Evolution of metabolic network organization

BMC Systems Biology

Table 1 Classification models performance.

Comparison	Accuracy	Kappa statistic	Classification model
Archaea (56) vs. Bacteria (600) vs. Eukarya (87)	93.54% (89.50%)	0.81 (0.62)	Functions.Logistic
Prokarya (656) vs. Eukarya (87)	98.25% (96.90%)	0.91 (0.84)	Functions.MultilayerPerceptron
Unicellular (44) vs. Multicellular (43) Eukarya	96.55% (96.55%)	0.93 (0.93)	Rules.JRip
Free-living (525) vs. Host-associated (61) Bacteria	91.98% (92.32%)	0.45 (0.48)	Rules.OneR
Immotile (202) vs. Motile (322) Bacteria	72.33% (72.14%)	0.40 (0.40)	Lazy.IB1
Anaerobe (253) vs. Facultative aerobe (170) vs. Aerobe (253)	61.20% (57.92%)	0.38 (0.33)	Trees.RandomForest
Halotolerant (4) vs. Halophile (15) Bacteria	78.95%	0.00	Functions.LibSVM
Psychrophile (24) vs. Mesophile (508) vs. Thermophile (61) Bacteria	86.00%	0.04	Functions.LibSVM

Scores obtained by training classification models to discriminate groups of taxa based on quantitative descriptors of the structure and complexity of their Networks of Interacting Pathways (NIPs). Data given is the number of taxa considered for each group, as well as the accuracy and Kappa statistics on 10-fold cross-validation of the best performing classification model when using all 52 NIP descriptors and, in parentheses, those obtained with the best subsets of descriptors identified (see Figure 1).

ISSN: 1752-0509