I am using Markov chain Monte Carlo techniques applied to unsupervised latent dirichlet allocation (LDA) methodologies to reduce the dimensionality of bioinformatic data. I use graph theoretic embeddings to visualize the results. I am developing modified LDA approaches to handle the peculiarities of this type of data to enhance convergence and repeatability of the stochastic computations.
For microbiome data, this approach allows me to identify sets of bacteria that may represent interacting ecosystems which are then related to disease phenotypes. I am exploring the possibility that pathogenicity may relate to bacterial interactions in the case of Alzheimer’s disease rather than single pathogens. I am also working with Crohn’s/inflammatory bowel disease and nasopharyngeal microbiome measurements.