Thesis
Machine learning prediction of genetic hearing loss via selective intra-ensemble data partitioning
University of Iowa
Master of Science (MS), University of Iowa
Spring 2024
DOI: 10.25820/etd.007520
Abstract
This thesis reports on enhancements to AudioGenev9, an ensemble machine learning model. These enhancements involve a novel method - intra-ensemble data partitioning - to produce a diverse set of sub-models, and heuristics derived from domain knowledge regarding typical deafness phenotype-genotype associations. The context for this prediction is personalized genomic, or precision, medicine which has the potential to improve diagnosis leading to improved treatments and outcomes for patients. Previously, AudioGenev4, a patient-level support vector-based ML tool, has filled this niche as a supplemental diagnostic tool for predicting a hearing loss patient’s disease-causing gene, when autosomal dominant inheritance has been identified, using multiple audiograms. AudioGenev9, an audiogram-level ensemble prediction model, is the latest version of this pioneering tool. It is a publicly available alternative that offers improved overall predictive performance, from 70.9% to 77.8%, on single audiograms, achieved through careful consideration of the effects of imbalanced training data and selective data partitioning for ensemble sub-model training.
Details
- Title: Subtitle
- Machine learning prediction of genetic hearing loss via selective intra-ensemble data partitioning
- Creators
- Sean Ryan
- Contributors
- Thomas L. Casavant (Advisor)Terry A. Braun (Committee Member)Kishlay Jha (Committee Member)Sarah Gerard (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Biomedical Engineering
- Date degree season
- Spring 2024
- DOI
- 10.25820/etd.007520
- Publisher
- University of Iowa
- Number of pages
- xii, 64 pages
- Copyright
- Copyright 2024 Sean Ryan
- Language
- English
- Description illustrations
- Illustrations, tables, graphs, charts
- Description bibliographic
- Includes bibliographical references (pages 49-51).
- Public Abstract (ETD)
Hearing loss is the most common sensory deficit in the world, affecting about 20% of the global population including about 15% of American adults. While a gradual loss of hearing ability is unavoidable with aging, a significant portion of hearing loss caused by disease causing genetic mutations. For these people, a strong understanding of the source of their hearing loss from a early age can greatly impact their ability to reduce the impact of hearing loss on their lives. This thesis introduces an enhanced version of a machine learning model designed with this goal in mind: AudioGenev9.
AudioGenev9 is a tool designed to contribute to the field of personalized genomic medicine in hearing loss. Personalized genomic medicine is the practice of utilizing information about the genetic background of a patient to tailor medical treatments. At the extreme, this could mean selecting from thousands of treatment avenues for a single disease with great certainty about its potential for success. However, medical research is still in the early stages of developing diagnostic and treatment protocols for most diseases.
Progress on precision genetic diagnosis is a process that requires a high degree of collaboration between clinicians, geneticists, and researchers as well as the utilization of many tools. For hearing loss, there are over 125 genes associated with the disease and thousands of known mutations in each. AudioGenev9 can assist these experts in the process of determining which mutations could contribute to hearing loss by detecting patterns in the hearing loss experienced by a patient and suggesting which genes the disease causing mutation is in.
AudioGenev9 does this through a novel implementation of machine learning that utilizes several different models working in unison. In contrast to previous work that has utilized multiple types of models, AudioGenev9 uses unique methods of inserting knowledge from hearing loss experts to guide the models towards better predictions
- Academic Unit
- Roy J. Carver Department of Biomedical Engineering
- Record Identifier
- 9984647557802771
Metrics
1 File views/ downloads
21 Record Views