Fication. In this section, we report the experimental results obtained from testing our subgraph search algorithm and the VF2 algorithm [18]. We chose to compare with the VF2 algorithm, because it is the most 1317923 efficient sub-graph isomorphism algorithm based on time [17].Experimental SetupThe computer system used in these experiments was equipped with 3.4 GHz Intel Core i7 processor (4 cores) with 4 GB RAM running Cent OS Linux 5.5. All implementations for these experiments were written in C++. The VF2 algorithm was the optimized versions as presented in the VFLib library.AccuracyWe evaluated the accuracy of our subgraph search algorithm by comparing the number of detected subgraphs between our algorithm and the VF2 algorithm. All graphs with size 3? nodes were generated from signaling network SN1 and SN2 by using the FANMOD and classified into non-isomorphic-graphs. Both algorithms were tested on the signaling networks SN1 and SN2 with non-isomorphic-graphs. The result shows that our algorithm could successfully detect all subgraphs in each signaling network as the VF2 algorithm could. (data not shown).RMOD: Regulatory Motif Detection ToolFigure 6. The run-time comparisons between the RMOD and the VF2 algorithm. The average run-times of searching for all occurrences of a subgraph were measured against various signaling networks. Illustrated results are for (a) 3-node subgraph search (b) 4-node subgraph search (c) 5node subgraph search (d) 6-node subgraph search. Times are given 1315463 in Panels show duration of scratching response and right panels show total milliseconds (ms). doi:10.1371/journal.pone.0068407.gScalabilitySince all the subgraphs in our test datasets were correctly identified by our algorithm, we attempted to test the speed and scalability of our algorithm with our signaling network datasets. Table 2. Computational cost for RMOD algorithm on large signaling networks.Query graph size Network SN5 SN6 3 2545.91 4223.84 4 51137.15 64478.95 5 446923.56 640834.Rows indicate the running time (milliseconds) of our subgraph search algorithm for each query graph size. doi:10.1371/journal.pone.0068407.tWe measured the average run-time for all occurrences of subgraph using 50 k-node query graphs (3#k#6), which are randomly selected non-isomorphic subgraphs generated by the FANMOD, and compared the performance of our algorithm with that of the VF2 algorithm. If the number of non-isomorphic subgraphs in signaling networks is less than 50, all non-isomorphic subgraphs in the signaling network were used as query graphs. Figure 6 shows the average run-time of searching for all occurrences of a subgraph in various sizes of signaling networks, where the size of a single query graph varies. We see that the runtime of our algorithm approximately increases in linear as the size of network increases. We also see that our algorithm shows a significantly smaller run-time than that of the VF2 algorithm, and the difference between our algorithm and the VF2 algorithm becomes even more prominent when the network is large. For example, our algorithm shows about 376 milliseconds (ms) in average run-time for detecting 6-node sub-graphs in signaling network SN4 whereas the VF2 algorithm shows about 14128 ms.RMOD: Regulatory Motif Detection ToolFigure 7. The network editor interface. The network editor allows users to R cells. Transfected ES cells underwent double-selection with the neomycin analogue create or edit input network. doi:10.1371/journal.pone.0068407.gThis difference results from the exponential increase in the path to be explored in the VF2 algorithm. Table 2 shows the experimental results obtained from.Fication. In this section, we report the experimental results obtained from testing our subgraph search algorithm and the VF2 algorithm [18]. We chose to compare with the VF2 algorithm, because it is the most 1317923 efficient sub-graph isomorphism algorithm based on time [17].Experimental SetupThe computer system used in these experiments was equipped with 3.4 GHz Intel Core i7 processor (4 cores) with 4 GB RAM running Cent OS Linux 5.5. All implementations for these experiments were written in C++. The VF2 algorithm was the optimized versions as presented in the VFLib library.AccuracyWe evaluated the accuracy of our subgraph search algorithm by comparing the number of detected subgraphs between our algorithm and the VF2 algorithm. All graphs with size 3? nodes were generated from signaling network SN1 and SN2 by using the FANMOD and classified into non-isomorphic-graphs. Both algorithms were tested on the signaling networks SN1 and SN2 with non-isomorphic-graphs. The result shows that our algorithm could successfully detect all subgraphs in each signaling network as the VF2 algorithm could. (data not shown).RMOD: Regulatory Motif Detection ToolFigure 6. The run-time comparisons between the RMOD and the VF2 algorithm. The average run-times of searching for all occurrences of a subgraph were measured against various signaling networks. Illustrated results are for (a) 3-node subgraph search (b) 4-node subgraph search (c) 5node subgraph search (d) 6-node subgraph search. Times are given 1315463 in milliseconds (ms). doi:10.1371/journal.pone.0068407.gScalabilitySince all the subgraphs in our test datasets were correctly identified by our algorithm, we attempted to test the speed and scalability of our algorithm with our signaling network datasets. Table 2. Computational cost for RMOD algorithm on large signaling networks.Query graph size Network SN5 SN6 3 2545.91 4223.84 4 51137.15 64478.95 5 446923.56 640834.Rows indicate the running time (milliseconds) of our subgraph search algorithm for each query graph size. doi:10.1371/journal.pone.0068407.tWe measured the average run-time for all occurrences of subgraph using 50 k-node query graphs (3#k#6), which are randomly selected non-isomorphic subgraphs generated by the FANMOD, and compared the performance of our algorithm with that of the VF2 algorithm. If the number of non-isomorphic subgraphs in signaling networks is less than 50, all non-isomorphic subgraphs in the signaling network were used as query graphs. Figure 6 shows the average run-time of searching for all occurrences of a subgraph in various sizes of signaling networks, where the size of a single query graph varies. We see that the runtime of our algorithm approximately increases in linear as the size of network increases. We also see that our algorithm shows a significantly smaller run-time than that of the VF2 algorithm, and the difference between our algorithm and the VF2 algorithm becomes even more prominent when the network is large. For example, our algorithm shows about 376 milliseconds (ms) in average run-time for detecting 6-node sub-graphs in signaling network SN4 whereas the VF2 algorithm shows about 14128 ms.RMOD: Regulatory Motif Detection ToolFigure 7. The network editor interface. The network editor allows users to create or edit input network. doi:10.1371/journal.pone.0068407.gThis difference results from the exponential increase in the path to be explored in the VF2 algorithm. Table 2 shows the experimental results obtained from.