PrediXcan, an imputed gene expression-trait association method, was compared to multiple linear regressions (MLR) of single nucleotide polymorphisms (SNPs) using the quantitative phenotypes serum total cholesterol (TC), low-density lipoprotein cholesterol (LDL), high-density lipoprotein cholesterol (HDL) and triglycerides (TG). The gene expression prediction models were trained using transcriptome- and genome-wide data from Depression Genes and Networks (DGN whole blood) and Genotype-Tissue Expression (GTEx) Project (GTEx whole blood, GTEx pancreas and GTEx liver). Linear combinations of the effect sizes derived using elastic net or least absolute shrinkage and selection operator (LASSO) with genotypes from 1304 European patients from the Diabetes Control and Complications Trial (DCCT) were used to estimate the genetically regulated expression (GReX) for genes. Different gene expression predictors were present in each training set. The 10-fold cross-validated predictive performance, estimated GReX, and p values from associations for matched genes were weakly correlated across training sets and strongly correlated for models derived using elastic net and LASSO. MLR models had more significant associations than PrediXcan models and larger inflation factors for p values. A comparison of p values for matched genes between PrediXcan and MLR models showed weak correlations but strong evidence for LDL and HDL associations with genes at locus 1p13.3 and 16q13, respectively. / Thesis / Master of Science (MSc)