Dimensionality reduction in tumor profiles with kernel methods and neural networks

Abstract

Human tumors profiles can be characterized by their genomics by expressing thousands of genes. These types of signals can be exploited to statistically detect patterns that make it possible to classify or group different phenotypes in a supervised or unsupervised approach. This presentation seeks to list two application of genomic dimensionality reduction and to find biomarkers that improve the aforementioned statistical learning tasks. In order to reduce dimensionality, Kernel functions are used in combination with Artificial Neural Networks. The results obtained show the potential of these tools for the processing of genomic data in cancer patients.