Abstract 5050: Comparative analysis of somatic copy number variations and allelic imbalance in whole-genome bisulfite sequencing: challenges and insights

Goal: To assess the feasibility, reliability, and limitations of using Whole Genome Bisulfite Sequencing (WGBS) for detecting somatic copy number variation (SCNV) and tumor-specific allelic imbalance (AI). Background: In cancer genomics, the use of WGBS for SCNV and AI detection enables simultaneous analysis of genetic variations and methylation landscapes, potentially providing insights into the interplay between somatic alterations and DNA methylation in tumorigenesis. However, WGBS data is inherently noisier compared to whole-genome sequencing (WGS) or microarray data due to artifacts introduced during bisulfite conversion. These challenges, including difficulties in read alignment, genotype calling, and phasing, the lack of strand complementarity, and the need to distinguish true C>T SNPs from bisulfite-induced changes, collectively complicate SCNV and AI detection. Method: We analyzed WGBS data from The Cancer Genome Atlas (TCGA), including 39 tumor samples and 8 adjacent normal tissues across 9 tumor types. SCNV detection was performed using GATK-CNV, while AI callings were conducted with BisSNP, BISCUIT, Revelio (with Unified Genotyper), and BSsnp. Confusion matrices were used to compare GATK-CNV (WGBS) and hapLOH (microarray) at the base, gene, and arm levels. F1 score, specificity, and sensitivity were calculated to evaluate their performance across different genomic scales. Additionally, AI identified from WGBS data using BisSNP, BISCUIT, Revelio (with Unified Genotyper), and BSsnp were compared to those detected from array and WGS data using hapLOH/seq. To determine detection thresholds and refine results, sequencing data were downsampled, with filters applied to exclude low-quality data and retain high-confidence SNPs and genotypes. Result: GATK-CNV exhibited variable performance across samples, with most samples showing high true positives and low false positives, indicating good sensitivity for detecting SCNVs and capturing a genome burden similar to hapLOHseq data. However, for samples with a high number of events, normalization disrupts event callings, resulting in inconsistencies in SCNV identification compared to array data. For AI detection, Revelio (with Unified Genotyper) showed the best overall performance among the tools evaluated. Downsampling analysis showed that reducing sequencing coverage impaired event detection, becoming less effective at lower coverage levels. BSsnp, BISCUIT, and BisSNP produced widespread and inconsistent variant calls, and initial filtering for high-quality SNPs showed limited improvement in precision. This study highlights the potential and challenges of WGBS as a tool for analyzing SCNV/AI alongside DNA methylation with phased data, offering insights into tumorigenesis and advancements in cancer diagnostics and treatment. Citation Format: Yiru Gu, Jian Gu, Paul Scheet, Justin Wong. Comparative analysis of somatic copy number variations and allelic imbalance in whole-genome bisulfite sequencing: challenges and insights [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5050.