Selected Research Projects:

Deep Neural Networks for Predicting Quantitative Structure-Activity Relationships in Drug Discovery

Implemented multi-task deep neural nets (DNN), which predict activities of compounds from multiple assays simultaneously. Developed a novel multi-task DNN approach that automatically build assistant data sets from candidate companion data sets, which efficiently borrows informative knowledge across multiple data sets and boost the prediction accuracy.

Dynamic Connectivity Detection Algorithm for Change-Point Detection in Brain Imaging Time Series

Developed a data-driven change points detection algorithm for partitioning the fMRI time course into distinct temporal intervals, each with different functional connectivity patterns between nodes. Proposed a novel sparse covariance matrix estimation method and significantly speed up the program compared to alternative algorithms, such as the dynamic connectivity regression technique.

Multilevel Binary Principal Component Analysis with Application to High-Dimensional Data

Developed a multilevel probabilistic PCA model to analyze high-dimensional binary data with replicate measurements. Proposed a nested variational EM approach for model fitting, which is computationally efficient and scalable for large data sets.

Developing N-leaping Algorithm for Accelerating Stochastic Simulation of Coupled Chemical Reactions

Developed a novel approximate stochastic simulation algorithm 'N-leaping', which shows favorable high efficiency and accuracy compared to the most popular approximate algorithms, such as the tau-leaping method.