Machine learning for biological networks

Genetic studies often involve huge number of covariants that interact with each other, in the form of expressions or mutations. It is crucial to mine important covariants associated with different diseases for better clinical treatment. Traditional statistical methods have been successful in testing single covariants, but are limited when studying the joint effect of multiple related genes. Hence, incorporating biological interaction networks becomes a promising approach for genetic association study. On the other hand, the advance of graph learning algorithms has made it possible to build data-driven models for large graph problems. These methods generally fall into two categories: 1) random walk and 2) deep graph neural net. We study how to leverage information from biological networks under these frameworks to solve genetic association problems on large scale. Towards this end, we have applied graph neural network to cancer prognostic prediction. We also develop a network diffusion method for variant association study for Parkinson's disease. Our results demonstrate the power of graph learning algorithms in biological domain.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Ding, Hantian

Contributors dc:contributor

• Peng, Jian

• Machine Learning
• Bioinformatics

Statement dc:rights

• Copyright 2020 Hantian Ding

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108040

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108040