Detecting repeated cancer evolution from multi-region tumor sequencing data

Abstract

Recurrent successions of genomic changes, both within and between patients, reflect repeated evolutionary processes that are valuable for the anticipation of cancer progression. Multi-region sequencing allows the temporal order of some genomic changes in a tumor to be inferred, but the robust identification of repeated evolution across patients remains a challenge. We developed a machine-learning method based on transfer learning that allowed us to overcome the stochastic effects of cancer evolution and noise in data and identified hidden evolutionary patterns in cancer cohorts. When applied to multi-region sequencing datasets from lung, breast, renal, and colorectal cancer (768 samples from 178 patients), our method detected repeated evolutionary trajectories in subgroups of patients, which were reproduced in single-sample cohorts (n = 2,935). Our method provides a means of classifying patients on the basis of how their tumor evolved, with implications for the anticipation of disease progression.

Identification of Conserved Evolutionary Trajectories in Tumors

“…Let P be a set of tumors and {G p }(p ∈ P) its corresponding collection of tumor graphs -a tumor graph G p is a directed graph corresponding to tumor p ∈ P where each node corresponds to either a somatically altered single gene in the tumor, or a copy number altered entire chromosomal arm (labelled by the gene or chromosome and the type of alteration affecting it) 2 . A directed edge from one node to another indicates temporal precedence of the former alteration event to the latter, possibly inferred through the use of time series, multi-region, single cell or single molecule sequencing data (similar to what was considered by [4]).…”

“…Since, until recently, no computational tool was purposefully developed to identify recurrent patterns of tumor evolution across multiple tumor samples, these studies typically relied on manual inspection of all possible sequences of alteration events, starting from the root of the tumor phylogeny (representing germline), all the way to leaf nodes [29]. We are aware of only two recent exceptions, which are both based on clustering tumor phylogenetic trees by the use of tree similarity/distance measures [25,4]. In particular, the REVOLVER method [4] employs a maximum-likelihood learning strategy to construct a joint hidden tree model for all tumors in a cohort, which is then used to infer an individual tree depicting the temporal order of clonal alterations in each individual tumor.…”

“…We are aware of only two recent exceptions, which are both based on clustering tumor phylogenetic trees by the use of tree similarity/distance measures [25,4]. In particular, the REVOLVER method [4] employs a maximum-likelihood learning strategy to construct a joint hidden tree model for all tumors in a cohort, which is then used to infer an individual tree depicting the temporal order of clonal alterations in each individual tumor. This is followed by a hierarchical clustering of the trees for the purpose of detecting edges shared within each cluster (and manually constructing likely evolutionary trajectories involving those edges).…”

Uncovering the subtype-specific temporal order of cancer pathway dysregulation

“…[22–28]), which is not possible in a tree model. A recent publication using an ensemble approach [29] leveraged non-binary cancer cell fraction data which provides more information about the timing of mutations.…”

Uncovering the subtype-specific temporal order of cancer pathway dysregulation

“…[25,22,3,6,7,45,46]), which is not possible in a tree model. A recent publication using an ensemble approach [12] leveraged non-binary cancer cell fraction data which provides more information about the timing of mutations.…”