Gene expression knowledge graph for patient representation and diabetes prediction

Sousa, Rita T.; Paulheim, Heiko

doi:10.1186/s13326-025-00325-6

Journal of Biomedical Semantics

Table 4 Mean and standard deviation for accuracy, precision, recall, f1-score, and weighted average f1-score (Acc, Pr, Re, F1, WAF), comparing several approaches. These include our methodology coupled with different embedding methods and different representation strategies, a variant employing a GNN, the gene expression baselines (GE when only the GSE123658 dataset is used and GE+ if the three datasets are used), and the omics integration baselines (SIMBA and MultiMAP). The bold value indicates the highest performance within each embedding method, while the italic value highlights the highest performance for each metric

From: Gene expression knowledge graph for patient representation and diabetes prediction

Metric	Embedding method		Composite gene embeddings			Direct patient embeddings
	Type	Model	not norm	pat-norm	gene-norm	Binning approach			Patient-gene links approach
	Type	Model	not norm	pat-norm	gene-norm	not norm	pat-norm	gene-norm	pat avg	gene avg
Acc	Walk-based	RDF2vec	0.695 ± 0.041	0.695 ± 0.041	0.707 ± 0.050	0.487 ± 0.238	0.564 ± 0.152	0.561 ± 0.119	0.879 ± 0.118	0.854 ± 0.060
	Semantic-based	ComplEx	0.671 ± 0.090	0.671 ± 0.090	0.755 ± 0.090	0.510 ± 0.098	0.524 ± 0.046	0.510 ± 0.125	0.633 ± 0.102	0.512 ± 0.115
		distMult	0.696 ± 0.065	0.696 ± 0.065	0.780 ± 0.085	0.512 ± 0.100	0.574 ± 0.032	0.488 ± 0.119	0.454 ± 0.101	0.512 ± 0.161
		HolE	0.647 ± 0.068	0.647 ± 0.068	0.721 ± 0.080	0.499 ± 0.052	0.598 ± 0.136	0.403 ± 0.065	0.621 ± 0.078	0.830 ± 0.042
	Translational	TransE	0.524 ± 0.023	0.499 ± 0.034	0.524 ± 0.023	0.512 ± 0.031	0.487 ± 0.031	0.524 ± 0.023	0.512 ± 0.129	0.649 ± 0.110
	Translational	TransR	0.524 ± 0.023	0.499 ± 0.034	0.499 ± 0.034	0.426 ± 0.083	0.440 ± 0.032	0.414 ± 0.075	0.706 ± 0.065	0.743 ± 0.047
	GNN								0.414 ± 0.096	0.478 ± 0.106
	GE		0.768 ± 0.120
	GE+		0.780 ± 0.083
	SIMBA		0.559 ± 0.090
	MultiMAP		0.535 ± 0.091
Pr	Walk-based	RDF2vec	0.772 ± 0.137	0.772 ± 0.137	0.718 ± 0.048	0.424 ± 0.318	0.558 ± 0.172	0.517 ± 0.170	0.923 ± 0.154	0.844 ± 0.051
	Semantic-based	ComplEx	0.743 ± 0.168	0.743 ± 0.168	0.756 ± 0.080	0.483 ± 0.094	0.519 ± 0.094	0.480 ± 0.112	0.654 ± 0.177	0.488 ± 0.174
		distMult	0.762 ± 0.148	0.762 ± 0.148	0.789 ± 0.088	0.482 ± 0.111	0.629 ± 0.189	0.448 ± 0.124	0.412 ± 0.138	0.519 ± 0.184
		HolE	0.717 ± 0.167	0.717 ± 0.167	0.710 ± 0.088	0.487 ± 0.053	0.566 ± 0.128	0.353 ± 0.086	0.606 ± 0.057	0.839 ± 0.036
	Translational	TransE	0.000 ± 0.000	0.087 ± 0.175	0.000 ± 0.000	0.094 ± 0.188	0.182 ± 0.223	0.000 ± 0.000	0.477 ± 0.110	0.797 ± 0.221
	Translational	TransR	0.000 ± 0.000	0.087 ± 0.175	0.087 ± 0.175	0.412 ± 0.083	0.372 ± 0.082	0.367 ± 0.085	0.746 ± 0.136	0.732 ± 0.050
	GNN								0.381 ± 0.109	0.439 ± 0.162
	GE		0.821 ± 0.141
	GE+		0.856 ± 0.084
	SIMBA		0.538 ± 0.124
