GKNnet: an relational graph convolutional network-based method with knowledge-augmented activation layer for...

Area of Science:

• Genomics and Bioinformatics
• Microbial Evolution and Adaptation
• Computational Biology

Background:

• Structural variants (SVs), especially deletions, significantly impact microbial phenotypes, adaptation, and evolution.
• Accurate identification of deletion variations is crucial for understanding microbial genomics.
• Long-read sequencing offers improved SV detection but suffers from high error rates, complicating existing algorithms.

Purpose of the Study:

• To develop an advanced method for precise and comprehensive identification of deletion structural variants in microbial genomes.
• To address the challenges posed by high error rates in long-read sequencing data for SV detection.
• To improve the accuracy and recall of structural variant detection algorithms.

Main Methods:

• A novel method employing graph convolutional networks (GCNs) to identify variant regions.
• Integration of a knowledge-augmented activation layer (KANLayer) to minimize noise and reduce false positives.
• Utilization of a clustering algorithm to consolidate overlapping variant regions, enhancing recall.

Main Results:

• The proposed GCN-based method demonstrates superior performance in identifying deletion structural variants.
• Achieved higher F1 scores compared to established benchmark methods (cuteSV, Sniffles, Svim, Pbsv) on simulated and real datasets.
• The method shows robustness and improved precision and recall in structural variant detection.

Conclusions:

• The GCN-based approach offers an innovative and effective solution for microbial genome structural variation research.
• This method significantly enhances the accuracy and reliability of deletion variant detection from long-read sequencing data.
• The findings contribute to advancing the study of microbial evolution and adaptation through improved genomic analysis.