Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science


convolutional neural networks, Gene similarity networks, graph embedding, multi-omics, representation learning, self-organizing maps


Rueda, L.




Deep learning models are currently applied in diverse domains, including image recognition, text generation, and event prediction. With the advent of new high-throughput sequencing technologies, a multitude of genomic data has been generated and made available. The representation of such data using deep neural networks, or for that matter, application of differential analysis has, however, not been able to match the growth of that data. One of the main challenges in applying convolutional neural networks on gene interaction data is the lack of understanding of the vector space domain to which they belong and also the inherent difficulties involved in representing those interactions on a significantly lower dimension viz Euclidean spaces. These challenges become more prevalent when dealing with various types of "omics" data with different forms. In this regard, we introduce a systematic, and generalized method, called iSOM-GSN, used to transform multi-omic genomic data with higher-dimensions into a two-dimensional grid. Afterwards, we apply a convolutional neural network (CNN) to predict disease states of various types. Based on the idea of the Kohonen's self-organizing map (SOM), we generate a two-dimensional grid for each sample for a given set of genes that represent a gene similarity network (GSN). The set of genes that are significantly highly mutated across the whole genome, are related to each other based on functional interactions. We then test the model to predict breast and prostate cancer stages using gene expression, DNA methylation, and copy number alteration, yielding accuracies in the 94-98% range for tumor stages of breast cancer and calculated Gleason scores of prostate cancer with just 14 input genes for both cases. To our knowledge, this is the first attempt to use self-organizing maps and convolutional neural networks on integrating high-dimensional multi-omics data. The scheme not only outputs nearly perfect classification accuracy, but also provides an enhanced scheme for visualization, dimensionality reduction, and interpretation of the results.