Somatic structural variation discovery on cancer pangenome graphs
- 1Bilkent University
Abstract
Somatic structural variants (SVs) drive tumorigenesis across most cancers, yet low variant allele frequencies in subclonal populations often necessitate multi-sample sequencing per patient. Existing somatic SV callers rely on whole-genome read alignments against a single linear reference, incurring high computational costs and reference bias. We present JANUS, which discovers somatic SVs directly within a personal pangenome, a colored de Bruijn-like graph of locally consistent parsing (LCP) cores built from matched normal and tumor PacBio HiFi reads. JANUS identifies candidate SV sites traversed exclusively by tumor reads without prior alignment, mapping only short read snippets surrounding each site. Across five tumor-normal pairs in the CASTLE panel, JANUS leads the 18-caller consensus with a mean F1 score of 0.887 (next-best caller: 0.858), widening its lead under low-signal regimes (0.489 versus 0.362 at 5x; 0.683 versus 0.520 at 20% purity). End to end, JANUS analyzes a tumor-normal pair in ~11 minutes on average with 16 threads, which is ~24x faster than read alignment alone on the same machine. In a 20-tumor patient simulation, joint graph calling outperforms independent pairwise runs in speed and accuracy, particularly at low depth. By processing multi-sample cohorts jointly and maintaining sensitivity at low coverage and purity, JANUS provides an efficient, reference-unbiased framework for cancer genomics. JANUS is available at https://github.com/BilkentCompGen/janus.
Related
- A Plasmodium falciparum Pangenome Resource to Drive Structural Variant Discovery and to assist Malaria Control
- Somatic Structural Variant Discovery from Tumor Specific Strings Computed Against a Draft Composite Germline Reference
- Somatic variant detection using an individual-specific pangenome reference: a case study of a pediatric acute myeloid leukemia patient
- Pangenome Analysis Reveals Gene Content Variation and Evolutionary Dynamics Across the Culex pipiens Species Complex
- Panagram reveals pangenome architectures underlying trait variation in plants and animals
BibTeX
@misc{kara2026somatic,
title = {{Somatic structural variation discovery on cancer pangenome graphs}},
author = {M. Kara and A. Ashyralyyev and C. Alkan},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.764632},
url = {https://www.biorxiv.org/content/10.64898/2026.11.05.764632v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Somatic structural variation discovery on cancer pangenome graphs
AU - M. Kara
AU - A. Ashyralyyev
AU - C. Alkan
PY - 2026
DA - 2026/11/05
PB - bioRxiv
DO - 10.64898/2026.11.05.764632
UR - https://www.biorxiv.org/content/10.64898/2026.11.05.764632v1
AB - Somatic structural variants (SVs) drive tumorigenesis across most cancers, yet low variant allele frequencies in subclonal populations often necessitate multi-sample sequencing per patient. Existing somatic SV callers rely on whole-genome read alignments against a single linear reference, incurring high computational costs and reference bias. We present JANUS, which discovers somatic SVs directly within a personal pangenome, a colored de Bruijn-like graph of locally consistent parsing (LCP) cores built from matched normal and tumor PacBio HiFi reads. JANUS identifies candidate SV sites traversed exclusively by tumor reads without prior alignment, mapping only short read snippets surrounding each site. Across five tumor-normal pairs in the CASTLE panel, JANUS leads the 18-caller consensus with a mean F1 score of 0.887 (next-best caller: 0.858), widening its lead under low-signal regimes (0.489 versus 0.362 at 5x; 0.683 versus 0.520 at 20% purity). End to end, JANUS analyzes a tumor-normal pair in ~11 minutes on average with 16 threads, which is ~24x faster than read alignment alone on the same machine. In a 20-tumor patient simulation, joint graph calling outperforms independent pairwise runs in speed and accuracy, particularly at low depth. By processing multi-sample cohorts jointly and maintaining sensitivity at low coverage and purity, JANUS provides an efficient, reference-unbiased framework for cancer genomics. JANUS is available at https://github.com/BilkentCompGen/janus.
ER -