Hierarchical Ensemble Methods for Protein Function Prediction

Table 1

Characteristics of some of the main hierarchical ensemble methods for AFP.

Methods	References	Multipath	Partial path	Class structure	Cost sens.	Sel neg.	Base learn	Improves on flat	Node process
HMC-LMLP	[124, 125]	√	√	TREE			any	√	TD
HTD-CS	[27]	√	√	TREE	√		any	√	TD
HTD-MULTI	[127]		√	TREE			any	√	TD
HTD-PERLEV	[128]			TREE			spec	√	TD
HTD-NET	[26]	√	√	DAG			any	√	TD
BAYES NET-ENS	[20]	√	√	DAG		√	spec	√	TD & BUP
HIER-MB and BFS	[30]	√	√	DAG			any	√	TD & BUP
HBAYES	[92, 135]	√	√	TREE		√	any	√	TD & BUP
HBAYES-CS	[123]	√	√	TREE	√	√	any	√	TD & BUP
Reconc-heuristic	[24]	√	√	DAG			any	√	TD
Cascaded log	[24]	√	√	DAG			any	√	TD
Projection-based	[24]	√	√	DAG			any	√	TD & BUP
TPR	[31, 142]	√	√	TREE		√	any	√	TD & BUP
TPR-W	[31]	√	√	TREE	√	√	any	√	TD & BUP
TPR-W weighted	[145]	√	√	TREE	√		any	√	TD & BUP
Decision-tree-ens	[29]	√	√	DAG			spec	√	TD & BUP