At this month’s journal club, James, an undergraduate student in Jane Foster’s laboratory at McMaster University, led the discussion on how we can better associate microbiome data with clinical data on health and disease. Beginning with a brief introduction to PERMANOVA and GLM testing, we focused on a new regression-based testing method that was described in “An adaptive association test for microbiome data” by Wu et al. (2016), called aMiSPU, or the adaptive microbiome-based sum of powered score.
After an extensive buildup to understand the mathematics behind the method that involved several powerpoint slides, many questions, and a lot of eureka moments, the group was satisfied with much of the paper’s claims. The aMiSPU method provided better power in many cases, especially at detecting differences at low percentages.
Importantly, however, the paper acknowledged that there was no clear cut best method for all data, as there were situations where the aMiSPU was not the optimal test. This highlighted the importance of understanding the central questions being asked before analyzing microbiome data.
Of particular interest to the group was the authors’ analysis of simulated data to compare and contrast several methodologies, including DESeq2 and Kruskal-Wallis. We were surprised by the number of false positives that were created by each testing method, including the aMiSPU (although it did perform much better at detecting true positives and reducing false positives than other tests). An important area for future exploration could be the repetition of this experiment with new simulated data and across a wider span of published tests.
The paper communicated a new approach to detecting differences in the microbiome between clinical metadata and while we were unsure if it would be applicable to all datasets, we did find it an intriguing approach worthy of exploration in our own analyses.