Molecular Dynamics of the Response to Breast Cancer Therapies

Surgery is still the gold standard approach for treating breast cancer. The majority of pharmacological therapies are given after tumour excision, rendering it difficult to
assess how cancer cells respond directly in patients. However, pre-surgical treatment has become more popular over the past decade for patients with breast cancer to downstage tumours, so that less invasive operations can be performed (e.g. avoiding mastectomy) and potentially improve predictions of patient response as it this is partly
determined by multiple factors, including unique individual characteristics. Over the last 10 years the Sims group have collected data from gene expression
profiling of breast cancer patients at different timepoints: pre-treatment, on-treatment (most commonly at the 2-week checkpoint) and at excision. Patients are labelled as
"responders" and "non-responders". Due to the data being collected from different studies, it is quite challenging to process them as a whole in their raw format. A
universal framework needs to be established in order to curate and integrate all necessary information and allow meaningful comparisons, increased statistical power
and more robust conclusions. Analysis of these high-dimensional datasets is likely to yield significant insight into inter-patient variability regarding the same treatment approach, as well as into the differences induced by different treatment approaches and finally the effect of each treatment at various timepoints. Also, it will allow for the identification of useful biomarkers that can help clinicians make early predictions, guide patient stratification and achieve enhanced outcomes, thus, ultimately, preventing invasive procedures and unnecessary treatment strategies with undesirable side-effects. Methodologically, the aggregated curated data will be analysed using advanced computational and statistical approaches. Machine Learning algorithms will be employed to understand the data's properties and structure at a deeper level. Modern dimension-reduction techniques, such as the Random Projection Ensemble classifier (co-developed by Dr. Timothy Cannings) will also be used to reduce the complexity of the data and facilitate analysis. Eventually, results from this analysis will be used to allocate patients to appropriate treatment groups (that will lead to improved outcomes) depending on their respective measurements regarding pre- and on-treatment biomarkers that have been found to be of particular predictive value.