Table 1.

Statistical comparison of the ability of best performing groups to identify the best models: summary of the two-tailed paired t-tests on per-target differences between the models predicted as the best and the actual best models. The summary is shown for top 10 groups according to the cumulative ranking. Groups are sorted according to the increasing cumulative rank. Single-model methods are in bold, quasi-single – in italic (not present in this table), clustering – in plain text. Each non-diagonal cell contains a 4-character string encoding relationship between the two groups. Each symbol of the string relates to a specific evaluation measure in order – GDT_TS, LDDT, CADaa and SG, and shows if the group in the row was statistically better (+), worse (−) or similar (=) to the group in the column. The data are provided for the statistical significance threshold of p=0.05. We define group A as statistically better/worse than group B if it was proven statistically better/worse according to at least two measures and was not proven worse/better according to any remaining measures. All other pairs of groups are statistically indistinguishable or such that unambiguous statistical difference in performance could not be established (shaded cells).

		1	2	3	4	5	6	7	8	9	10
ProQ3	1	X	====	====	====	==+=	=+==	=+==	=++=	=++=	=+++
SVMQA	2	====	X	====	====	====	====	+===	====	+=+=	=+++
FDUBio	3	====	====	X	====	====	====	====	====	====	=++=
MESHI_con_serv	4	====	====	====	X	====	====	====	====	====	=++=
MESHI_server	5	==−=	====	====	====	X	====	====	====	====	=++=
ProQ3_1_diso	6	=−==	====	====	====	====	X	====	====	====	==+=
ProQ3_1	7	=−==	−===	====	====	====	====	X	====	====	==+=
ProQ2	8	=−−=	====	====	====	====	====	====	X	====	==+=
MUfold2	9	=−−=	−=−=	====	====	====	====	====	====	X	==+=
Pcomb-domain	10	=−−−	=−−−	=−−=	=−−=	=−−=	==−=	==−=	==−=	==−=	X