Table 2. Toxicity Annotation Model Statistics^a.

Toxicity class	Total number of positively annotated tox labelsb	Number of tox labels in train/test modelc	Number of drug forms in train/test modeld	Loss	Precision	Sensitivity	F1-Score
Carcinogenicity	287	154	75	0.576	1	1	1
Cardiotoxicity	482	316	119	4.584	1	1	1
Dermatological toxicity	77	46	27	0.011	1	1	1
Gastrotoxicity	317	208	65	0.881	0.99	0.99	0.99
Hematological toxicity	195	136	61	0.954	1	1	1
Hepatotoxicity	301	209	78	0.977	0.99	0.99	0.99
Immune system toxicity	136	98	53	7.137	0.98	0.98	0.98
Infections	97	71	39	0.973	0.99	0.99	0.99
(Energy) Metabolism toxicitye	130	88	33	0.016	1	1	1
Misusef	383	256	132	3.599	0.99	0.99	0.99
Musculoskeletal toxicity	85	60	14	0.038	1	1	1
Nephrotoxicity	79	58	24	0.874	0.99	0.99	0.99
Neurotoxicity	394	252	94	0.888	0.99	0.99	0.99
Psychiatric toxicityf	274	157	62	0.8	1	1	1
Respiratory toxicity	291	192	95	2.799	0.98	0.98	0.98
Teratogenicity	357	246	78	0.898	1	1	1
Vascular toxicity	236	151	86	0.701	0.98	0.98	0.98

^aThe performance statistics are given for the trained model; see definitions in Abbreviations.

^bThe total number of positively annotated toxicity labels for each toxicity class. These labels were then divided into a set of training and testing labels and a validation set of labels.

^cThe number of positively annotated toxicity labels used in the model training and testing. At least one positively annotated toxicity label per active ingredient was selected for the model training and testing, since this represents the broadest diversity of boxed warning text descriptions.

^dThe number of drug forms (i.e., active ingredients) for positively annotated toxicity labels used in the model training and testing. Examples of diseases for each toxicity class are given in Table 1.

^eThe class of “metabolism toxicity” is toxicity due to energy metabolism processes.

^fThe class of “misuse” includes accidental poisoning, drug misuse, and overdose but not drug dependence or suicide attempt that is classed as “psychiatric toxicity” (see Table 1).