Improving Genomic Alignment Accuracy and Variant Detection for Acute Lymphoblastic Leukaemia Patients: A Pan-Genome Graph Approach

Abstract Background Acute lymphoblastic leukaemia (ALL) is characterised by extensive genomic heterogeneity, including single nucleotide variants, insertions, deletions, and complex rearrangements that are critical for diagnosis, subclassification, and risk stratification. Conventional short-read mRNA sequencing workflows frequently rely on alignment to a single linear reference genome, which incompletely represents disease-associated variation, leading to reference bias, misalignment, and reduced variant detection sensitivity. Pan-genome graphs provide an alternative reference structure by explicitly representing known genomic variation. Results We developed a spliced pan-genome graph tailored to acute lymphoblastic leukaemia by integrating disease-associated variants and transcript annotations into the human GRCh38 reference genome. Performance was evaluated using simulated datasets with known ground truth and mRNA sequencing data from ten ALL patients. Graph-based alignment consistently outperformed linear alignment, demonstrating higher mapping precision, improved mapping quality calibration, and increased proportions of Primary and Properly Paired reads. These improvements led to a substantial enhancement in variant detection, as the graph-based approach detected over seven times more unique insertions and deletions and nearly three times more unique single nucleotide variants relative to the linear workflow. Performance gains were most evident for insertions and deletions greater then 10 base pairs, which are commonly overlooked by linear reference alignment approaches when using short-read sequencing data. While graph-based alignment required longer runtime, it achieved lower peak memory usage compared with the linear approach. Conclusions A disease-specific spliced pan-genome graph significantly enhances transcriptomic alignment accuracy and variant detection in ALL. The marked improvement in insertion and deletion recovery underscores the clinical value of graph-based references for identifying functionally disruptive variants that are critical for genomic classification. These findings support the adoption of disease-aware pangenome graphs as a practical strategy to improve genomic analyses in precision oncology.