Figure 1.
Influence of noise level on the performance of ssMIKANA, tsMIKANA and cMIKANA network inference methods.
Different MIKANA network models were inferred from 100-gene scale-free networks. The sensitivity and false discovery rate (FDR) from MIKANA inference methods with steady-state data only (ssMIKANA), time-series data only (tsMIKANA) and the combination of steady-state and time-series data (cMIKANA) are compared. Different noise levels, 1%, 3%, 5%, 8%, 10%, 13%, 15%, 18% and 20%, were added to data, respectively.
Figure 2.
The performance of tsMIKANA and cMIKANA inference methods on the same size of data samples for scale-free networks with 100 genes.
The sensitivities and false discovery rates (FDRs) from MIKANA inference methods with time-series data only (tsMIKANA) and with the combination of steady-state and time-series data (cMIKANA) are compared. We reconstructed tsMIKANA network models from time-series datasets – each contained 5, 10, 20 and 40 data samples in 3 replicates, providing 15, 30, 60 and 120 data samples, respectively. We also reconstructed cMIKANA network models from the combined datasets with the same size – containing time-series data sample from 1 temporal experiment (5, 10, 20 and 40 time-series data samples, respectively) and the remainder samples were collected from several knockdown experiments (containing 10, 20, 40 and 80 steady-state data samples, respectively). 10% noise was added to both time-series and steady-state data. Being reconstructed from the same sizes of data, cMIKANA models showed higher sensitivity and relatively lower FDRs compared with tsMIKANA models, suggesting the prediction of gene regulatory interactions could be improved by incorporating steady-state data.
Figure 3.
Comparison of the directionality of edges in the ssMIKANA, tsMIKANA and cMIKANA network models.
Scale-free networks with 100 genes were generated and related steady-state datasets were simulated. Time-series data were collected at each time point in 10 replicates. 10% noise were added to data. The proportion of edges in the canonical networks found in the forward network models (left column) and in the reversed network models (right column) were computed.
Figure 4.
Venn diagram showing the network edges present in three inferred models.
The edge-wise comparison between the ssMIKANA, tsMIKANA and cMIKANA models which were reconstructed from the related endothelial dataset is made.
Table 1.
Overlap of edges between all pairs of 50-gene network models.
Table 2.
Top 10 hubs from the ssMIKANA, tsMIKANA and cMIKANA network models respectively, reconstructed from endothelial datasets.
Table 3.
Biological enrichment analysis for the top 10 hubs in the ssMIKANA, tsMIKANA and cMIKANA network models.