Conditional prediction of consecutive tumor evolution using cancer progression models: What genotype comes next?

Abstract

Accurate prediction of tumor progression would be helpful for adaptive therapy and precision medicine. Cancer progression models (CPMs) can be used to infer dependencies in mutation accumulation from cross-sectional data and provide predictions of tumor progression paths. But their performance when predicting the complete evolutionary paths is limited by violations of assumptions and the size of available data sets. Instead of predicting full tumor progression paths, we can focus on short-term predictions, more relevant for diagnostic and therapeutic purposes. Here we examine if five distinct CPMs can be used to answer the question 'Given that a genotype with n mutations has been observed, what genotype with n + 1 mutations is next in the path of tumor progression' or, shortly, 'What genotype comes next'. Using simulated data we find that under specific combinations of genotype and fitness landscape characteristics CPMs can provide predictions of short-term evolution that closely match the true probabilities, and that some genotype characteristics (fitness and probability of being a local fitness maximum) can be much more relevant than global features. Thus, CPMs can provide short-term predictions even when global, long-term predictions are not possible because fitness landscape- and evolutionary model-specific assumptions are violated. When good performance is possible, we observe significant variation in the quality of predictions of different methods. Genotype-specific and global fitness landscape characteristics are required to determine which method provides best results in each case. Application of these methods to 25 cancer data sets shows that their use is hampered by lack of the information needed to make principled decisions about method choice and what predictions to trust. Fruitful use of these methods for short-term predictions requires adapting their use to local genotype characteristics and obtaining reliable indicators of performance, as well clarifying the interpretation of their results when key assumptions do not hold.

Joint inference of exclusivity patterns and recurrent trajectories from tumor mutation trees

“…Since clonal interactions have a major impact on intratumor heterogeneity and the observed evolutionary trajectories, it would be beneficial to exploit the pairwise frequencies of mutational events within and between subclones when modelling tumor progression. For cross-sectional bulk sequencing data, a collection of probabilistic methods for inferring the temporal order of mutations and predicting tumor progression is called cancer progression models (CPMs) (Beerenwinkel et al, 2015;Diaz-Colunga and Diaz-Uriarte, 2020). In light of the observation that mutually exclusive mutations are often associated with genes in the same functional pathway (Yeang et al, 2008;Vandin, 2017), Raphael and Vandin (2015) developed the first CPM that jointly infers sets of mutually exclusive mutations and the temporal order among them using a chain model.…”

“…For cross-sectional bulk sequencing data, where each tumor is summarized by a binary genotype, a collection of probabilistic methods for inferring the temporal order of mutations and predicting tumor progression is called cancer progression models (CPMs) [25, 26]. In light of the observation that mutually exclusive mutations are often associated with genes in the same functional pathway, Raphael & Vandin developed the first CPM that jointly infers sets of mutually exclusive mutations and the temporal order among them using a chain model [27].…”

EvAM-Tools: tools for evolutionary accumulation and cancer progression models

“…These dependencies can be deterministic, as in Oncogenetic Trees (OT) (Desper et al ., 1999; Szabo and Boucher, 2008), Conjunctive Bayesian Networks (CBN) (Gerstung et al ., 2009; Montazeri et al ., 2016), Disjunctive Bayesian Networks (DBN) (Nicol et al ., 2021), and Hidden Extended Suppes-Bayes Causal Networks (H-ESBCNs) (Angaroni et al ., 2021), or stochastic as in Mutual Hazard Networks (MHN) (Schill et al ., 2020). These models also implicitly encode the possible mutational trajectories with predictions about their probability, and have been used for predicting cancer evolution, both long-term (Diaz-Uriarte and Vasallo, 2019; Hosseini et al ., 2019) and short-term (Diaz-Colunga and Diaz-Uriarte, 2021). Although developed in the field of computational oncology, these models are not limited to cancer: they can be applied to other questions involving the (irreversible) accumulation of discrete items (Gotovos et al ., 2021), and have been used to examine tool use in animal taxa (Johnston and Røyrvik, 2020).…”