Machine learning methods for metabolic pathway prediction

BMC Bioinformatics

Table 2 Best-performing Boolean features, ordered by information gain

Feature	ACC	SN	SP	FM	PR	RC	IG
has-enzymes	0.821	0.914	0.796	0.681	0.543	0.914	0.188
has-reactions-present	0.797	0.919	0.765	0.655	0.509	0.919	0.173
majority-of-reactions-present	0.872	0.707	0.916	0.699	0.69	0.707	0.165
some-initial-reactions-present	0.84	0.724	0.87	0.654	0.597	0.724	0.138
some-initial-and-final-reactions-present	0.864	0.605	0.933	0.651	0.706	0.605	0.136
mostly-absent-not-unique	0.215	0.163	0.229	0.08	0.053	0.163	0.133
all-initial-reactions-present	0.825	0.747	0.845	0.641	0.561	0.747	0.133
every-reaction-present	0.871	0.508	0.968	0.623	0.807	0.508	0.132
every-reaction-present-or-orphaned	0.871	0.508	0.968	0.623	0.807	0.508	0.132
taxonomic-range-includes-target	0.795	0.813	0.79	0.624	0.506	0.813	0.131

See section "Feature Extraction and Processing" and Section 1 of Additional file 2 for description of features.
Columns 2 through 8 correspond to various performance measures: ACC = accuracy; SN = sensitivity; SP = specificity; FM = F-measure; PR = precision; RC = recall; IG = information gain.

ISSN: 1471-2105