Table 1.

Performance of machine learning and deep learning models using only sequence knowledge compared to graph-based models fusing structure knowledge.

Category	Model	Feature Representation	MAE	RMSE	PCC	Spearman	CI	R 2
Machine learning models	RR	label encoding	1.471	1.864	0.515	0.521	0.685	0.261
	KNN	label encoding	1.377	1.732	0.604	0.565	0.703	0.362
	DT	label encoding	1.409	1.882	0.590	0.577	0.707	0.247
	SVR	label encoding	1.749	2.150	0.156	0.231	0.527	0.017
	RF	label encoding	1.221	1.527	0.726	0.713	0.760	0.504
Sequence-based deep learning models	1D-CNN	label encoding	1.569	1.938	0.535	0.531	0.683	0.201
		one-hot encoding	1.159	1.473	0.752	0.741	0.776	0.539
	RNN	label encoding	1.815	2.227	0.175	0.329	0.611	−0.054
		one-hot encoding	1.669	2.045	0.459	0.460	0.657	0.111
Structure-based deep learning models	GraphSAGE	label encoding	1.288	1.599	0.675	0.664	0.739	0.449
		one-hot encoding	1.127	1.417	0.758	0.745	0.778	0.573