Table S4. Comparison between PHENOstruct vs. other methods only leaf terms.

Subontology	Method	F-max	Precision	Recall	mac-AUC
Organ	PhenoPPIOrth	0.03	0.05	0.03	0.51
	Struct->Dis->HPO	0.01	0.01	0.02	0.50
	Binary SVMs	0.20	0.19	0.22	0.66
	Clus-HMC-Ens	0.08	0.04	0.31	0.64
	PHENOstruct	0.30	0.21	0.50	0.77
Inheritance	PhenoPPIOrth	0.11	0.15	0.09	0.55
	Struct->Dis->HPO	0.01	0.01	0.02	0.46
	Binary SVMs	0.69	0.62	0.78	0.71
	Clus-HMC-Ens	0.72	0.63	0.84	0.73
	PHENOstruct	0.74	0.68	0.82	0.74
Onset	PhenoPPIOrth	0.19	0.20	0.17	0.52
	Struct->Dis->HPO	0.03	0.02	0.14	0.49
	Binary SVMs	0.29	0.21	0.47	0.62
	Clus-HMC-Ens	0.28	0.21	0.43	0.58
	PHENOstruct	0.35	0.28	0.48	0.68

The performance is evaluated by using the exact annotations (i.e. only leaf terms) as ground truth. In other words, true-path rule is not applied. Performance is presented using macro AUC and protein-centric F-max, Precision and Recall as defined above.