Minerva: High-Resolution Allele-Specific Copy Number and Complex SV Harmonization with Long Reads

Copy-number alterations (CNAs) and structural variations (SVs) are major drivers of cancer evolution, yet accurately resolving allele-specific copy number (ASCN) in highly rearranged tumor genomes remains challenging. Conventional approaches often struggle with limited phasing information and inaccurate segmentation. Long-read sequencing offers unique advantages for genome reconstruction, but current CN tools do not fully exploit its long-range phasing or breakpoint-level resolution.We present Minerva, a long-read framework for ASCN calling and complex SV clustering. Minerva integrates high-fidelity structural variation calls from Severus into a breakpoint graph that defines segmentation directly on rearrangement breakpoints. This approach enables chromosome-arm-scale phasing, allowing haplotypes to be followed through long and structurally complex regions. It further provides haplotype-specific coverage estimates and accurate CN assignment even across SV clusters. Minerva supports both tumor-normal and tumor-only workflows, delivering stable ploidy estimation and haplotype-aware CN inference even in the absence of matched normals. Using the CASTLE long-read somatic cancer cell line panel, we benchmarked Minerva against established short-read and long-read CN methods. Across nearly all cell lines, Minerva achieved higher chromosome-scale CN accuracy, more consistent purity/ploidy estimates, and markedly improved detection of focal CN changes, including deletions, duplications, or more complex events like translocations with deletions. Leveraging the precision of long-read SV breakpoints, Minerva resolves sub-100 bp CN segments, representing orders-of-magnitude finer resolution than short-read approaches. Performance gains were strongest in genomes with dense rearrangements, including fold-back inversions, sysmic amplifications, templated insertions, and ecDNA. We further applied Minerva to (i) a breast cancer long-read cell-line panel and (ii) a long-read breast tumor cohort. Across both datasets, Minerva identified distinct and recurrent amplification patterns in key oncogenic regions, including MYC, ERBB2, and CCND1, and revealed allele-specific focal events and complex rearrangement architectures missed by existing CN/SV methods. Minerva provides a unified, phasing-aware solution for high-resolution CN and SV interpretation in tumor-only long-read cancer genomics, enabling more accurate reconstruction of cancer genome structure, selective pressures, and oncogenic amplification landscapes.