Identification of protein functions using a machine-learning approach based on sequence-derived properties

Table 4 Selection ratios for traditional and new features in the CFS method

Protein class	Number of selected features	Merit value	Traditional features (n = 451)	New features (n = 33)
Transport	38	0.302	8.43%	0%
Transcription	51	0.387	11.31%	9.09%
Translation	76	0.499	16.85%	27.27%
Gluconate utilisation	59	0.59	13.08%	15.15%
Amino acid biosynthesis	52	0.309	11.53%	15.15%
Fatty acid metabolism	90	0.303	19.96%	24.24%
Acetylcholine receptor inhibitor	52	0.974	11.53%	9.09%
G-protein coupled receptor	39	0.487	8.65%	9.09%
Guanine nucleotide-releasing factor	69	0.36	15.30%	27.27%
Fibre protein	31	0.481	6.87%	3.03%
Transmembrane	35	0.443	7.76%	9.09%

The merit value is the highest merit calculated for an optimal subset of the features for each class. The selected features are highly correlated with the class and have low inter-correlation with each other.

ISSN: 1477-5956