Somatic structural variation discovery on cancer pangenome graphs

Somatic structural variations (SVs) play a significant role in the development of most cancers, but identifying these alterations can be challenging due to their low frequency in subclonal populations. Traditional somatic SV callers depend on whole-genome read alignments against a single reference genome, resulting in extensive computational demands and potential reference bias. To address these issues, researchers have developed JANUS, a novel method for discovering somatic SVs within a personal pangenome.

JANUS constructs a colored de Bruijn-like graph of locally consistent parsing (LCP) cores from matched normal and tumor PacBio HiFi reads. This graph allows JANUS to identify candidate SV sites traversed exclusively by tumor reads without the need for prior alignment. Instead, the method maps short read snippets surrounding each SV site.

The approach has been tested across five tumor-normal pairs in the CASTLE panel, where JANUS outperformed the 18 other callers in the consensus analysis, achieving a mean F1 score of 0.887. JANUS also demonstrates superior performance in low-signal regimes, where its scores of 0.489 at 5x coverage and 0.683 at 20% tumor purity significantly surpass the next-best caller's scores of 0.362 and 0.520, respectively.

The efficiency of JANUS is further highlighted by its ability to analyze a tumor-normal pair in approximately 11 minutes on average using 16 threads. This speed is more than 24 times faster than relying solely on read alignment processes on the same machine. In a simulation involving 20 tumors, joint graph calling using JANUS proves to be faster and more accurate than running independent pairwise runs. This method is particularly advantageous at low depths, ensuring both speed and accuracy.

By processing multi-sample cohorts jointly and maintaining sensitivity at low coverage and purity, JANUS offers a practical, reference-unbiased framework for cancer genomics. The software is freely accessible online at https://github.com/BilkentCompGen/janus, making it a valuable resource for researchers in the field of cancer genomics.