Skip to main content
Fig. 2 | Genome Biology

Fig. 2

From: MITRE: inferring features from microbiota time-series data linked to host status

Fig. 2

Cross-validated predictive performance of MITRE and comparator methods on semi-synthetic and real data. ad Results on semi-synthetic data. A parametric bootstrapping-type method was used to generate simulated data from an underlying real dataset. Simulated cases were generated by randomly selecting and perturbing bacterial clades over a randomly selected limited time window (~ 20% of the duration of the study); an equal number of control subjects were simulated. For the one-clade perturbation scenarios, the clade remained unperturbed for the simulated cases; for the two-clade perturbation scenarios, one clade was perturbed in the simulated control subjects, and both were perturbed in the simulated cases. a, b One or two clades randomly perturbed in simulated subjects, 18 time points, varying numbers of subjects. c, d One or two clades randomly perturbed in simulated subjects, 32 subjects, varying numbers of time points. e Results on real data. The different methods were used to predict the indicated categories in the datasets shown. F1-score is the harmonic mean of precision and recall; higher scores indicate superior results

Back to article page