Inferring genetic interactions via a nonlinear model and an optimization algorithm

BMC Systems Biology

Table 3 Performances of GASA, TSNI, NCA, GAGA and GA-regular SA applied to data simulated from Eq. (2) with high level of noise, where the averaged results of five repeats are reported

		# int^a	# pc^b	TPR^c	TNR^d	FPR^d	mFPR^e
GASA	AIC/no power law			0.66	0.97	0.03	0.19
	BIC/power law			0.65	0.97	0.03	0.19
GA-regular SA	AIC/no power law			0.5	0.95	0.05	0.30
	BIC/power law			0.47	0.96	0.04	0.31
NCA	100% true connectivity			0.58	0.94	0.06	0.31
	50% true connectivity			0.25	0.85	0.15	0.73
TSNI	inputting prior knowledge: 26 true links	3	1	0.50	0.89	0.11	0.50
		3	2	0.50	0.89	0.11	0.50
		3	3	0.50	0.89	0.11	0.50
GA-GA	AIC/no power law			0.31	0.83	0.17	0.71
	BIC/power law			0.31	0.90	0.10	0.60

^a '# int' denotes the number of interpolations.
^b '# PC' denotes the number of principal components
^c TPR is the percentage of correctly predicted links out of the total number of existing links in a simulated network. Note signs of interactions were not accounted toward TPR and other performance measure.
^d TNR (FPR) is the ratio of correctly predicted non-existing links (false positives) over the total true negatives.
^e mFPR is the ratio of incorrectly predicted links to the total predicted links.

ISSN: 1752-0509