SVHunter: long-read-based structural variation detection through the transformer model.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Structural variations (SVs) are genomic rearrangements (>50 bp) prevalent in the human genome and linked to complex diseases.
• Current long-read SV detection tools use heuristic algorithms, limiting flexibility, accuracy, and robustness.
• Oversimplification of SV signatures by existing methods hinders comprehensive analysis.

Purpose of the Study:

• To introduce SVHunter, a transformer-based approach for enhanced long-read structural variation detection.
• To improve the accuracy and robustness of identifying complex genomic rearrangements.
• To provide a more flexible and powerful tool for SV analysis in research and clinical settings.

Main Methods:

• SVHunter integrates convolutional neural networks (CNNs) and transformers to capture local and global SV signatures.
• Utilizes mean shift clustering with dynamic bandwidth adjustment for precise breakpoint clustering.
• Employs a transformer-based architecture for advanced pattern recognition in genomic data.

Main Results:

• SVHunter demonstrates superior performance in detecting diverse SV types across multiple sequencing platforms and datasets.
• Achieved a significant reduction in the false discovery rate compared to existing methods.
• Accurate identification of SVs, including complex rearrangements, was validated.

Conclusions:

• SVHunter offers a robust and accurate solution for long-read SV detection.
• The method shows significant potential for advancing genomic research and clinical diagnostics.
• Transformer-based approaches represent a promising direction for complex genomic variation analysis.