. Author manuscript; available in PMC: 2022 Mar 8.

Published in final edited form as: J Chem Inf Model. 2020 Sep 10;60(9):4200–4215. doi: 10.1021/acs.jcim.0c00411

Table 4:

Effect of using an ensemble of models compared to average of individual model performance. BP: Best possible fraction of low RMSD poses; Rand: expected fraction of randomly sampled low RMSD poses.

Train	Test	Model	Evaluation	RMSE	R	AUC	Top1	BP	Rand
CrossDock	CCV	Dense	Average	1.55	0.547	0.944	0.615	0.970	0.0321
CrossDock	CCV	Dense	Ensemble	1.42	0.612	0.956	0.684	0.970	0.0321
CrossDock	CCV	Def2018	Average	1.47	0.577	0.906	0.537	0.970	0.0321
CrossDock	CCV	Def2018	Ensemble	1.45	0.587	0.914	0.574	0.970	0.0321
General	Core	Dense	Average	1.490	0.733	0.942	0.788	0.946	0.135
General	Core	Dense	Ensemble	1.348	0.788	0.960	0.836	0.946	0.135
General	Core	Def2018	Average	1.383	0.787	0.943	0.802	0.946	0.135
General	Core	Def2018	Ensemble	1.368	0.796	0.946	0.814	0.946	0.135
Refined	Core	Def2018	Average	1.503	0.720	0.932	0.766	0.946	0.135
Refined	Core	Def2018	Ensemble	1.438	0.749	0.941	0.800	0.946	0.135