Extended Data Figure 5: The training sample size needed to reach a great predictive performance scales linearly with the number of species in synthetic data for MelonnPan, MiMeNet, and mNODE.

Synthetic data in this figure are generated by the microbial consumer-resource model with nutrient sampling probability $p_{\mathrm{n}}=1.0$. For the case with 100 species and varying number of metabolites (100, 200, or 300), three metrics are used for comparing model performances: the mean SCC $\bar{\rho }$, the top-50 mean SCC ${\bar{\rho }}_{50}$, and the number of metabolites with SCCs larger than 0.8 divided by the number of metabolites $N_{\rho >0.8}/N_{\mathrm{m}}$.