Deep Learning to Predict Outcome in Cancer Patients

Longitudinal surveillance of cancer evolution is pivotal for patient's individual medicine. Liquid biopsies including blood, urine, etcof cancer patients areideal to measure diverse tumour-derived substances, such as cell-free tumour DNA (ctDNA). The presence of certain microbiota, on the other hand,can influencesystemically tumour growth. Thus, making us of allthese factors may provide models supportive for clinical decision making.Diverse molecular data includinggene expression, mutational patternsand structural genomic alterationas well asdiverse microbiotas, will be collected across several solid cancers. Where possible,clinical-pathological data, treatment response and overall survival of cancer patients, as well as drug-sensitivity data from in vitromodelswill be obtained. The sum of all will be exploited, to identify robust markers in ctDNAs and the microbiome from cancer patients,indicativeof patient's responses to different treatment strategies. By implementing and comparing Bayesian statistical methods with deep learning techniques, the pros and cons of these analytical avenues will be tested and inform their application and molecular marker selection in ctDNA and microbiota compositions to predict treatment response.