Speaker: Kendrick Li, Graduate Student, UW Biostatistics
Abstract: Human bodies habitat numerous microbiome, including bacterial, fungi and virus. One goal in microbial research is to correctly infer the interaction network among microbiome community. This can be derived from co-occurrence pattern of the microbiomes. Microbial data are compositional and high-dimensional in nature, which brings inferential and computational challenges into this problem. Several methods, including SparCC and CCREPE , adjust for composition effect. These methods consider marginal correlation, while the marginal correlation of two species can be mediated by their correlation with a common third species. Methods are still elusive that adjust compositionality and infer conditional correlation, which can be interpreted as direct interactions. In 2016, Kurtz and colleagues developed SPIEC-EASI (SParse InversE Covariance Estimation for Ecological Association Inference) , a pipeline for inferring the microbial interaction network . SPIEC-EASI transforms the abundance data to adjust for composition effect, and then utilizes neighborhood selection and graphical lasso to learn the underlying correlation graph. The graphical model approach gives results about direct interaction, which is of more interest to researchers. Spiec-Easi applies StARs (Stability Approach to Regularization Selection) for model selection. StARs repeatedly draws subsamples from the original data and estimates a network with each subsample, and chooses the tuning parameter that leads to the most “similar” graphs estimated from the subsamples. To set up a benchmark for comparing different methods, Spiec-Easi accompanies a data generator, which simulates microbial synthetic data from a real dataset. With the synthetic data based on a dataset from American Gut Project, Spiec-Easi outperforms the state-of-art methods.  Faust, Karoline, et al. “Microbial co-occurrence relationships in the human microbiome.” PLoS computational biology 8.7 (2012): e1002606.  Kurtz, Zachary D., et al. “Sparse and compositionally robust inference of microbial ecological networks.” PLoS computational biology 11.5 (2015): e1004226.