Overview of the Swave method.

Professor Ye Kai's team at Xi'an Jiaotong University has published a research paper titled Population-level structural variant characterization using pangenome graphs in Nature Genetics. The study introduces Swave, a new method for resolving population-level structural variants (SVs) using pangenome graphs.

This work addresses the challenge of population structural variant detection and genotyping, providing a new tool for characterizing complex structural variant maps. It marks another innovative achievement by Professor Ye Kai's team at the intersection of artificial intelligence and biomedicine.

Genomic variation is a vital driver of core biological processes, such as disease progression and species evolution. Although rapid developments in genome assembly and pangenome technology have made population-scale SV analysis possible, several long-standing bottlenecks remain, such as difficulty identifying complex variants, strong noise interference in repetitive regions, and high rates of missing genotypes in cross-sample analysis.

Accurately identifying and analyzing complex SVs in large-scale population data has become a crucial challenge for both genomics research and clinical applications.

To address these challenges, the Swave method employs a dot-plot projection wave strategy to perform dimensionality elevation, denoising, and feature extraction on sequences. It then utilizes an AI-based recurrent neural network (RNN) to automatically determine variant types and perform sample genotyping.

Swave outperforms existing methods in key metrics, including SV identification accuracy, pedigree consistency, and population genotype completeness.

The team further applied this method to pangenome analyses of healthy populations and rare disease cohorts. They discovered a significant number of rare, low-frequency events within complex structural variants that were previously underestimated, highlighting their importance in studying pathogenic and evolutionary mechanisms.

This research underscores the unique advantages of AI in recognizing complex patterns within biological information and demonstrates how innovative computational methods support biological and medical research.

Notably, this achievement, alongside the team's previous work on individual SV detection (SVision, Nature Methods, 2022), cross-sample SV comparison (SVision-pro, Nature Biotechnology, 2024), and recently published high-precision gene annotation (ANNEVO), forms a comprehensive strategic framework for AI-driven genomic analysis. It reflects the team's sustained ability to bridge the gaps between AI methodologies, core genomic questions, and biomedical applications.