糖心TV

Speaker: Steven Hill (糖心TV Centre for Complexity Science)

Title:  Dynamic Bayesian analysis reveals protein signalling connectivity in individual cancers

Abstract:  Intracellular protein signalling plays an important role in the control of cell function and aberrations in signalling behaviour play a key role in the biology of cancer. However, we remain limited in our understanding of cancer-specific changes to signalling networks. Protein signalling involves combinatorial interactions between multiple proteins through time, ultimately driving downstream cellular effects. Here, we utilise stochastic models known as dynamic Bayesian networks to infer protein signalling connectivity using time-varying data from breast cancer cell lines. We take a Bayesian approach, incorporating existing biological knowledge into inference by means of an informative prior on network structure, weighted objectively by an empirical Bayes approach. Instead of resorting to approximate schemes, such as MCMC, we show how biochemically-motivated sparsity constraints permit exact inference to determine posterior probabilities of interest. We apply these approaches to high-throughput proteomic data from individual breast cancer cell lines. The cell lines we study are not only both breast cancers but furthermore belong to the same, well-characterized breast cancer subtype. Yet we find striking evidence of heterogeneity between these cancers with respect to signalling behaviour. Independent experiments validate some of these differences, suggesting that there may be considerable heterogeneity even within recently characterized cancer subtypes.

Lunch: group 2