Table 2.

The descriptions of some computational methods in drug combination prediction.

Methods	Algorithms	URL	Characteristics	Reference
Traditional machine learning	Support vector machine	—	Do well in identifying subtle patterns in complex data sets; poor interpretability; run slowly on large data sets.	[58]
	Decision tree	https://github.com/Lianlian-Wu/ForSyn (accessed on 18 July 2023)	Display visually; easy to over fit; accuracy may decrease when processing data with complex relationships.	[59]
	Gradient boosting	—	Do well in handling nonlinear relationships and high dimensional data; easy to over fit; hyperparameters tuning is complex.	[60]
Deep learning methods	Feedforward neural network	—	Do well in handling nonlinear relationships and high dimensional data; easy to over fit; poor interpretability; processing large data takes a long time	[61]
	Autoencoder	https://github.com/qiaoliuhub/drug_combination (accessed on 19 July 2023)	Feature learning ability is strong; poor interpretability.	[62]
	Graph convolutional network	https://github.com/Sinwang404/DeepDDS/tree/master (accessed on 19 July 2023)	Being able to capture the relationship and topological information between the nodes in the graph, poor interpretability, and robustness is of concern.	[63]
	Deep belief network		Perform well in supervised study; easy to over fit; poor interpretability.	[64]
Mathematical methods	Network analysis	—	Be able to capture complex interactions; good interpretability.	[65]
	Dynamic mathematical model	—	Be able to simulate drug reaction more accurately; poor interpretability.	[15]
Search algorithms	Breadth first search algorithm	—	Be able to consider a large amount of potential drug-target interactions; robustness is of concern.	[66]
Systems biology methods	Signature-based model	https://tanlab.ucdenver.edu/kMap(accessed on 20 July 2023)	Understand drug action mechanisms and influencing factors more comprehensively; high requirements on data quality.	[67]