HPOAnnotator: improving large-scale prediction of HPO annotations by low-rank approximation with HPO semantic similarities and multiple PPI networks

Table 12 Performance results on Data-201706 focusing on the sub-ontology Organ abnormality

Method	AUC	AUPR	micro-AUC	micro-AUPR	macro-AUC	macro-AUPR	leaf-AUC	leaf-AUPR
NMF-Organ	0.955	0.507	0.883	0.250	0.745	0.127	0.682	0.077
NMF-PPN-Organ	0.962	0.555	0.889	0.276	0.755	0.144	0.701	0.091
NMF-NHPO-Organ	0.962	0.535	0.888	0.264	0.756	0.141	0.702	0.089
NMF-All	0.956	0.512	0.884	0.258	0.755	0.129	0.685	0.083
NMF-PPN-All	0.962	0.553	0.889	0.273	0.755	0.143	0.698	0.089
NMF-NHPO-All	0.962	0.556	0.889	0.274	0.755	0.144	0.699	0.090
HPOAnnotator-All	0.963	0.559	0.891	0.278	0.759	0.146	0.702	0.094

The first three rows of methods with “Organ” are trained by HPO terms on Organ abnormality, while the others with “All” are trained by considering all sub-ontologies.
Method performs best in terms of this evaluation metric are in boldface

ISSN: 1755-8794