Pansoma, a machine learning tool for identifying somatic variants using pangenome graphs

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Somatic variant calling identifies acquired mutations crucial for cancer research and precision oncology.
• Linear reference genomes introduce bias, limiting accuracy in somatic variant detection.
• Graph-based pangenome references offer improved representation of human genetic diversity.

Purpose of the Study:

• To introduce Pansoma, a novel pangenome-native tool for somatic variant calling.
• To leverage machine learning and graph pangenome references for enhanced variant detection.
• To provide bioinformatics tools for graph-based genomic data management and analysis.

Main Methods:

• Developed Pansoma, a machine learning-based tool utilizing tensor representations of alignment on graph nodes.
• Applied Pansoma to both short- and long-read sequencing data for somatic variant detection.
• Created accompanying bioinformatics tools for graph-based data and variant analysis.

Main Results:

• Pansoma improves tumor-only somatic variant detection accuracy.
• The tool preserves unique variant representations from the pangenome graph.
• Outputs include graph-anchored variants and remapped linear-reference variants.

Conclusions:

• Pansoma offers a significant advancement in somatic variant calling using pangenome graphs.
• The tool addresses limitations of linear references, improving accuracy and data representation.
• Pansoma facilitates more robust genomic studies in cancer and precision medicine.