Active machine learning for transmembrane helix prediction

BMC Bioinformatics

Table 2 Comparison of TMpro NN: applying active vs. passive learning algorithms for updating training set from MPtopo (101 proteins, 443 TM segments).

	Methods	# of Proteins in Training-Set	Qok	Qhtm	Qhtm	Qhtm	Q2
				Fscore	%obs	%prd
1	Random	1	14	40	41	41	63
		2	25	60	59	61	67
		5	36	84	88	81	74
		10	35	79	81	78	74
2	Node-Coverage	1	44	91	92	90	78
		2	44	91	91	90	79
		5	44	91	92	90	79
		10	46	91	92	90	79
3	Confusion-Rated	1	26	68	75	65	67
		2	34	85	90	81	75
		5	41	89	91	87	78
		10	45	91	91	90	79
4	Node-Coverage & Confusion-Rated	1	44	91	92	90	78
		2	44	90	91	89	79
		5	44	91	91	90	79
		10	45	90	91	90	78

For description of columns, see caption of Table 1. Q_htm^%obs and Q_htm^%pred have been computed per-protein and averaged over all the proteins.

ISSN: 1471-2105