- The authors investigate intratumoral genetic heterogeneity by performing Bayesian inference on a population-genetics model of asexual evolution, using data from high-coverage bulk sequencing data. The model combines a generative model for tumour development with an error model for sequencing.
- Previous work by the authors showed that, under a neutral evolutionary model, the variant-allele fractions (VAFs) (i.e. the percentage of genomes which are mutated in a particular allele) follow a power-law distribution. Subsequent work by the authors showed that by modelling spatial effects and selection, the authors could infer whether a particular variant was neutral or non-neutral
- In this work, the authors use a stochastic branching process model, whereby cells divide and die with particular rates, and acquire de novo mutations upon division. Mutant subclones are assigned a fitness advantage, which is related to the ratio of replication rate of the mutant to the background host population. Clones which have a selective advantage induce an additional peak in the distribution of VAFs.
- The mean VAF of a particular cluster is a measure of its relative size within the tumour; the total number of distinct mutations in the cluster is a measure of its age, since older subclones have had more time to accrue mutations. These two pieces of information allow the replication rate of the particular subclone to be constrained, and its selective advantage to be inferred.
- Note that mutations can hitchhike with the actual driver event, so it is not necessarily the case that all mutations with a surprisingly high VAF cause a selective advantage. The driver event may not necessarily even be genetic in origin.
- Using this framework, the authors discover detectable subclones which were consistently present, with a fitness advantage >20%.