Table 3.

Drug repositioning methods integrated into DrugRepoBank

Methods class	Specific method	Detail	Ref
Similarity-based methods (drug–drug similarity prediction)	Chemical structure similarity	Chemical structure similarity is estimated with atom pairs using the Tanimoto coefficient, which is defined as the proportion of atom pairs shared among two compounds divided by their union.	(40)
	Target protein sequence-based similarity	Pairwise protein sequence comparison is performed using the standard Needleman-Wunsch dynamic programming algorithm for global alignment, and the percentage of pairwise sequence identity is reported as the corresponding sequence similarity	(41)
	Target Protein functional similarity [GO Cellular Component (CC)]	Each drug was annotated with enriched GO Cellular Component (CC) terms and the functional similarity between any two drugs is determined by the semantic similarity of their associated GO terms using the topology of the GO graph structure	(42)
	Target Protein functional similarity [GO Molecular Function (MF)]	Each drug was annotated with enriched GO Molecular Function (MF) terms and the functional similarity between any two drugs is determined by the semantic similarity of their associated GO terms using the topology of the GO graph structure	(42)
	Target Protein functional similarity [GO Biological Process (BP)]	Each drug was annotated with enriched GO Biological Process (BP) terms and the functional similarity between any two drugs is determined by the semantic similarity of their associated GO terms using the topology of the GO graph structure	(42)
	Drug-induced pathway similarity	Pairwise similarity between any two pathways was estimated based on the similarity of their constituent genes using dice similarity	(43)
Similarity-based method	Target-target similarity	Pairwise target protein sequences are compared based on the Needleman-Wunsch algorithm, which is designed based on dynamic programming	(41)
Artificial-intelligence-based methods	CPI_Prediciton	CPI_Prediciton is a CPI prediction approach by combining a graph neural network (GNN) for compounds and a convolutional neural network (CNN) for proteins	(44)
	TransformerCPI	TransformerCPI is a sequence-based deep learning method with a self-attention mechanism for compound-protein interaction prediction	(45)
	CapBM-DTI	CapBM-DTI is a drug-target interaction prediction method with capsule network and transfer learning	(46)
Signature-based methods	GSEAweight0	GSEAweight0 is derived from the KS-like statistic with weighted KS enrichment statistic (ES): p = 0	(47)
	GSEAweight1	GSEAweight1 is derived from the KS-like statistic with weighted KS enrichment statistic (ES): p = 1	(47)
	GSEAweight2	GSEAweight2 is derived from the KS-like statistic with weighted KS enrichment statistic (ES): p = 2	(47)
	KS	KS is derived from the KS-like statistic with the rank of fold changes as weight	(6)
	XSum	The XSum method was focused on the top genes ranked by fold changes of gene expression	(48)
	ZhangScore	The rank-based weights are set to all genes in one gene signature in ZhangScore	(49)
Network-based method	DRviaSPCN	DRviaSPCN is an approach to prioritize cancer candidate drugs by considering drug-induced subpathways and their crosstalk effects	(51)
	DrugSim2DR	DrugSim2DR is a tool that systematically predicts drug functional similarities within the context of specific diseases to facilitate drug repurposing	(52)