Skip to main content

Table 3 Correlation between GO functional annotation and link density in networks from ResNet 4.0 database

From: Automatic extraction of gene ontology annotation and its correlation with clusters in protein networks

Network Annotation ResNet 4.0 Physical Regulation Expression Molecular Transport Promoter Binding
Biological process GOA (public) 62.9% (1247/1982) 64.1% (1247/1946) 48.0% (909/1895) 28.8% (527/1833) 25.7% (383/1491) 8.3% (103/1234)
Biological process GOA (MedScan) 41.8% (777/1858) 56.2% (1026/1825) 14.2% (262/1851) 7.4% (136/1836) 22.6% (389/1722) 7.6% (121/1595)
Biological process GOA (combined) 47.6% (1363/2861) 59.2% (1662/2809) 21.5% (595/2762) 13.8% (376/2722) 21.6% (524/2428) 6.6% (142/2166)
Cellular component GOA (public) 80.9% (361/446) 78.7% (351/446) 43.0% (165/384) 34.6% (128/370) 43.1% (118/274) 1.0% (2/209)
Molecular function GOA (public) 56.6% (489/864) 53.3% (434/815) 35.1% (269/767) 27.3% (196/719) 19.5% (99/507) 2.5% (9/357)
  1. Table 3 shows the percentage of GO subnetworks with high link density for six network types in ResNet 4.0 for five GOAs. A GO sub-network was created by connecting all proteins within a GO group by all possible links from the given network type. All proteins that form child groups were included into the parent group recursively. The numbers in bold face indicate the fraction of these subnetworks that are densely linked, while the numbers in brackets show how this fraction was calculated. The first number in brackets indicates the total number of densely linked sub-networks and the second number shows the total number of all GO sub-networks. ResNet 4.0 contains Physical interaction relations (Binding + Protein Modification) and regulatory relations (Regulation, Expression, MolTransport, PromoterBinding). The statistics for all networks was described previously [29] and is summarized in Table 5. Regulation network was considered without DirectRegulation relations indicating the regulation by means of physical interaction.
  2. To compute the number of densely linked GO sub-networks, we have randomized corresponding network using the algorithm that preserves the number of interaction partners of each protein [30]. After 10 randomizations, we calculated the average number of relations and their standard deviation in every GO sub-network and compared them to the number of the original network relations in every GO group. We then counted number of GO groups for which the number of actual links is greater than the number of randomized links by at least five standard deviations. Assuming that the randomized number of links in a GO group is normally distributed, this means that the selected GO groups could have such elevated number of links by pure chance only with a p-value less than 10-6.