PangenomeX: a graph convolutional network-based pangenome framework for unbiased population-scale genomic variation...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Pangenomes are crucial for identifying population-specific genomic variations from shallow whole genome sequencing.
• Current pangenome frameworks struggle with copy number variation (CNV) analysis due to population representation bias and difficulty distinguishing benign copy number polymorphisms (CNPs) from pathogenic CNVs.

Purpose of the Study:

• To develop a novel pangenome framework, PangenomeX, for accurate population-scale CNV analysis using low-coverage sequencing data.
• To address the challenges of population representation bias and the accurate identification of pathogenic CNVs amidst common CNPs.

Main Methods:

• PangenomeX utilizes a graph-convolutional network (GCN) framework.
• It embeds known CNPs as prior knowledge and constructs a CNV relationship network guided by a phylogenetic tree.
• A GCN learns interactions between CNV and CNP nodes, aggregating information from local neighborhoods to mitigate population bias.

Main Results:

• PangenomeX demonstrates superior performance in distinguishing pathogenic CNVs from common population CNPs compared to existing methods.
• Evaluations on simulated data and 561 real samples validate its effectiveness.

Conclusions:

• PangenomeX provides a robust methodological blueprint for large-cohort variant screening, particularly for CNVs.
• It offers a practical approach for integrating graph-based genomics into clinical practice for improved variant analysis.