Hierarchical model for neutral-evolution-aware subclonal deconvolution over different karyotypes. This model is a bayesian formulation and expansion of the mobster algorithm (Caravagna et al; PMID: 32879509).
The main novelties are:
- Support for more than one karyotype at the time (mobsteh works with karyotypes 1:0, 1:1, 2:0, 2:1, 2:2)
- Full bayesian implementation with meaningfull priors (no need to bootstrap anymore)
- Hierachical estimation of the mutation rate from different karyotypes (with credible intervals)
- Explicit separation of clonal and subclonal clusters
- GPU support and fast inference with VI and Pyro
To install just run:
pip install HMOBSTER
This package has very few functionalities other than the model inference itself.
We suggest using directly the mobsterh_fit function in the R pakage which provides a complete interface to the python backend.