Confidence metrics

See Figure 1e

The new metrics, edge score (ES) and edge rank score (ERS), allow one to assess the confidence of an inferred edge. They are determined by comparing inferred weights (IW) from the true-data model and null weights (NW) from null-data models. ES enables like-for-like comparisons of IW between algorithms, and ERS additionally accounts for the specific network context. Both metrics augment the standard interpretation of IW.

Click here to browse confidence metrics.

Dynamical simulations

See Figure 2a

Timecourse in silico data were produced from dynamical simulations of each combination of the different motifs, gates, stimulus conditions, noise levels, and parameter values for logic gate edges.

Click here to browse dynamical simulations.

Kinetic landscapes

See Figure 2b

Network inference performance varies heavily as a function of the kinetic parameters for logic gate edges. Kinetic landscapes exhibit a variety of patterns, and many landscapes have intricate combinations of features resembling phase diagrams. Outcomes are shown for each combination of the different motifs, gates, stimulus conditions, noise levels, parameters for logic gate edges, algorithms, time intervals of input data provided, and metrics.

Click here to browse kinetic landscapes.

Time interval

See Figure 3a

We considered outcomes derived from three time intervals of the dataset: the first half for the transition from resting state to activation, the second half for relaxation toward the initial state or continued activation, and the full timecourse. The analysis shows that there are certain types of dynamics from which algorithms consistently will make confident or non-confident inferences.

Click here to browse time interval.

Stimulus location

See Figure 3c

The decision of which node to stimulate shapes the outcomes of the simulations and inference in consistent ways. Most notably, the gate edge emanating from the node that does not receive the stimulus is inferred with greater confidence.

Click here to browse stimulus location.

Noise level

See Figure 4

Each algorithm differs in its robustness to noise in the data. We introduce chaos as a new metric to assess whether an algorithm's performance is reliable as a function of estimated noise in a dataset.

Click here to browse noise level.

Inferred networks

See Figure 5

Visualization of the full networks produced by each algorithm, using both the inferred weight and the confidence metrics.

Click here to browse inferred networks.