Table 1.

Assessed network inference models.

Method	Type	Lags	Category	Features
catnet	Static	–	Bayesian network	– Categorization of data
				– Stochastic search (simulated annealing) in the network space
pcalg	Static	–	Bayesian network	– Progressive removal of edges (backwards selection)
				– Conditional dependence estimated with partial correlation
GeneNet	Static	–	Graphical Gaussian Model	– Full partial correlations estimated through shrinkage
				– Edges are directed from the most to the less exogenous variable
VAR I +lars	Dynamic	Fixed (first)	VAR	–VAR(I) model subject to a LI penalty term
				– Regression coefficients estimated with least angle regression (lars)
simone	Dynamic	Fixed (first)	VAR	–VAR(I) model subject to a variable penalty term (to favor the selection of transcription factors)
				– Regression coefficients estimated through optimization
GI DBN	Dynamic	Fixed(first)	Dynamic Bayesian network	– Estimation of a number of first order partial regression coefficients, for each possible interaction
				– Predictors and target are lagged by I time point
Time Delay ARACNE	Dynamic	Estimated(one)	Information–theoretic	– Mutual information used to infer dependencies (MI estimated with a copula–based approach)
				– Estimation of the lag between two genes
				– Use of the DPI to break up fully connected triplets
Time lagged MRNET	Dynamic	Estimated(one)	Information–theoretic	– Mutual information used to infer dependencies (Gaussian assumption)
				– Estimation of the lag between two genes
				– mRMR feature selection
Time lagged CLR	Dynamic	Estimated(one)	Information–theoretic	– Mutual information used to infer dependencies (Gaussian assumption)
				– Estimation of the lag between two genes
				– Normalization of MI