Misannotated Multi-Nucleotide Variants in Public Cancer Genomics Datasets Lead to Inaccurate Mutation Calls with Significant Implications

Although next-generation sequencing is widely used in cancer to profile tumors and detect variants, most somatic variant callers used in these pipelines identify variants at the lowest possible granularity, single-nucleotide variants (SNV). As a result, multiple adjacent SNVs are called individually instead of as a multinucleotide variants (MNV). With this approach, the amino acid change from the individual SNV within a codon could be different from the amino acid change based on the MNV that results from combining SNV, leading to incorrect conclusions about the downstream effects of the variants. Here, we analyzed 10,383 variant call files (VCF) from the Cancer Genome Atlas (TCGA) and found 12,141 incorrectly annotated MNVs. Analysis of seven commonly mutated genes from 178 studies in cBioPortal revealed that MNVs were consistently missed in 20 of these studies, whereas they were correctly annotated in 15 more recent studies. At the BRA]: V600 locus, the most common example of MNV, several public datasets reported separate BRAF V600E and BRAF V600M variants instead of a single merged V600K variant. VCFs from the TCGA Mutect2 caller were used to develop a solution to merge SNV to MNV. Our custom script used the phasing information from the SNV VCF and determined whether SNVs were at the same codon and needed to be merged into MNV before variant annotation. This study shows that institutions performing NGS sequencing for cancer genomics should incorporate the step of merging MNV as a best practice in their pipelines. Significance: Identification of incorrect mutation calls in TCGA, including clinically relevant BRAF V600 and KRAS G12, will influence research and potentially clinical decisions.