Evaluating Discordant Somatic Calls Across Mutation Discovery Approaches to Minimize False-Negative Drug-Resistant Findings

Evaluating robustness of somatic mutation detections is essential when using whole-exome sequencing (WES) for treatment decision-making. A comprehensive evaluation was conducted using tumor WES from the US Food and Drug Administration-led Sequencing Quality Control Phase 2 project, in which multiple library kits sequenced identical DNA materials across three laboratories to benchmark analytical validity. These workflows included various read aligner (BWA, Bowtie2, DRAGEN-Aligner, DRAGMAP, and HISAT2) and mutation caller (Mutect2, TNscope, DRAGEN-Caller, and DeepVariant) combinations. The results revealed that DRAGEN exhibited superior performance, achieving mean F1 scores of 0.966 and 0.791 for single-nucleotide variant and insertion/deletion detection, respectively. Among open-source software, BWA Mutect2 and HISAT2 Mutect2 combinations showed the highest mean F1 scores for single-nucleotide variant (0.949) and insertion/deletion (0.722), respectively. The analyses indicated that high-quality data can be analyzed as having worse results, and vice versa. Evaluations of Catalog of Somatic Mutations in Cancer reported mutations unveiled discrepancies across enrichment kits. IDT enrichment kits showed a higher false-negative rate, whereas Agilent WES kits tended to miss mutations in CBL and IDH1, and Roche library kits tended to miss the mutations in PIK3CB. For drug-related biomarkers, Sentieon TNscope tended to underestimate tumor mutation burden and overlook crucial drug-resistance mutations, such as FLT3 (c.G1879A: p.A627T) for cytarabine resistance in leukemia and MAP2K1 (c.G199A:p.D67N) for BRAF inhibitors in melanoma. The findings highlight the importance of robust bioinformatic analysis in identifying tumor mutations and guiding clinical decision-making.