Computing paths and cycles in biological interaction graphs

BMC Bioinformatics

Table 2 Benchmarks for calculation of shortest signed paths between all pairs of nodes.

Network	Number of uSCCs (with number of nodes)	Algorithm
		approximation with DLACC-TI			TSA	DFT
		[s]	TI corrections	remaining errors	[s]	[s]
T-cell	1 (33)	0.79	71	2	0.34	0.02
EGFR	1 (33)	1.18	183	1	0.61	0.26
T-cell+EGFR	2 (33, 33)	6.0 ± 0.04	879 ± 79	3 ± 0	3.1 ± 0.03	419 ± 82
Regulon DB	1 (30)	103	145	0	11.8	1.0
CA1 neuron	1 (154)	25	1869	43*	582*	2213*
Cancer signaling	4 (2, 2, 2, 445)	243	2161	n/a	>12 h	>12 h

The running times for the different algorithms are shown and the quality of the approximation with DLACC-TI is assessed. Also, the number of uSCCS in the networks together with the number of nodes that they contain is shown. In the column "TI corrections" the number of shorter paths that can be identified with transitive inference after having run DLACC is given. The "remaining errors" column shows how many shortest paths (after the TI step) differ in their length compared to the exact results delivered by TSA or DFT. When an algorithm ran longer than 12 hours it was considered impractical and terminated. Therefore, no exact results were determined for the cancer signaling network and consequently the quality of the approximation with DLACC-TI cannot be given (*) For the CA1 neuron, the search depths in the two-step algorithm and in the exhaustive search were limited to length 18 to make calculations practicable. Therefore some paths may have been missed. The longest shortest path identified for this network with the DLACC-TI is also of length 18. The computational environment is the same as in Table 1.

ISSN: 1471-2105