AnnoPRO: a strategy for protein function annotation based on multi-scale protein representation and a hybrid deep learning of dual-path encoding

Table 5 A comparison among the predictive performances of AnnoPRO and three representative methods for the functional annotations of two well-known heat shock 70kDa proteins (HSPA1A, HSPA2)

Protein Name	Methods	BP		CC		MF
Protein Name	Methods	Recall	Precision	Recall	Precision	Recall	Precision
HSPA1A	DeepGOPlus	0.358	0.357	0.410	0.889	0.605	0.812
	PFmulDL	0.635	0.457	0.615	0.800	0.814	0.500
	NetGO3	0.286	0.876	0.634	0.605	0.809	0.884
	*AnnoPRO*	0.641	0.715	0.595	0.962	0.917	0.936
HSPA2	DeepGOPlus	0.375	0.284	0.394	0.867	0.765	0.765
	PFmulDL	0.344	0.386	0.419	0.812	0.788	0.605
	NetGO3	0.346	0.605	0.419	0.684	0.757	0.903
	*AnnoPRO*	0.470	0.851	0.594	0.670	0.868	0.943

Those values indicating the best performances among all methods were highlighted in BOLD, and AnnoPRO performed the best in the vast-majority (9/12) of the Gene Ontology (GO) classes (BP, CC, MF) under both evaluating criteria (recall, precision). All methods were ordered based on their publication dates. BP biological process, CC cellular component, MF molecular function, HSPA1A heat shock 70 kDa protein 1A, HSPA2 heat shock 70 kDa protein 2

ISSN: 1474-760X