Research

My research interests include

  1. causal mediation analysis of big data problem;

  2. (generalized) regression for covariance matrix outcomes.

Mediation analysis with high-dimensional mediators

Causal diagram with multiple causally dependent mediators.
Causal diagram of proposed reduced model relaxing the ordering assumption in the mediators.
Identified brain pathways using Pathway Lasso in a classification learning task fMRI study with reaction time as the outcome of interest.


Granger mediation analysis for multiple time series

  • Mediation model with correlated errors for unmeasured confounding.

  • Integrate Granger causality to account for spatio-temporal dependency.


Functional mediation analysis

  • Treatment, mediator and outcome are all functional data.

  • Study dynamics in the causal effects.


Multiview Data Integration

  • Integrate data collected from two sources following a biological mechanistic assumption.

  • Treat data from different imaging modalities as two blocks of high-dimensional mediators.


Regression for covariance matrix outcomes

Two-dimensional data contour plot. Covariance matrices vary as continuous X varies.

  • Integrate principal component analysis principle with generalized linear model.

  • In the projection space, data variation depends on covariates through a log-linear model.

  • Identify linear projection and model coefficient simultaneously.

  • Application in resting-state fMRI study.

Brain regions with high positive loadings.
Brain regions with high negative loadings.

  • Brain subnetwork (component) identified in a resting-state fMRI study shows significant treatment effect in semantic aphasia patients.