Extended Data Fig. 4: MDSINE2 outperforms state-of-the-art methods on a 10 taxa synthetic benchmarking standard.
From: Learning ecosystem-scale dynamics from microbiome data with MDSINE2
(a) Underlying dynamical systems network topology16, which was used to simulate data under a gLV model (n = 10 seeds). Spearman correlation between model predictions for (b) growth rate strengths and (c) interactions strengths (higher is better, 1 is maximum) (d) Area under the receiver operating curve (AUC-ROC) for interaction presence/absence (higher is better, 1 is maximum). (e) RMSE of log abundance for model forecast from initial conditions compared to ground truth trajectory (lower is better). Pairwise comparisons performed with BH corrected two-sided Wilcoxon signed-rank test * p < 0.05, ** p < 0.01. Boxes denote interquartile region with a line for the median. Whiskers denote 95% interval. p values provided in Supplementary Data Table.
