Table 2

The predictive performance of the naïve Bayesian inference program, achieved when implementing a Gaussian likelihood function of a.) the observed structural characteristics alone, b.) when implementing the observed protein family frequencies alone as likelihoods and c.) when combining the observed protein family frequencies with the Gaussian likelihood functions of observed structural characteristics.

Class

Test set


a. Structural features

MCC

% Correct predictions


Thermophiles

0.24

80.0

Mesophiles

0.36

50.0

Psychrophiles

0.47

25.0


b. Protein families

MCC

% Correct predictions


Thermophiles

0.60

92.9

Mesophiles

0.13

28.6

Psychrophiles

0.51

50.0


c. Combined

MCC

% Correct predictions


Thermophiles

0.67

92.0

Mesophiles

0.40

57.1

Psychrophiles

0.68

50.0


(For the individual predictions, see Additional file 5, 6 and 7)

Jensen et al. BMC Genomics 2012 13(Suppl 7):S3   doi:10.1186/1471-2164-13-S7-S3

Open Data