• Explored linear mixed-effect models to quantify mediating effects of genetically regulated expression on complex traits, which was increased by 14% by incorporating functional gene annotations• Developed a Bayesian model to identify cell-type specific and shared single-cell eQTLs by using an EM algorithm to leverage cell type similarities, which was applied in the OneK1K dataset, resulting in the identification of 18% more signal genes and approximately two-fold shared effects• Collaborated with two research scientists at Boehringer Ingelheim to conducted genome-wide association studies (GWAS) to identify novel genetic variants statistically significantly associated with complex traits, such as idiopathic pulmonary fibrosis (IPF), metabolic dysfunction-associated steatohepatitis (MASH) and systemic sclerosis (SSc)

• Teaching Fellow for Machine Learning with Biomedical Data (Fall 2022).• Tutored a Ph.D. student to prepare for the Biostatistics qualifying exam, which they successfully passed

Worked at Base Pricing team on a core model supporting pricing strategy decisions, assessing the causal impact of the price changes on ride volume and financial outcomes• Extended the model to forecast the long-term quarterly effects of the pricing adjustments• Built a PyTorch-based Bayesian model to estimate credible intervals of the prediction results, accurately capturing uncertainties in regional validation experiments• Introduced a new metric to evaluate the profit-ride trade-off for the pricing business decision, incorporating risk adjustment through the newly developed credible interval feature

• Developed a novel variable selection method using grouped pliable lasso for a Cox model in survival analysis, which improved the identification of brain regions associated with critical impairments of neurodegenerative diseases

• Developed a new framework applying a resampling method for multi-collinearity diagnostics and visualization called “mcvis” (https://cran.r-project.org/web/packages/mcvis/index.html), which can identify the sources of collinearity in complex data, particularly where traditional methods like VIF may not perform adequately