Functional discrimination of membrane proteins using machine learning techniques

BMC Bioinformatics

Table 3 Discrimination of channels/pores, electrochemical potential-driven transporters and primary active transporters using different machine learning approaches with amino acid composition as features

Method	5-fold cross-validation
	Sensitivity			Precision			F-Measure			Accuracy
	F1	F2	F3	F1	F2	F3	F1	F2	F3	(%)
Bayesnet	0.582	0.777	0.538	0.606	0.643	0.612	0.594	0.703	0.573	62.1
Naive Bayes	0.496	0.823	0.534	0.626	0.597	0.606	0.554	0.692	0.568	60.7
Logistic function	0.535	0.695	0.619	0.615	0.638	0.601	0.572	0.665	0.610	61.6
RBF network	0.543	0.735	0.625	0.640	0.666	0.603	0.587	0.699	0.614	63.3
Support vector machine	0.469	0.757	0.642	0.675	0.620	0.603	0.553	0.682	0.622	62.4
k-nearest neighbor	0.525	0.707	0.572	0.586	0.588	0.615	0.554	0.642	0.593	59.8
Bagging meta learning	0.541	0.679	0.677	0.646	0.660	0.618	0.589	0.669	0.646	63.6
Classification via Regression	0.492	0.695	0.630	0.599	0.628	0.599	0.540	0.660	0.614	60.8
Decision tree J4.8	0.506	0.580	0.572	0.529	0.581	0.554	0.517	0.580	0.563	55.5
NBTree	0.512	0.669	0.569	0.569	0.610	0.568	0.539	0.638	0.569	68.2
Partial decision tree	0.473	0.649	0.550	0.544	0.551	0.568	0.506	0.596	0.559	55.6
Neural network	0.549	0.717	0.642	0.636	0.659	0.619	0.589	0.687	0.630	63.6
Jack-knife test	0.571	0.709	0.676	0.664	0.660	0.644	0.571	0.709	0.676	65.4
Equal data	0.635	0.713	0.624	0.689	0.698	0.591	0.661	0.705	0.607	65.7

F1: channels/pores; F2: electrochemical potential-driven transporters; F3: primary active transporters. Equal data: Results obtained with a dataset of 502 proteins each in all the three classes of transporters.

ISSN: 1471-2105