NEURAL NETS ARE NOT ALL YOU NEED: EVALUATING THE EFFECTS OF DEEP LEARNING ON TRANSCRIPTOMIC ANALYSIS

Loading...
Thumbnail Image
Degree type
Doctor of Philosophy (PhD)
Graduate group
Genomics and Computational Biology
Discipline
Bioinformatics
Computer Sciences
Biology
Subject
Computational biology
Deep learning
Machine learning
Funder
Grant number
License
Copyright date
2022
Distributor
Related resources
Author
Heil, Benjamin, J.
Contributor
Abstract

Technologies for quantifying biology have undergone significant advances in the past few decades, leading to datasets rapidly increasing in size and complexity. At the same time, deep learning models have gone from a curiosity to a massive field of research, with their advancements spilling over into other fields. Machine learning is not new to computational biology, as machine learning models have been used frequently in the field to account for the aforementioned size and complexity of the data. This dissertation asks whether the paradigm shift in machine learning that has led to the rise of deep learning models is causing a paradigm shift in computational biology. To answer this question, we begin with chapter 1, which gives background information helpful for understanding the main thesis chapters. We then move to chapter 2, which discusses standards necessary to ensure that research done with deep learning is reproducible. We continue to chapter 3, where we find that deep learning models may not be helpful in analyzing expression data. In chapters 4 and 5 we demonstrate that classical machine learning methods already allow scientists to uncover knowledge from large datasets. Then in chapter 6 we conclude by discussing the implications of the previous chapters and their potential future directions. Ultimately we find that while deep learning models are useful to various subfields of computational biology, they have yet to lead to a paradigm shift.

Advisor
Greene, Casey, S.
Holmes, John, H.
Date of degree
2023
Date Range for Data Collection (Start Date)
Date Range for Data Collection (End Date)
Digital Object Identifier
Series name and number
Volume number
Issue number
Publisher
Publisher DOI
Journal Issue
Comments
Recommended citation