A Survey on Identification of Protein Complexes in Protein–protein Interaction Data: Methods and Evaluation - Computational Intelligence Techniques for Comparative Genomics - page 60

The average weight of graph G and the weighted density of the core are cal-
culated with Eqs. ( 2 ) and ( 3 ). The neighborhood af nity for the two graphs and to

measure the similarity between the graphs, it lters the redundant core and deter-

mines whether the predicted protein complex matches the real protein complex by
using the Eq. ( 1 ).
Next, the method considers the pairwise proteins with high af nity score as core

proteins and forms protein complex core from edge. Firstly, the preliminary cores
are detected from the neighborhood graph of the pairwise proteins in the PPI
network. If the neighborhood graph satis

es the requirement, it is directly regarded
as preliminary core, otherwise, will be dealt with until it meets the requirements.
Then, due to different pairwise proteins may produce same or similar preliminary
cores, which will lead to high redundancy. In order to eliminate the redundant
cores, we use redundancy

ltering procedure to process preliminary cores and

nally get the protein complex cores.

Detecting the Attachment

It identi es the attachment proteins for each core to form protein complex. In this

method, it identi es attachment proteins by edge [ 17 ]. The criteria for selecting

attachment protein not only apply the properties of network topology but also
consider the internal af

nity between core and attachment.
Generally, the attachment proteins densely interact with core proteins than other
attachment proteins. The af nity score between core and attachment is high [ 16 ]. It

may participate in multiple complexes unlike the unique core. To identify such
proteins, a function is needed to measure the closeness between attachment and
core.

Þ¼ j

N v \ V CG j

closeness ð v ; CG ð 10 Þ

j V CG j

Using this method, those proteins closely associated with complex core can be
selected as attachment proteins. However, there exists many attachment proteins,
they only interact with a few core proteins, but the gene co-expressed consistency
and the function similarity is high. The average weight between attachment proteins
and core proteins is de

ned as,

j W C j

AW ð v ; CG Þ¼ ð 11 Þ

j N V \ V CG j

Finally, in this method, it identi es the attachment proteins for each core to form

protein complex.