TABLE 2.

Method comparisons based on Stability Index and prediction metric (MSE/AUC) in experimental microbiome data sets with continuous or binary outcomes (methods ordered in terms of best MSE/Stability performance, followed with raw MSE/AUC/Stability values in parentheses)

Dataset (Continuous)	$n\ast p$ $(p/n)$	MSE (lower is better)	Stability (higher is better)
BMI Gut	98 * 87 (0.9)	Random forests (4.99) Compositional lasso (21.59) Lasso (24.07) Elastic Net (25.33)	Elastic Net (0.23) Compositional lasso (0.22) Lasso (0.14) Random forests (0.02)
pH Soil	89 * 2183 (24.5)	Elastic Net (0.23) Random forests (0.26) Lasso (0.34) Compositional lasso (0.46)	Compositional lasso (0.39) Lasso (0.31) Elastic Net (0.16) Random forests (0.04)

Dataset (Binary)	$n\ast p$ $(p/n)$	AUC (higher is better)	Stability (higher is better)
BMI Gut	98 * 87 (0.9)	Random forests (1.00) Compositional lasso (0.85) Elastic Net (0.78) Lasso (0.63)	Compositional lasso (0.29) Elastic Net (0.19) Lasso (0.14) Random forests (0.01)
pH Soil	89 * 2183 (24.5)	Random forests (1.00) Compositional lasso (0.96) Elastic Net (0.94) Lasso (0.90)	Compositional lasso (0.46) Elastic Net (0.32) Lasso (0.28) Random forests (0.03)