Grant: Doctoral Research Award - PA - Bioinformatics and Computational Biology

There are two schools of statistics that are prevalent. The frequentist school and the Bayesian school. The frequentist approach relies on large sample properties and is ineffective when the sample size is small. The Bayesian approach relies on the choice of prior and is prone to high type-I error when the prior is wrongly chosen. Due to their limitation, it's often difficult of them to be used for prevention of a future pandemic or targeting treatment at the early phase of pathogenic emergence. In this situation, the sample sizes are small and no prior is available because the future doesn't resemble the past. The Dempster-Shafer approach circumvents this issue by enabling posterior inference without the need of a prior. Evaluating the Dempster-Shafer approach on complex medical genetic, proteomic, blood biochemistry, cardiovascular symptom datasets of hospitalized patients with sepsis and pneumonia will be the focus of this study. We evaluate our method in tasks such as gene-expression analysis, drug target discovery, novel pathogen detection, life support assignment, and causal inference.