field of comparative genomics and SV detection methodologies.