	MultiMAP		0.447 ± 0.257
Re	Walk-based	RDF2vec	0.586 ± 0.158	0.586 ± 0.158	0.636 ± 0.118	0.371 ± 0.300	0.486 ± 0.140	0.432 ± 0.213	0.875 ± 0.158	0.843 ± 0.102
	Semantic-based	ComplEx	0.564 ± 0.097	0.564 ± 0.097	0.711 ± 0.169	0.536 ± 0.136	0.461 ± 0.123	0.536 ± 0.193	0.539 ± 0.146	0.404 ± 0.160
		distMult	0.614 ± 0.139	0.614 ± 0.139	0.743 ± 0.113	0.464 ± 0.176	0.461 ± 0.166	0.464 ± 0.193	0.386 ± 0.139	0.539 ± 0.094
		HolE	0.536 ± 0.111	0.536 ± 0.111	0.718 ± 0.093	0.518 ± 0.109	0.614 ± 0.210	0.357 ± 0.162	0.618 ± 0.172	0.793 ± 0.104
	Translational	TransE	0.000 ± 0.000	0.200 ± 0.400	0.000 ± 0.000	0.200 ± 0.400	0.400 ± 0.490	0.000 ± 0.000	0.521 ± 0.209	0.393 ± 0.170
	Translational	TransR	0.000 ± 0.000	0.200 ± 0.400	0.200 ± 0.400	0.411 ± 0.049	0.307 ± 0.125	0.382 ± 0.190	0.636 ± 0.118	0.739 ± 0.125
	GNN								0.382 ± 0.172	0.407 ± 0.211
	GE		0.671 ± 0.165
	GE+		0.671 ± 0.200
	SIMBA		0.432 ± 0.142
	MultiMAP		0.625 ± 0.371
F1	Walk-based	RDF2vec	0.639 ± 0.050	0.639 ± 0.050	0.667 ± 0.078	0.390 ± 0.296	0.516 ± 0.152	0.462 ± 0.199	0.878 ± 0.105	0.842 ± 0.072
	Semantic-based	ComplEx	0.625 ± 0.062	0.625 ± 0.062	0.724 ± 0.123	0.506 ± 0.110	0.470 ± 0.080	0.501 ± 0.142	0.579 ± 0.127	0.431 ± 0.154
		distMult	0.657 ± 0.047	0.657 ± 0.047	0.761 ± 0.087	0.464 ± 0.131	0.491 ± 0.077	0.450 ± 0.156	0.397 ± 0.137	0.521 ± 0.128
		HolE	0.590 ± 0.035	0.590 ± 0.035	0.710 ± 0.074	0.492 ± 0.039	0.582 ± 0.153	0.348 ± 0.119	0.598 ± 0.095	0.811 ± 0.062
	Translational	TransE	0.000 ± 0.000	0.122 ± 0.243	0.000 ± 0.000	0.128 ± 0.256	0.250 ± 0.306	0.000 ± 0.000	0.492 ± 0.154	0.504 ± 0.156
	Translational	TransR	0.000 ± 0.000	0.122 ± 0.243	0.122 ± 0.243	0.407 ± 0.055	0.331 ± 0.108	0.368 ± 0.126	0.671 ± 0.073	0.728 ± 0.063
	GNN								0.367 ± 0.134	0.410 ± 0.161
	GE		0.730 ± 0.135
	GE+		0.729 ± 0.130
	SIMBA		0.475 ± 0.135
	MultiMAP		0.489 ± 0.247
WAF	Walk-based	RDF2vec	0.683 ± 0.033	0.683 ± 0.033	0.702 ± 0.052	0.473 ± 0.246	0.559 ± 0.151	0.546 ± 0.130	0.870 ± 0.129	0.854 ± 0.060
	Semantic-based	ComplEx	0.662 ± 0.086	0.662 ± 0.086	0.751 ± 0.094	0.509 ± 0.098	0.515 ± 0.040	0.506 ± 0.122	0.625 ± 0.104	0.499 ± 0.114
		distMult	0.685 ± 0.063	0.685 ± 0.063	0.779 ± 0.085	0.506 ± 0.099	0.553 ± 0.014	0.483 ± 0.121	0.449 ± 0.105	0.505 ± 0.167
		HolE	0.636 ± 0.060	0.636 ± 0.060	0.719 ± 0.079	0.490 ± 0.060	0.594 ± 0.135	0.389 ± 0.064	0.613 ± 0.081	0.829 ± 0.043
	Translational	TransE	0.361 ± 0.027	0.333 ± 0.037	0.361 ± 0.027	0.348 ± 0.035	0.320 ± 0.034	0.361 ± 0.027	0.508 ± 0.132	0.619 ± 0.119
	Translational	TransR	0.361 ± 0.027	0.333 ± 0.037	0.333 ± 0.037	0.422 ± 0.083	0.425 ± 0.043	0.407 ± 0.070	0.698 ± 0.065	0.740 ± 0.047
	GNN								0.396 ± 0.091	0.461 ± 0.110
	GE		0.764 ± 0.121
	GE+		0.771 ± 0.093
	SIMBA		0.549 ± 0.096
	MultiMAP		0.430 ± 0.150

Back to article page

ISSN: 2041-1480

Contact us

General enquiries: journalsubmissions@springernature.com

You are viewing the site in preview mode

Journal of Biomedical Semantics

Contact us