Fuscan: a robust DNA fusion caller for targeted sequencing data in cancer diagnostics.

Motivation: Gene fusions resulting from genomic structural variation in somatic cells have been increasingly identified as central events driving oncogenesis. Ultra-deep targeted sequencing of driver fusions informs therapeutic selection in precision oncology. However, most structural variant (SV) callers were primarily architected for whole genome sequencing, failing to resolve the artifacts and alignment errors that drive false-positive calls in high-depth targeted data. Results: Here, we describe Fuscan, a robust DNA fusion caller specifically optimized for targeted sequencing data to identify oncogenic drivers. Fuscan improves sensitivity by focusing alignment on targeted driver sequences while simultaneously filtering homologous genomic regions to prevent false-positive partner-gene breakpoints. We performed targeted sequencing on 85 non-small cell lung cancer clinical specimens (comprising tissue and body fluids), four SV reference standards at 0.5% allele frequency, and 282 healthy-control leukocyte samples. We benchmarked Fuscan against established SV callers, achieving an area under the curve (AUC) of 0.992 and demonstrating its robustness in challenging clinical scenarios, including low-tumor-content tissues and liquid biopsies. Availability and implementation: Fuscan is available on our GitHub repository: https://github.com/YJmedLab/Fuscan.