Assistant Professor, School of Engineering
Computer Science and Engineering
Current Research Interests:
Broadly, I am interested in both basic and applied research in algorithms and Bayesian (probabilistic) machine learning.
Specifically, my research aims to develop probabilistic machine learning models, combinatorial algorithms, and scalable inference methods to better understand high-dimensional data, particularly genomics and genetics data applied to complex disease.
Current Research Projects:
I’m working on many projects, here are some (in no particular order):
1) Drug Design: Given observations of adverse reactions (side effects) to pharmaceutical drugs, predict the adverse reactions to a newly synthesized drug. In collaboration with Pfizer.
2) Admixture models in biology: Developing mixture models for various applications in biology. For example, the X chromosome genes are differentially expressed based on X chromosome inactivation patterns. We hypothesize that these patterns can be associated with some X-linked diseases. Other problems in this area include genomic mosaicism and mixtures of cancer signatures.
3) Predicting the outcomes of motions submitted during trial proceedings. These methods typically combine natural language processing with supervised learning. In collaboration with the Law School.
4) A single gene can express many different proteins which have divergent function and can be related to disease development. The reconstruction and quantification of these different proteins is a difficult problem. We develop probabilistic machine learning methods for variations of this problem.
5) Given genomic sequencing for an individual, reconstruct their sequences of genetic variation. Very similar to “genome assembly” but a bit different. A mix between algorithms and machine learning.
There are some other projects as well on clustering of genomic sequences, trying to estimate risk of opioid use disorder from genetics and environmental factors, and text mining with biomedical abstracts, but these are less well formulated.