Tumours are mixtures of phylogenetically related cancer cell populations or *clones*, which are subject to a process of Darwinian evolution in response to selective pressures in their local micro-environment. `clonosGP` is a statistical methodology for tracking this latent heterogeneity continuously in time based on longitudinally collected tumour samples. In technical terms, it combines Dirichlet Process Mixture Models with Gaussian Process Priors to identify clusters of mutations and track their cellular prevalence continuously in time. If only cross-sectional data are available, then it performs standard non-parametric clustering of mutations based on their observed frequency, similarly to `PyClone` and other software in the same category. The statistical models underlying `clonosGP` were implemented in the excellent probabilistic programming system `PyMC3` on which we also rely for inference using variational methods.
0 commit comments