Deep Sparse Autoencoder-based Feature Selection for SNPs validation in Prostate Cancer Radiogenomics

Prostate cancer is the most diffused cancer affecting the male population. As therapies improve their effectiveness, surviving patients might be affected by complications induced by radiotherapy in the long run. To predict the onset of such rare late toxicities, because of the failure of phenotypic characteristics, the attention is shifting towards identifying specific genetic locations (Single Nucleotide Polimorphisms, or SNPs) associated with them. Because of the complexity of the problem, SNPs identified in a study are rarely validated on a different cohort of patients. In this case study we apply a novel approach for feature selection (namely a Deep Sparse Autoencoder-based Feature Selection method), to validate SNPs associated with radiotherapy-induced late toxicity causing Urinary Frequency Variation (UFV).