Establishing a baseline for literature mining human genetic variants and their relationships to disease cohorts

Table 4 Relation extraction results over Variome corpus relations with at least 100 examples, based on 10-fold cross-validation and assuming gold-standard entity annotations

Relation	N	Baseline			PKDE4J
		P	R	F	P	R	F
Disease-has-ConceptIdeas	431	0.704	0.922	0.764	0.799	0.746	0.781
Disease-has-Physiology	188	0.752	0.873	0.788	0.885	0.684	0.806
Disease-has-Disorders	349	0.754	0.884	0.793	0.773	0.713	0.752
Disease-relatedTo-BodyPart	445	0.763	0.853	0.791	0.794	0.666	0.746
Mutation-relatedTo-Disease	126	0.702	0.986	0.777	0.683	0.973	0.758
Gene-has-Physiology	180	0.844	0.897	0.861	0.857	0.723	0.807
Gene-has-Mutation	538	0.835	0.866	0.845	0.910	0.569	0.758
Cohort-has-Mutation	307	0.873	0.921	0.888	0.909	0.726	0.839
Cohort-has-Disease	717	0.715	0.813	0.745	0.865	0.654	0.781
Cohort-has-Size	669	0.857	0.793	0.835	0.910	0.736	0.844
Cohort-has-Disorders	119	0.859	0.918	0.878	0.903	0.748	0.845

P = Precision, R = Recall, F = F0.5 score (weighting P more than R due to the importance of Precision). The {P,R}-Base results refer to results from a simple co-occurrence baseline. The best F score (F-Base or F-PKDE4J) is bolded

ISSN: 1472-6947