Table 2

A comparison of the contributions of each dataset to the composite human protein interaction map, with network properties of each of the datasets

Dataset

Version

Number of interactions

Number of proteins

Clustering <C>

Connectivity <#interactions/protein>

Reactome

08/03/04

9,987

619

0.74

15.4

BIND

08/03/04

1,536

1,212

0.1

1.3

HPRD

04/12/04

6,054

2,747

0.09

2.2

Orthology inferred (core)

03/31/04

9,283

3,469

0.13

2.7

Co-citation

This paper

6,580

3,737

0.3

1.8

Total

This paper

31,609

7,748

0.24

4.1

An analysis of network features (clustering coefficient [38] and degree of connectivity) of each of the datasets indicates low degree (<k>) for all except Reactome, which is by far the most densely sampled protein interaction dataset. The final combined network is modular in structure and shows extensive, non-random clustering of proteins as compared to randomly generated networks with equal numbers of proteins and interactions (<C> = 9 × 10^-3 ± -3 × 10^-5; average of 10 trials).

Ramani et al. Genome Biology 2005 6:R40   doi:10.1186/gb-2005-6-5-r40