4. Conclusion
Essential proteins are of great significance for the survival and
development of organism. Recent studies have exerted consider-
able efforts in predicting essential proteins. Many centrality
measures have been proposed to identify essential proteins based
on PPI networks. However, dynamics exist in the PPI networks. This
important inherent property has been ignored by previously
proposed methods. Basing on this fact, we take advantage of
dynamic network topology to predict essential proteins and
propose a new method named CDLC. CDLC predicts essential
proteins by combining dynamic local average connectivity with
in-degree of proteins in complexes.
We can draw the following four conclusions from the
experimental results. First, the quality of our dynamic PPI network
is higher than that of the static PPI network. Second, topology-
based methods can achieve better effects when they are
implemented in the dynamic PPI network than when they are
implemented in the static PPI network. Third, CDLC can outper-
form five previously proposed methods (DC, LAC, SoECC, PeC, and
CoEWC). Furthermore, the prediction precision can be improved by
more than 45% when comparing CDLC with DC. Fourth, CDLC can
achieve higher prediction precision than CEPPK which is the latest
method for discovering essential proteins.
Although CDLC performs better than most previously proposed
methods, our method for the construction of dynamic PPI network
has limitations. The expression value of a gene is obtained by
analyzing the level of mRNA related to it. However, the level of
mRNA is just one of the factors which influence the level of protein
in a cell. That is, the level of protein in a cell is not always correlated
with the expression of the gene. Therefore, the dynamic PPI
network in this study can not perfectly match the real one. This
finding may partially account for the inaccurate prediction of
essential proteins. The contents of dynamic data are expected to
increase in the future. The performance of our method can be
improved by integrating several types of dynamic data to construct
a realistic dynamic PPI network. Further efforts can also be exerted
to improve prediction precision from the aspect of the inherent
property of essential proteins.