Table 1.
Algorithm 1 From Time-series to Boolean Networks.
Figure 1.
Iterative k-means clustering with (direct binarization) vs.
.
More refined binarization is achieved with higher values of .
Figure 2.
True dynamics (left column) and the dynamics based on asynchronous simulation of the best-scoring Boolean networks learned from the data (right column) of the four systems: toy network (a–b), Jak-Stat (c–d), Smad (e–f), and budding yeast cell cycle (g–h).
The Boolean network simulated for each system is one with minimum error obtained by the KM3:REVEAL method (see Table 2).
Table 2.
Evaluation results for different combinations of binarization and learning methods on the four networks.
Figure 3.
Dynamics of Boolean networks learned from 16 time-points of the toy network.
(a) Time points correspond to 0 min, 5 min, 15 min, 30 min, 45 min, 1 hr, 2hr, 3hr, 6hr, 8 hr, 10 hr, 12 hr, 15 hr, 18 hr, 21 hr, 24 hr. (b) Time points are manually selected to capture the oscillatory patterns of the original system. Left panels show the time points selected, and right panels show the binary data obtained by applying KM3 to the measurements at the selected time points in the left panels. Binarized data are shifted vertically for readability. Blue, green, red, and cyan curves correspond to species A, B, C, and D, respectively.