Bayesian Inference

Quantifying the dynamics of Kaposi's sarcoma-associated herpesvirus persistence

In collaboration with the Kaye lab at Brigham and Women's Hospital, we are quantifying the dynamics of latent Kaposi's sarcoma-associated herpesvirus (KSHV) persistence. We developed a mathematical model and a statistical inference framework to infer viral dynamics from fluorescence microscopy images of cells in culture. Forward simulations were used to understand decades-long viral persistence and evaluate latent KSHV replication as a potential therapeutic target to disrupt KSHV-dependent tumor growth.

The impact of measurement noise in mediation analysis

As a post-bacc at the Jackson Laboratory, my main research focus has been mediation analysis. A primary objective of mediation analysis is to infer the relationship among three variables, and it is becoming increasingly common to use it with multi-omics data to understand causal pathways underlying a phenotype. Mediation analysis is often done without distinguishing variation due to causal relationships from variation due to measurement noise, which can have a profound effect on inferences. In this analysis, we address the impact of applying a standard mediation analysis to data as if it is measured without error and identify ways to diagnose the accuracy of results from real data.

A Bayesian model selection approach to mediation analysis

In collaboration with the Valdar lab at UNC, we developed a Bayesian model selection approach to mediation analysis implemented in the bmediatR R package. This approach allows for flexibility in both data inputs and potential inferences and uses conjugate priors to increase efficiency. I am currently extending the framework to allow for the inference of moderated mediation.

A Bayesian model selection approach to mediation analysis

In collaboration with the Valdar lab at UNC, we developed a Bayesian model selection approach to mediation analysis implemented in the bmediatR R package. This approach allows for flexibility in both data inputs and potential inferences and uses conjugate priors to increase efficiency. I am currently extending the framework to allow for the inference of moderated mediation.