MRC DiMeN Doctoral Training Partnership: Deep learning approaches to discover new roles for non-canonical signalling in cancer

Abnormal protein phosphorylation is strong associated with cancer aetiology. Considerable efforts have been undertaken to profile phosphorylation networks via serine, threonine and tyrosine residues and the responsible kinases/phosphatases. There is growing evidence that other amino acids, such as Histidine can be phosphorylated, but to date, the study of so-called non-canonical phosphorylation (on residues other than S, T and Y) has been severely limited due to lack of experimental tools. We have recently developed a method for unbiased phosphorylation site identification by mass spectrometry, leading to the discovery of widespread phosphorylation (through many 100s of sites) on non-canonical (pX) sites.
We are now exploring the complete phosphoproteome in key tumour cell models, such as breast cancer. In addition, there is a wealth of publicly available data on gene expression in various breast cancer sub-types. At present, it is largely unknown how the coherent system signals through canonical and non-canonical kinase functions, and why particular breast cancer sub-types respond poorly to therapy. In this project, we will integrate a wide array of data sets, both those publicly available, and those being generated in the Eyers’ laboratory, using machine learning to improve predictions of kinases-substrate relationships, and to determine new biomarkers for patient stratification.