We introduce the probabilistic model class of mixtures of oncogenetic trees for estimating typical tumor-specific pathogenetic routes. In these models, progression is characterized by the ordered accumulation of permanent genetic changes. We derive a genetic progression score (GPS) from the tree models that estimates the genetic status of a tumor. Using Cox regression models we demonstrate that the GPS is a medically relevant prognostic factor. For several cancer types, the GPS can be used to discriminate between patient subgroups with different clinical outcome.
Our tree models for estimating tumor progression are currently applied to chromosomal alterations. Extending our approach to changes on gene level will enable a more precise determination of relevant genomic regions and potential target genes.