Deletion variants calling in third-generation sequencing data based on a dual-attention mechanism

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic structural variations, particularly deletions, are significant contributors to genetic diseases.
• Third-generation sequencing technologies offer enhanced capabilities for analyzing complex genomic structures and understanding variant impacts.
• Accurate detection of deletion variants is crucial for disease research and genetic diagnostics.

Purpose of the Study:

• To introduce Dual Attention Structural Variation (DASV), a novel computational method for precise deletion structural variant calling.
• To leverage deep learning and attention mechanisms for improved analysis of genomic sequencing data.
• To evaluate DASV's performance against existing state-of-the-art tools.

Main Methods:

• DASV converts gene alignment information into image representations.
• A dual attention mechanism integrates image data with genomic sequencing data.
• A multi-scale convolutional neural network is employed for precise identification of deletion regions.
• Performance was benchmarked against cuteSV, SVIM, Sniffles, and PBSV across multiple datasets.

Main Results:

• DASV demonstrates superior performance in deletion variant calling compared to established tools.
• The method achieves a favorable balance between precision and recall, leading to enhanced F1 scores.
• Consistent improvements were observed across diverse genomic datasets, highlighting DASV's robustness.

Conclusions:

• DASV represents a significant advancement in the accurate detection of deletion structural variants.
• The proposed image-based deep learning approach offers a powerful new perspective for genomic variation analysis.
• This method has the potential to improve the understanding and diagnosis of genetic diseases associated with deletions.