Fig 1.
Schematic diagram of LBN method.
(1) process the data, (2) construct the initial network (a large-scale network) by CMI or MI, (3) decompose the network into local networks (a number of small-scale networks) by kNN with k = 1, (4) perform BN to obtain local BNs (a number of small-scale networks), (5) integrate local BNs into a candidate network (a large-scale network), (6) perform CMI to obtain the tentative network (a large-scale network). Iteratively performing BN and CMI with kNN (k = 2) until GC topological structure tends to stable, the final network or GRN can be inferred. The solid lines denote the true regulations and the dashed lines denote redundant correlations between two genes.
Table 1.
Comparison of different methods on dataset10, dataset50 and dataset100.
Table 2.
Comparison of Grow-shring, IAMB and LBN methods on dataset10.
Fig 2.
(a) True network. (b) Inferred network with LBN (α = β = 0.01). The solid lines are correctly inferred regulatory relationships, and the dotted lines are false regulatory links.
Table 3.
Comparison of different methods on SOS DNA repair network.
Table 4.
Comparison of different methods on the large-scale gene regulatory network.
Fig 3.
Gene regulatory networks composed of 10 genes.
(a) The true network with 10 genes and 10 edges. (b) The network inferred by BN method. (c) The network inferred by MI+BN. (d) The network inferred by MI+BN+CMI. (e) The network inferred by MI+BN+CMI+kNN+BN. The solid lines are correctly inferred regulatory relationships, and the dotted lines are false regulatory links.
Table 5.
Results of different combination ways on the dataset10.
Fig 4.
Effect of parameters α and β for LBN on dataset10.