Abstract:
|
We apply new panomic distance mediation analysis with feature selection to understand the effect of polyp infiltrating microbes on host immune response. Multivariate omnibus distance mediation analysis (MODIMA) allows for testing for mediation in multivariate exposure-mediator-response triplets. Our extension, MODIMA-fs, selects individual response and mediator features associated with the effect. We perform a simulation to confirm the statistical power of our method. We apply the method to data from 30 colorectal polyps with data on 594 immune-related gene expression from a NanoString panel and microbiome from 16S rRNA gene sequencing. The polyp descriptors including sex, race, age, colonic location, degree of dysplasia, and histology are expanded into a complete 6-way interaction design matrix. Only the species and genera with literature evidence for association with colorectal polyps and cancers are included. Euclidean distance MODIMA-fs suggests a significant mediation effect of 2 microbes on the association between polyp characteristics and the expression of 9 immune genes (P=0.01). This work illustrates a top-down approach to hypothesis generation analysis with panomic data.
|