Graduate Studies
Utilizing Statistical and Machine Learning Techniques to Improve Tools Identifying Genetic Basis Underlying Complex Traits
Abstract
dc:description.abstractIn genetic studies, understanding the genetic basis underlying complex traits takes center stage. Modern high-throughput technologies have empowered researchers to generate large amounts of multi-scale omics data that bridge the gap between genotype and phenotype. With the availability of omics data, researchers have successfully identified genetic variants and genes underlying complex traits using genome-wide association studies (GWAS) and other association mapping methods using omics data as mediators, e.g., transcriptome-wide association studies (TWAS). To provide better statistical tools unraveling the genetic basis of complex traits, I have developed three statistical methods: AE-TWAS, edLMM, and Kernel-smoothed permutation, respectively focusing on three stages of association mapping, namely input data denoising, statistical test to detect low-effect variants, and non-parametric generation of small P-values. My first project delves into the limitations of TWAS from the perspective of transcriptomic data denoising by transformation. I introduced AE-TWAS (AutoEncoder-transformed TWAS), which incorporates autoencoder-transformed transcriptome data into a standard TWAS protocol. We showed that AE-TWAS boosts the heritability of low-heritability genes in transcriptomic data and the transformed transcriptome leads to better performance of TWAS. In my second project, I introduced edLMM (expression- directed Linear Mixed Model) that incorporates weights derived from transcriptomic data in a linear mixed model (LMM). These weights provide an alternative estimate of the null hypothesis in an LMM by re-scaling the genetic relationship matrix (GRM). Through real data analysis and simulation studies, we demonstrated that edLMM is more powerful than the benchmark LMM method in GWAS (i.e., EMMAX), especially in its capacity to enhance the identification of genetic variants with low effects. My third project focused on the challenge of accurately estimating small p-values in genetic association studies using permutations when the closed-form distribution of test statistics is unknown. Specifically, I developed Kernel-smoothed permutation, a method that constructs the null distribution of a test statistic through a novel kurtosis-driven Box-Cox transformation, followed by a kernel- based density estimation (KDE). We demonstrated that our approach requires a magnitude of lower rounds of permutations than the conventional Na ̈ıve permutation. Taking together, my thesis contributed meaningful additions to the toolbox of association mapping from three abovementioned aspects, enhancing our ability to identify the genetic basis of complex traits in the cases of low expression heritability genes, low effect-size genetic variants, and low numbers of sampled permutations. These works push the boundary of association mapping to the corners that may be missed by using conventional tools.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (PhD)
- Discipline thesis:degree_discipline
- Mathematics & Statistics
- Grantor dc:publisher.institution
- Graduate Studies
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bian, Jiayi
- Advisors dc:contributor.advisor
-
- Wu, Jingjing
- Long, Quan
- Committee members dc:contributor.committeemember
-
- de Koning, Jason
- Chekouo Tekougang, Thierry
- Lu, Xuewen
- Bureau, Alexandre
Rights
dc:rights- Statement dc:rights
-
- University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Citation
- Bian, Jiayi (2023). Utilizing statistical and machine learning techniques to improve tools identifying genetic basis underlying complex traits (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.