Abstract

Cancer genome analysis still relies largely on universal reference genomes, which can obscure patient-specific variation in repetitive, polymorphic, and structurally complex regions. Here, we generated personalized, haplotype-resolved near-gapless assemblies from paired hepatocellular carcinoma (HCC) tumor and matched normal tissues using PacBio HiFi, ultra-long Oxford Nanopore, and Hi-C sequencing. The matched normal assembly served as a personalized reference for tumor analysis, improving read alignment relative to GRCh38, reducing multi-mapping, and achieving alignment performance comparable to T2T-CHM13. This patient-specific coordinate framework identified additional somatic small variants and structural variants, including variants missed or ambiguously represented by standard references. These assemblies also improved access to centromeric regions that are poorly captured by standard references, identifying patient-specific tumor-associated changes in assembled centromeric sequence content and repeat composition. Haplotype-resolved transcriptomic analysis identified 205 allelic pairs with allele-specific expression, including six dual protein-coding tumor-restricted candidates. Multi-omic integration further prioritized candidate regulatory alterations in MET and HSD17B2 , supported by allele-dependent luciferase reporter activity. These findings support a patient-centric assembly framework that complements standard references for resolving coding and non-coding somatic alterations in HCC and provides a foundation for future cohort-scale validation.