A comparison of embedding aggregation strategies in drug–target interaction prediction

Iliadis, Dimitrios; De Baets, Bernard; Pahikkala, Tapio; Waegeman, Willem

doi:10.1186/s12859-024-05684-y

BMC Bioinformatics

Table 3 Performance results obtained for every possible combination of two datasets (KIBA, DAVIS), three prediction settings (random, cold-drug, cold-target), three embedding aggregation strategies and three compound-target branch pairs (MLP–MLP, CNN–CNN, MPNN–CNN)

From: A comparison of embedding aggregation strategies in drug–target interaction prediction

Dataset	Pred setting	Aggregation	MLP–MLP				CNN–CNN				MPNN–CNN
Dataset	Pred setting	Aggregation	MSE	R\(^2\)	CI	Params	MSE	R\(^2\)	CI	Params	MSE	R\(^2\)	CI	Params
DAVIS	Random	Dot prod.	0.2656	0.6733	0.8702	2.1	0.2622	0.6776	0.8727	0.6	0.5561	0.3161	0.7687	0.2
		Tensor prod.	0.2828	0.6522	0.8622	6.7	0.2708	0.6669	0.8699	0.4	0.5537	0.3190	0.7720	0.4
		MLP	0.2722	0.6652	0.8759	2.4	0.2489	0.6938	0.8791	0.8	0.2647	0.6745	0.8722	1.9
	Cold	Dot prod.	0.6189	− 0.0426	0.7304	6.3	0.6593	− 0.1107	0.6657	0.1	0.5112	0.1387	0.7287	0.2
	Drug	Tensor prod.	0.6828	− 0.1502	0.6929	1.4	0.6807	− 0.1467	0.6142	0.3	0.5125	0.1366	0.7326	0.4
		MLP	0.6764	− 0.1394	0.6949	4.9	0.6464	− 0.0889	0.6484	0.5	0.5123	0.1370	0.7385	3.3
	Cold	Dot prod.	0.5895	0.3369	0.7974	1.1	0.4839	0.4556	0.8137	0.6	0.6783	0.2369	0.7294	0.1
	Target	Tensor prod.	0.5750	0.3532	0.8077	1.6	0.4803	0.4596	0.8069	0.5	0.6653	0.2515	0.7471	0.1
	Target	MLP	0.5966	0.3288	0.8023	2.0	0.4905	0.4482	0.8092	1.7	0.5626	0.3671	0.8014	0.9
KIBA	Random	Dot prod.	0.1994	0.7187	0.8379	2.1	0.2169	0.6940	0.8329	0.6	0.5490	0.2456	0.7154	0.2
		Tensor prod.	0.2092	0.7050	0.8408	4.4	0.2222	0.6865	0.8281	0.4	0.5534	0.2194	0.6996	0.3
		MLP	0.2093	0.7048	0.8496	2.6	0.1953	0.7245	0.8565	1.2	0.2544	0.6412	0.8190	2.3
	Cold Drug	Dot prod.	0.4167	0.4498	0.7510	9.9	0.4730	0.3756	0.7274	0.5	0.5930	0.2171	0.6985	0.4
		Tensor prod.	0.4345	0.4264	0.7460	7.8	0.4779	0.3691	0.7310	0.4	0.5965	0.2124	0.6978	0.3
		MLP	0.4258	0.4378	0.7538	3.8	0.4802	0.3661	0.7348	1.1	0.5172	0.3172	0.7211	2.6
	Cold Target	Dot prod.	0.4197	0.3666	0.7069	1.3	0.3839	0.4206	0.7281	0.7	0.6328	0.0450	0.6023	0.3
		Tensor prod.	0.4126	0.3773	0.7142	1.7	0.3904	0.4108	0.7192	0.5	0.6326	0.0453	0.6040	0.4
		MLP	0.3856	0.4181	0.7203	1.8	0.3795	0.4272	0.7308	0.5	0.4172	0.3704	0.7168	2.6

For every combination, the average MSE, R\(^2\), CI and number of trainable parameters of the top-5 performing (lowest overall loss) configurations are reported

Back to article page

ISSN: 1471-2105

Contact us

General enquiries: journalsubmissions@springernature.com