VOLARE: Visual analysis of disease-associated microbiome-immune system interplay

Background: Associations between the microbiome and the immune system are well documented. However, elucidating specific mechanisms is an active research area. High-dimensional "omic" assays such as 16S ribosomal RNA (rRNA) sequencing and CyTOF immunophenotyping support hypotheses generation, identifying possible interactions that can be further explored experimentally. Linear regression between microbial and host immune features is useful for quantifying relationships between mi-crobes and immune readouts. But, vetting dozens of significant associations can be cumbersome, especially when a project involves experts from different disciplines. In order to facilitate communication and sense-making across disciplines, we performed a design study for visual analysis of these relationships with a goal of helping researchers prioritize results for experimental follow-up. Results: Using data from paired 16S ribosomal RNA (rRNA) sequencing and CyTOF immunophenotyping on gut biopsy samples from people with and without HIV, we fit a regression model to each microbe:immune cell pair, accounting for differences by disease status. We used permutation testing to control the false discovery rate, resulting in a top table of microbe:immune cell pairs. After identifying essential tasks in the further analysis of this top table, we designed VOLARE (Visualization Of LinEar Re-gression Elements), a web application that integrates a searchable top table, a network summarizing this table, sparkline-inspired graphs of fitted regression models, and detailed regression plots showing sample-level detail. We applied this application to two case studies--microbiome:immune cell data from gut biopsies and microbiome:cytokine data from fecal samples. Conclusions: Systematically integrating microbiome-immune system data through linear regressions and presenting the top table results in an interactive environment supports the Shneiderman mantra ("Overview first, filter, details-on-demand"). Our approach allows domain experts to control the analysis of their results, empowering them to screen dozens of candidate relationships with ease. Our contributions include characterizing the exploration of microbiome-immune system data in a team science context, and the support of an associated workflow by integrating existing visualization approaches. Availability: R scripts, the web application, and sample data are available at https://sourceforge.net/projects/cytomelodics