Neural Latent Filtering for Gene Discovery in Breast Cancer Subtypes

Gene selection from expression data represents a challenging task, primarily due to the high data dimensionality and the vast number of genes that would be identified, many of which may be unrelated to cancer-relevant biological processes. To tackle this issue, filtering methods constitute an effective solution to identify the most informative genes, which can serve as potential biomarkers to tailor cancer therapies. This work proposes a novel neural-based filtering approach which aims to identify genes by means of their latent representation extracted from RNA Sequencing expression data. This approach has been applied to study breast invasive carcinoma dataset, aiming to identify the most relevant genes of two breast cancer subtypes, Luminal-A and Basal-like, to better investigate their molecular landscape.