Implementation of multiple-instance learning in drug activity prediction

BMC Bioinformatics

Table 4 Optimization of tuning parameter λ for MILES

Data set	Dissimilarity measure	Cross-validationa	λ	n ^b
I	Soergel	0.777	8.000	196
	Dice	0.761	4.400	165
	Manhattan^c	0.803	4.400	130
	Rogers-Tanimoto	0.801	4.000	153
II	Soergel	0.865	0.001	103
	Dice	0.865	0.001	85
	Manhattan^c	0.877	0.022	63
	Rogers-Tanimoto	0.868	0.069	72
III	Soergel	0.899	0.001	94
	Dice	0.901	0.001	75
	Manhattan	0.934	0.550	63
	Rogers-Tanimoto^c	0.935	4.400	46
IV	Soergel	0.579	0.003	125
	Dice	0.544	0.031	111
	Manhattan	0.690	0.550	87
	Rogers-Tanimoto^c	0.689	6.800	78

^a The median classification accuracy for 5 replications of 5-fold cross-validation; ^b the number of prototype conformers selected in the set Γ; ^c The model selected based on the number of prototype conformers.

ISSN: 1471-2105