The spread of cells from a primary tumour to a secondary site remains one of the most life-threatening pathological events. Epithelial-Mesenchymal Transition (EMT) is a cell programme involving loss of cell-cell adhesion, gain of motility, invasiveness and survival; these properties are fundamental for metastasis. Epithelial remodelling is also crucial for development. Reactivation of a programme resembling EMT is a credible mechanism for key aspects of the invasion-metastasis cascade and an MET-like process may produce the differentiation frequently observed in secondary tumours. Indeed, oncofetal signalling pathways (e.g. Hedgehog, Wnt, TGF-beta) activate EMT, and promote metastasis in multiple cancers.

Navigating from molecular measurements to phenotype implies understanding gene function. Many genes are poorly characterised, but coordinately regulated, and new functions continue to be discovered even for deeply studied genes. Thus, a substantial portion of gene function is uncharted. Data driven networks provide useful abstractions to fill these knowledge gaps.

We have developed techniques for mapping context-specific cell process networks and applied these to study EMT/MET; including to identify new EMT players, pathway crosstalk, functional transcription factor targets and drivers of metastasis. Orthologues of predicted EMT transcription factor targets in fly discriminated human breast cancer molecular subtypes and our analysis predicted new gene functions; for example, evidencing networks that reshape Waddington’s epigenetic landscape in epithelial remodelling. Predicted invasion roles were followed up in a tractable cell model and a novel druggable target was validated in an organotypic invasion assay. I will also discuss a novel algorithm for causal network inference, applied to combine ex vivo immunohistochemical measurements with clinical data in order to gain mechanistic insight and to predict molecular control of clinical parameters. Multivariate modelling controlling for clinical variables demonstrates that causal network based risk groups have significant prognostic value.