Hire John C. - Data Scientist

Finding the rare and important amid the noise

For the last 3 years I have used the probabilistic programming languages JAGS and Stan to develop regression models for analyzing compositional data from microbial DNA. These data have multiple challenges: multivariate non-linear responses, high variability, sum to one constraint, and spatial and temporal auto-correlation. To overcome these challenges, I modeled the data using the Dirichlet-multionmial distribution as the base for an extensible model. The regression parameters informed the Dirichlet distribution, which allowed for greater variance than the multinomial alone. To test the models, I simulated data to evaluate what truth the model could recover and when it broke down. This research was part of a broad research program with over 15 graduate students. I provided stastical training and consulting as well as generalizable code for them to analyze their data.