Researchers Develop Framework for Using Cancer Progression Models to Identify Therapeutic Targets

A preprint posted to arXiv introduces a structural causal framework for applying Pearl's do-operator and conditional interventions to evolutionary accumulation models (EvAMs), also called cancer progression models (CPMs). These models use cross-sectional genomic data to infer dependencies in the order of mutation accumulation during tumor progression, and have long been suggested as tools for identifying therapeutic targets. The authors demonstrate that a naive approach — simply conditioning on the absence of a mutation — yields incorrect predictions, and they provide correct intervention procedures for seven major EvAM methods: OT, OncoBN, CBN, H-ESBCN, MHN, HyperHMM, and HyperTraPS. For each method, the paper specifies how to implement interventions, often as parameter modifications, and examines whether the modularity assumption required for well-defined interventions is satisfied. The framework also distinguishes between two biologically distinct intervention types — 'killing' and 'inactivating' a mutation — which are conflated in standard EvAM representations, and recasts the problem as one of ranking candidate targets rather than binary prediction. The authors note the framework is general and applies to any fitted computational model interpretable as a structural causal model, with accompanying code made publicly available.