Table 1 Signaling networks used for validation of the algorithm.

B2009^15,41

22

46

2

11

4.090

Signaling network of EGF-EGFR and insulin-IR, where the outputs are ERK and AKT. Control goal is to decrease the activities of both ERK and AKT. The inhibitory regulation of AKT against RAF was negligible in the original ODE model since the inhibition of RAF by AKT was found to be very weak in HEK293 cells¹⁵. We also found that the removal of the link between AKT and RAF improves the prediction accuracy of signal flow estimation algorithm in our previous study⁴¹. Hence, the link between AKT and RAF was removed from the original network.

S2015^47,66

69

134

1

13

5.333

Signaling network of TGFβ-driven EMT, where EMT is the only phenotypic output node. Control goal is to decrease the activity of EMT node.

Z2015³⁸

60

142

2

13

4.642

Signaling network of T-LGL leukemia survival, where the two phenotypic nodes of proliferation and apoptosis are defined as the outputs. Control goal is to decrease proliferation and increase apoptosis.

F2013⁴⁸

97

253

2

12

4.641

Signaling network of colorectal tumorigenesis, where the two phenotypic nodes of proliferation and apoptosis are the outputs. We added ‘Proliferation’ node to the original model. Control goal is to decrease proliferation and increase apoptosis.

C2016⁶⁷

200

696

3

13

5.198

Signaling network of colorectal tumorigenesis, where the three phenotypic nodes of proliferation, apoptosis, and metastasis are the outputs. We added ‘Proliferation’, ‘Apoptosis’ and ‘Metastasis’ nodes to the original model. Control goal is to decrease proliferation and metastasis, and increase apoptosis. We set all the ambiguous signs of 51 links as plus in this network.

T2016⁶⁸

4388

15305

—

15

6.522

Literature-curated human signaling network. We obtained a subnetwork by identifying the giant component of the directed and signed links from the original database. No output was defined since the network was used to examine the performance of the algorithm.