Fig. 6

Association between kNN efficacy and local samples’ density in post-perturbation experiments. a Principal coordination analysis of the Jaccard distances between 42 samples from 21 subjects, before (hollow circles) and after antibiotic administration (filled circles). Dashed lines are drawn between samples from the same subject. b The kNN gain \(\Delta\) versus the mean distance of the closest three neighbor samples \(\langle D_3 \rangle\) calculated for 21 post-antibiotic microbial samples. Samples in denser regions in the assemblage space tend to have better predictions of the kNN method compared with samples in sparse regions. The orange straight line represents the linear regression model, represented by the equation \(Y = -0.5X + 0.25\), with p-value = 0.02. The light-blue curves indicate the 95% confidence envelop around the regression line