Revealing and avoiding bias in semantic similarity scores for protein pairs

Table 2 The performance of different data transformation methods*.

Measure	Estimated λ**	λ = 1	Inverse (λ = -1)	Cube-root (λ = 1/3)	Square-root (λ = 1/2)	Square (λ = 2)	Log
Resnik(AVG)	0.878	0	-***	0.645	0.370	0	-
Lin(AVG)	0.890	0	-	0.659	0.474	0	-
RS(AVG)	0.925	0	-	0.632	0.355	0	-
Jiang(AVG)	0.812	0	0.081	0	0	0	0.002
Resnik(BMA)	0.938	0.661	-	0.025	0.248	0	-
Lin(BMA)	0.940	0.706	-	0.012	0.156	0.002	-
RS(BMA)	0.927	0.650	-	0.004	0.042	0.001	-
Jiang(BMA)	0.010	0	0	0	0	0	0
TO	0	0	0	0	0	0	0
NTO	0.555	0.001	0	0.366	0.478	0	0.009
Dice	0.926	0.014	0	0.384	0.890	0	0.001
Kappa	0.896	0.010	-	0.518	0.866	0	-
GIC	0.552	0	-	0.096	0	0	-
VSM	0.291	0	-	0.006	0	0	-

* The numbers in the table represent the percentages of the scores that fitted normal distributions after data transformation, among all group pairs with different length combinations.
** λ was estimated by the method described in the Methods section.
*** "-" indicates the transformation method was not suitable for the similarity measure.