Table 2.

Overview of selected network reconstruction algorithms

GRN algorithms	Concept	Assumptions about gene-gene relation	Data prerequisite	Edge predictions
PIDC	mutual information	possibly non-linear	discrete	undirected
GENIE3	tree-based	possibly non-linear	continuous	directed
GRNBoost2	tree-based with gradient boosting	possibly non-linear	continuous	directed
PPCOR	partial correlation	Gaussian, linear	continuous	undirected

GRN algorithms are summarized based on their underlying methodology and theoretical assumptions about gene-gene relation. Except PPCOR, the algorithms do not have strong assumptions about gene-gene relation, and thus non-linear interactions can possibly be inferred. PIDC requires a data discretization, whereas all other methods work on continuous data. The inferred network may contain causal interactions (directed) or associated interactions (undirected).