GIMIC - Smoothed Graph-Image representationof Microbiome Samples induce an optimal distance

Abstract An essential tool in microbiome is the distance between samples. This serves to detect groups of similar samples. However, current methods lack interpretability. We have recently shown that combining the cladogram with the abundance of each microbe into a sorted normalized abundance tree can improve the accuracy of Machine Learning and Differential Analysis. We show that smoothing these tree-based images to produce GIMIC (Smoothed Graph IMages of the MICrobiome) leads to an interpretable visualization of microbiome sets and that the difference between the smoothed tree-based images is a better metric than the current state-of-the-art in a wide array of tasks.