Page 70 - GPD-1-2

Gene & Protein in Disease Recent advances and challenges of network biology

network is 8.9745; the APL is 3.5550; the graph density world biological networks can be generated. Based on
is 0.0040, which is a very sparse network; the PCC the artificial DD model, the evolution characteristics of
is −0.1222; and the average clustering coefficient of the biological networks can be explored [66,77,82] . However, the
network is 0.1830. The PPI network shown in Figure 1B artificial models can only be used to theoretically explore
has typical scale-free, small-world, disassortative, and possible features of biological networks. There are still some
sparse characteristics. gaps between real-world and artificial biological networks,
which limit the real-world applications of artificial
3. Recent advances of network biology biological networks. Third, by having insights on part of

Some recent advances of network biology are reviewed in the network topology and node dynamics, the unknown
this section. We mainly consider three aspects: Network topological connections can be predicted by dynamical
construction, important node identification in biological complex network theory [84-89] . However, this approach
networks, and applications of biological networks. requires node dynamics, which is generally difficult to
be applied to infer biological networks. The last, but the
3.1. Biological network construction most intriguing is data-driven approaches, which have
Network construction is the first step in network biology. been the focus in the field of network inference. Based on
Biological networks can be determined either from experimentally collected biological data, researchers have
experimental detection or data-driven/model-driven developed various mathematical and statistical models to
inference. Experimental detection is generally costly and infer biological networks, such as those methods based
inefficient, but it is more reliable than model inference. To on correlation analysis [22-26,90-94] , information theory [95,96] ,
experimentally determine whether there are interactions regression [36,37,39,42,88] , Granger causality [40,97] , Bayesian
among pairs of biomolecules, researchers have developed inference [98-101] , and Gaussian graphical model [33-35] . Popular
many experimental methods, such as ChIP-Seq, CLIP-Seq, software packages for biological network construction
[22]
yeast-two-hybrid and yeast-three-hybrid (Y3H), rec- include WGCNA , iDirect [102] , and many others [3,4,93,103] .
[69]
YnH, phage display technology, surface plasmon resonance Data-driven approaches are efficient in inferring biological
(SPR), fluorescence resonance energy transfer (FRET), networks from data; however, for organisms without known
coimmunoprecipitation, glutathione S-transferase (GST) biological network information, it is difficult to determine
pull-down, and so on [4,69-71] . It has been reported that the whether the inferred networks coincide with real-world
recently developed rec-YnH can simultaneously detect ones; furthermore, determining the cutoff threshold values
putative, multi-domain direct protein-protein, and multi- for some methods is also a problem [22-26,90-94] .
protein-RNA interactions under physiological conditions . Other than traditional biological networks, there are
[71]
There are four approaches to constructing biological also studies on differential co-expression networks [49,103-111] .
networks (Figure 2 and Table 1). First, and most Differential co-expression networks reveal the differential
conveniently, existing network data can be downloaded co-expression patterns among genes when comparing
from online databases, such as TRANSFAC, OPHID, between treatments and controls, which are effective
MIPS, DIP, MINT, STRING, BioGRID, HPRD, KEGG, tools for exploring omics data with different experimental
BBID, Reactome, and BIGG [4,67,72,73] . These online settings. The tools or algorithms that can be used to
databases collect both experimentally determined and perform differential co-expression analysis include
literature-curated interaction data among biomolecules, DiffCoEx [102] , CoXpress [103] , DINGO [104] , and many others.
which provide timely and updated valuable resources for Both co-expression and differential co-expression analyses
biological networks. However, the databases only collect are application orientated; the edges in the co-expression
interaction data for some model organisms or organisms or differential co-expression networks do not necessarily
that have been investigated by researchers. The data for indicate physical interactions between two genes; instead,
many organisms are still in a state of uncertainty. Second, they only reflect the similarity between their expression
it is feasible to artificially construct biomolecular networks profiles. Differential co-expression networks facilitate
through computer algorithms. Classical algorithms are researchers to effectively explore bioinformatics from
based on the duplication-divergence (DD) model [66,74-83] . omics data, which have important applications in network
In the DD model, duplication processes of biomolecules biology.
are mimicked by node duplication, while divergence
processes are expressed as random node deletion, edge 3.2. Recent advances in the explorations and
deletion, dimerization, and so on. By appropriately applications of network biology
tuning parameters in the DD model, artificial biological In network biology, guilt-by-association and guilt-by-
networks that have similar topological features with real- rewiring are two frequently used principles to explore

Volume 1 Issue 2 (2022) 4 https://doi.org/10.36922/gpd.v1i2.101