A nearest-neighbors network model for sequence data reveals new insight into genotype distribution of a pathogen

BMC Bioinformatics

Table 3 Structural comparison of networks on subsets of the data (resilience to incomplete data)

	Avg. degree	D	CC	Number Comp.	Largest Comp.	\|C₁\|	\|C_>2\|/\|V\|	Genus prec.	Genus recall	Species prec.	Species recall
GroEL Threshold
Full	10.5	5	0.98	304	46	188	62.3%	90.3%	47.3%	60.6%	63.2%
Sample Avg.	6.2	4	0.97	221	29	150	56.1%	89.7%	46.7%	62.8%	64.5%
Sample 1	6.2	5	0.97	221	30	151	53.8%	90.8%	47.2%	62.2%	65.4%
Sample 2	6.3	4	0.97	229	29	158	52.0%	89.3%	45.4%	62.1%	62.6%
Sample 3	6.5	4	0.98	223	27	152	53.2%	89.2%	45.5%	61.0%	63.0%
Sample 4	6.3	5	0.96	212	32	141	56.7%	87.7%	46.2%	63.4%	64.9%
Sample 5	5.9	4	0.98	218	25	146	58.5%	91.6%	49.4%	65.3%	66.7%
GroEL DiWANN
Full	2.6	7	0.19	179	34	0	85.4%	80.4%	43.9%	59.5%	61.8%
Sample Avg.	2.7	6	0.41	119	26	0	86.4%	75.8%	51.1%	55.2%	67.8%
Sample 1	2.8	7	0.52	100	33	0	86.4%	73.6%	52.1%	53.4%	68.4%
Sample 2	2.4	5	0.21	111	22	0	85.6%	78.2%	50.4%	56.4%	66.2%
Sample 3	2.8	6	0.58	113	24	0	84.0%	73.6%	52.7%	54.0%	69.2%
Sample 4	2.6	5	0.38	105	23	0	87.3%	77.9%	46.0%	57.9%	67.6%
Sample 5	2.7	7	0.36	105	29	0	88.9%	75.8%	54.1%	54.5%	68.0%

This table shows a comparison of both structure and clustering results for the GroEL dataset for networks generated from a random sample of 60% of the sequences. D denotes diameter, and CC denotes clustering coefficient. Also shown are the number of connected components, and the size of the largest component. |C₁| gives the number of nodes in clusters of size 1 (singletons), and |C_>2|/|V| shows the percentage of nodes in a cluster of size 3 or above. The full network is also included for comparison. For the threshold based networks, we use a threshold of 30, which had a good trade-off of precision and recall in the community analysis. The full networks contain 812 nodes, while each reduced network contains 487 nodes

ISSN: 1471-2105