Prediction performance of the pre-trained model, average performance and standard deviation of the individual fine-tuned models that comprise the ensemble, and performance of the ensemble, for comparable output sizes. The table indicates the percentage of drugs for which at least one, at least half and all reference metabolites have been correctly identified, as well as, the total number of identified metabolites.

Model	Output size	At least one metabolite (%)	At least half metabolites (%)	All metabolites (%)	Total identified metabolites	Precision (%)	Recall (%)
Pre-trained (beam 15)	9.1	39.3	27.4	13.1	49	6.4	22.6
Average (beam 15)	9.3 ± 0.4	78.8 ± 4.6	61.7 ± 5.7	33.1 ± 4.1	102.3 ± 8.0	13.1 ± 0.8	47.2 ± 3.7
Ensemble (beam 5)	10.2	90.5	77.4	42.9	125	14.5	57.6