. Author manuscript; available in PMC: 2016 Aug 11.

Published in final edited form as: J Chem Theory Comput. 2015 Aug 11;11(8):3714–3728. doi: 10.1021/acs.jctc.5b00271

Table 2.

abs_rmsd, rel_rmsd and rad_rmsd for individual training sets.

	Solvation energy rmsd (kcal/mol)				Inverse of effective radii rmsd (1/Å) (DNA, natom = 632)		obj_funct
	dnadup natom = 632		rnadup natom = 640		A-form	B-form
	abs_rmsd w = 1.0	rel_rmsd w = 5.0	abs_rmsd w = 1.0	rel_rmsd w = 5.0	rad_rmsd w = 2.5	rad_rmsd w = 2.5
GB-neck	68.3	29.5	144.6	13.8	0.068	0.075	0.850
GB-neck2	14.0	10.3	25.4	11.7	0.045	0.038	0.338
%reduced_error	80%	65%	82%	15%	34%	49%	60%

%reduced_error shows degree of improvement of GB-neck2 compared to GB-neck, defined by %reduced_error = 100*(X_GB-neck-X_GB-neck2)/X_GB-neck where “X” is either abs_rmsd, rel_rmsd, rad_rmsd or obj_funct. “natom” is the number of atoms for each structure in the training set. Weighting factors “w” are also shown for each set.