Techniques to Improve Deep Learning for Phenotype Prediction from Genotype Data
Date
2020-11-24Author
Kopas, Logan
ORCID
0000-0002-5525-7001Type
ThesisDegree Level
MastersMetadata
Show full item recordAbstract
We show that by representing Single Nucleotide Polymorphism (SNP) data to a neural network in a way that incorporates quality scores and avoids filtering out low quality SNPs we are able to increase the effectiveness of a deep neural network for phenotype prediction from genotype in some cases. We also show that we are able to significantly increase the predictive power of a neural network by making use of transfer learning. We demonstrate these results on a Whole Genome Sequencing (WGS) Neisseria gonorrhoeae dataset where we predict Antimicrobial Resistance (AMR) as well as on an exome sequencing Lens culinaris dataset where we predict 3 growing rate phenotypes.
Degree
Master of Science (M.Sc.)Department
Computer ScienceProgram
Computer ScienceSupervisor
Kusalik, Tony; Schneider, DaveCommittee
Stavness, Ian; Bett, Kirsten; Zhang, XuekuiCopyright Date
August 2020Subject
Deep learning
bioinformatics
genotype
phenotype prediction