Statistical Challenges and Opportunities in Personalized Medicine
Personalized medicine is the future of healthcare, yet conventional statistical methodologies remain largely unavailable for single-subject transcriptome analysis due to the ``single-observation'' challenge. To this end, we develop and study statistical learning approaches and large-scale inferences to learn most information from single-subject transcriptome data and to identify differentially expressed genes (DEG) between two transcriptomes for each individual. I will present my research works in single-subject transcriptome analytics, including three projects with different focuses. The first project describes a two-step approach to identify differentially expressed pathways by employing a non-parametric clustering technique, k-means, followed by Fisher's exact tests. The second project proposes a novel variance stabilizing framework to transform raw gene counts before identifying enriched gene sets, and the transformation strategically by-passes the challenge of variance estimation in single-subject transcriptome analysis. In the third project, we proposed a new type of biomarkers and a solution to identify them. I will discuss statistical methods and computational algorithms for all the three projects, as well as their real-data applications to personalized treatment.