Interpreting Mutation Co-Occurrence in Cancer Genomics Under Biological Context.

Abstract

Somatic mutation patterns observed in cancer genomes are widely used to generate hypotheses about functional relationships among cancer genes and signaling pathways. However, mutation co-occurrence and mutual exclusivity are assessed at multiple levels, including cohorts, bulk specimens, lesions, regions, clones, and individual cells, although each observational level supports a different scope of inference. In this structured narrative review, we clarify these inferential boundaries and distinguish marginal from conditional association, as well as negative association from complete mutual exclusivity. A hypothetical numerical example of Simpson's reversal illustrates how marginal and conditional associations can differ and why negative association with non-zero overlap should be distinguished from complete mutual exclusivity. We then synthesize evidence from bulk, multi-region, phylogenetic, and single-cell analyses to examine spatial and clonal localization, interclonal cooperation, single-cell error and detection power, and genetic versus non-genetic resistance. We also provide a decision guide for method selection and a staged framework for functional validation. Overall, statistical association, physical localization, and functional interaction are related but distinct inferential targets that require different data, assumptions, and forms of validation.