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

Table 11 Comparative performance of the novel feature set and traditional feature set using the random forest

Protein class	Novel feature set (33 features)					Traditional feature set (451 features)
	Training accuracy	Test accuracy	Sensitivity	Specificity	AUC	Training accuracy	Test accuracy	Sensitivity	Specificity	AUC
Transport	91.3688	90.30	86.6	93	0.968	92.9764	93.39	89.9	95.9	0.975
Transcription	90.9659	91.26	93.7	88.8	0.98	94.4252	95.33	96.4	94.3	0.99
Translation	97.679	97.78	68.8	100	0.95	98.0741	97.78	68.8	100	0.996
Gluconate utilisation	96.4059	98.11	85.7	100	0.997	97.2516	98.11	85.7	100	0.992
Amino acid biosynthesis	93.7676	94.52	91.7	96.3	0.983	94.836	95.46	94.8	95.9	0.991
Fatty acid metabolism	95.7242	94.44	72.8	99.2	0.97	96.2926	94	69.1	99.5	0.964
Acetylcholine receptor inhibitor	99.6896	100	100	100	1	99.8965	100	100	100	1
G-protein coupled receptor	94.4679	95.92	94.3	96.9	0.991	96.8745	97.18	94.7	98.7	0.993
Guanine nucleotide-releasing factor	96.7429	96.67	62.9	99.3	0.956	98.4985	97.92	74.3	99.8	0.992
Fibre protein	97.4771	95.89	33.3	98.6	0.798	99.2355	99.31	83.3	100	0.998
Transmembrane	93.555	93.52	87.4	96.7	0.978	95.9719	97.53	94.2	99.3	0.995

ISSN: 1477-5956