Table 2.

Performance comparison across models.

	All metabolites	Tyrosine and Proline	Tyrosine	Proline	All metabolites
	No tuning	No tuning	No tuning	No tuning	Parameter tuning
Positive predictive value	0.865	0.707	0.727	0.429	0.943
Negative predictive value	0.814	0.718	0.66	0.444	0.844
F1	0.831	0.716	0.657	0.400	0.853
Accuracy	0.837	0.712	0.687	0.437	0.887
AUC	0.897	0.753	0.732	0.509	0.927

We built five models with (1) all 173 metabolites, (2) only tyrosine and proline, (3) only tyrosine, and (4) only proline. Additionally, we show the results of the tuned parameters (tenfold CV).