An overarching theme to my research is studying complex systems in high dimensional settings, but the specific projects are quite diverse. I approach all of these problems by first considering the associated data generating mechanism and relating it to a graphical Markov model. My research typically involves strong statistical computing skills. Specific applications include:
- Modelling plant-pollinator networks. We have developed a grouped Dirichlet-multinomial regression model with regularization for model selection, and adapted latent Dirichlet allocation to the plant-pollinator network setting. Future work involves accommodating sampling weights, zero-inflation, and the development of statistical to compare networks in space and/or time.
- Risk factors for disease in a longitudinal setting. Causal models can be used to provide unbiased parameter estimates in the presence of time-dependent confounding. I have applied marginal structural models to natural history data and developed a Cox score bootstrap for fast, reliable bootstrapping to be used in large scale simulation studies. I am interested in studying time-dependent weighting within the context of cure rate models and gee models for correlated outcomes.
- Model selection/Structure learning/Machine learning. I am interested in several applied problems for the genetic selection of livestock. We have developed models that exploit the graphical structure of predictors for the prediction of an outcome, whether continuous or binary, using regularization. A specific application involves the selection of candidate genes in the prediction of boar taint. We have also analyzed milk spectral data for the purpose of improving the nutraceutical properties of milk and overall health of cows. This project also involves developing methods to link the spectral data with genetic, physical and performance traits in a longitudinal setting.
I am also involved with the Bioinformatics program and can supervise students on research projects.