(2021.10.13, 2021.10.14 updated)
For SiGN-BN HC+BS, there is an example result file, GN-527-100-result.sgn3, in the DOWNLOAD page. Estimate a network with the bootstrap method using GN-527-100.edf.txt and compare your result with it. For example, execute the network estimation with 100 iterations of the bootstrap method.
Then, you can compare your result using the SiGN-Proc tool.
$ ./signproc -t sgn3 result.sgn3 --comp file=GN-527-100-result.sgn3 ... COML: TP RTP FP FN TN Sn Sp COMP: 8429 1378 3559 1492 123743 0.850 0.631
If your result is similar to this, the network estimation is OK. Note that 100 iterations of the bootstrap method is not sufficient. If you perform 1000 iterations, values of Sn and Sp above will be around 0.9 or so. Also, Here Sp is not the general specificity. It is defined as TP / (TP + FP).
For SiGN-BN NNSR, you can also find an example result file, GN-10k-500-result.sgn3, estimated using GN-10k-500.edf.txt. Both files can be downloaded from the DOWNLOAD page. Estimate a network using signbnnnsr.X.Y.Z, and compare the result with it.
$ mpirun -np 64 signbnnnsr.0.16.8 --total-mem 1000 -T 15000 -o result.sgn3 GN-10k-500.edf.txt $ ./signproc -t sgn3 result.sgn3 --comp file=GN-10k-500-result.sgn3 ... COML: TP RTP FP FN TN Sn Sp COMP: 86242 3790 6565 6663 55437270 0.928 0.893
If you get similar results, network estimation worked properly.
(2022.9.18)
The hubness of a gene is the number of its children (outdegree). In bootstrap method, each edge has its bootstrap probability, and the HubIndex is incorporating this value and it is the sum of the bootstrap probabilities of the edges connecting to its children.
(2022.9.18)
The estimated edges have the EdgeType property, and it is one of 0 (up-regulation), 1 (down-regulation), and 3 (unknown). The EdgeTypeRatio represents the ratio of them (up:down:unknown) after compiling bootstrapped networks.
If the estimated B-spline regression curve is not complex, the program compares the left-most and right-most coefficients. If the right-most coefficient is larger than the left-most one, the curve is regarded as up-regulation, and vice versa. If the curve is complex, it is regarded as unknown regulation.