Date of Award

11-21-2019

Publication Type

Master Thesis

Degree Name

M.Sc.

Department

Computer Science

First Advisor

Dan Wu

Keywords

Deep Neural Network, Gene Interaction Network, Gleason Score prediction, Graph Convolutional Neural Networks

Rights

info:eu-repo/semantics/openAccess

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Abstract

From social networks to biological networks, graphs are a natural way to represent a diverse set of real-world data. This research presents attributed graph convolutional neural network with a pooling layer (AGCP for short), a novel end-to-end deep neural network model which captures the higher-order latent attributes of weighted, labeled, undirected, attributed graphs of arbitrary size. The architecture of AGCP is an efficient variant of convolutional neural network (CNN) and has a linear filter function that convolves over the fixed topological structure of a graph to learn local and global attributes of the graph. Convolution is followed by a pooling layer that coarsens the graph while preserving the global structure of the original input graph using information gain. On the other hand, advances in high throughput technologies for next-generation sequencing have enabled machine learning research to acquire and extract knowledge from biological networks. We apply AGCP on three bioinformatics networks, ENZYMES, D&D, and GINA a graph dataset of gene interaction networks with genomic mutation attributes as the attributes of the vertices. In several experiments on these datasets, we demonstrate that AGCP yields better results in terms of classification accuracy relative to the previously proposed models by a considerable margin.

Share

COinS