Fig 1.
Jaccard Distance to Donor following FMT intervention, as assessed using qiime fmt cc.
This visualization allows for users to interactively investigate whether recipient microbiomes become more similar to donor microbiomes with treatment (Fig 1A), and investigate statistical significance in the corresponding table (Fig 1B). In this figure, FMT intervention was at timepoint 3. Since this is an autoFMT study, the ‘donor’ is the recipient before cancer treatment. This figure shows that spontaneous recovery from cancer treatment leads to some individuals recovering quickly, while others do not. The FMT intervention decreases the variability in distance to donor of all recipients. We see that their distance to donor increases at the last time point, but this is up to two years after the initial FMT, and may indicate the development of an “individualized microbiome.” Presented data is from Taur et al. (2018) [2]. An interactive version of this figure is available in the project documentation at https://q2-fmt.readthedocs.io.
Fig 2.
Schematic of PEDF permutation test.
To test if PEDF is capturing donor overlap caused by microbiome transfer versus general microbiome similarity that could be observed across any two microbiomes in the study, we developed the PEDF permutation test. The PEDF permutation test breaks connections between factual donor-recipient pairs and tests to see if the unrelated donor-recipient pairs are more similar than the related donor-recipient pairs (via FMT transplant). A) The combinations of actual donor-recipient pairs in this example. Note Donor 1 is a donor to two recipients, to illustrate how PEDF permutation handles repeated transfers from one donor. PEDF values would be calculated for each of these pairs, and these would be considered our factual PEDF values. B) To create a null (i.e., counterfactual) distribution to compare factual PEDF values against, donors are paired up with every recipient that they did not donate to, to form our counterfactual pairs. The counterfactual pairs are sampled 999 times, and PEDF is calculated for each counterfactual pair to provide a robust null distribution. If the recipients have significant donor similarity following the FMT, it is expected that our factual donor-recipient pairs will have higher PEDF than our counterfactual donor-recipient pairs. The fraction of counterfactual PEDF values that are greater than an individual factual donor-recipient pair PEDF value is the p-value representing the significance of the factual donor-recipient pair PEDF value. C) A distribution of factual donor-recipient pair PEDF p-values with no globally significant signal will be distributed uniformly. D) If factual donor-recipient PEDF p-values are clustered around significant values, this suggests global statistical significance. This figure is a schematic representation and all data is simulated. Created in BioRender.
Fig 3.
Schematic representation of possible engraftment assessments with q2-fmt.
A) How did the donated microbiome alter the recipient’s microbiome? This subfigure illustrates that the community-coalescence pipeline and PEDF action could be used to evaluate how the donated microbiome alters the recipient’s microbiome. With longitudinal data, as presented by our tutorial data, q2-fmt also allows for investigation of the resilience of those alterations. B) Which features are important in the engraftment? This subfigure illustrates that the detect-donor-features pipeline and PRDF action can both assess what features are important to the engraftment, although they investigate this in different ways. The detect-donor-features pipeline evaluates which features are over-represented in the donor/reference relative to the recipient (seen in orange diverging barplots), while PRDF identifies features that successfully engrafted in multiple recipients, suggesting microbes that might be more amenable to engrafting. C) Are recipient features maintained or lost after FMT intervention? This subfigure illustrates the usage of detect-donor-features and PPRF to assess recipient features that are over-represented in the recipient relative to the donor/reference (seen in blue diverging barplots), and to investigate what proportion of the recipient features persist after FMT engraftment. Presented data is from Taur et al. (2018) [2]. Interactive versions of all figures presented here are available in the project documentation at https://q2-fmt.readthedocs.io.