Seminar Room, Herschel Smith Building, University Forvie Site
Please note the change of time and location. The Hershel Smith Building is across the car park from the Institute of Public Health. For a map, see https://map.cam.ac.uk/Forvie+Site.
Dr Paul Kirk
University of Oxford
“Bayesian Inference for Tumour Heterogeneity using the Hamming Ball Sampler”
Abstract:The characterisation of sub-clonal structure from whole genome sequencing of heterogeneous tumours is a computationally challenging problem due to the large number of mutations and unknown number of tumour sub-populations. Many recent approaches have adopted methods based on Bayesian nonparametric models (the Dirichlet or Indian Buffet Process) but standard computational methods for inference for these models can be slow as they rely on inefficient Markov Chain Monte Carlo methods. We have developed a novel Markov Chain Monte Carlo sampling algorithm that allows efficient and exact Bayesian inference for multiple membership models that is scalable to massive datasets. While standard approaches use asymmetric updates that are prone to becoming trapped in local modes, our algorithm uses locally symmetric modes that allows joint updating of the latent variable structure. We illustrate the utility of the sampling algorithm for tumour deconvolution problems and demonstrate its advantage over standard computational alternatives using in vitro mixtures of tumour DNA.
To see the full programme for the MRC Biostatistics Unit Seminar Series this term please visit https://www.mrc-bsu.cam.ac.uk/news-and-events/seminars/