Keynote Speech :
Title: Matrix Trifactorization and Convolutional Neural Network Integrating Gene Expression Data and Molecular Networks
Abstract:
Novel methods for integrating different types of "omics" data, from gene expression profiles and sequences
biomolecular networks, are promising to not only increase accuracy on disease gene identification but also to shed light on our understanding
of complex molecular processes and human diseases.Exhaustive experimental studies have unveiled common roots between certain
human diseases and specific mutations and deregulations of noncoding RNAs (ncRNAs). In parallel to these biological discoveries, researchers have developed several
computational models to predict ncRNA-disease associations. Here, we integrate sequence information from ncRNAs and proteins with a tripartite molecular network model
defined by ncRNA-protein targets and human disease relationships. By using this model, we predict associations between ncRNAs and human diseases.
In particular, we devised a variant of the nonnegative matrix trifactorization data integration approach that does not require input information
for known ncRNA-disease associations. On the other hand, while convolutional neural network methods have shown its potential in several fields, its application to certain
"omics" data such as molecular network structures and gene expression data has been less investigated. Here, we show a novel
method that allows us to analyze protein-protein interaction networks integrating transcriptomics data using convolutional neural network techniques. The
developed method is then applied to classify cancer samples, showing a better predictive performance than other existing methods.