Hidden Structure of Brain
Hidden Structure of Brain
Hidden Structure of Brain
forces, the geometrical shapes and compatibility of the building blocks are decisive factors in each
phase of growth. Therefore, a comprehensive understanding of these processes is essential for
the design of large assemblies of desired properties. Here, we introduce a computational model
for cooperative self-assembly with simultaneous attachment of structured groups of particles,
which can be described by simplexes (connected pairs, triangles, tetrahedrons and higher order
cliques) to a growing network, starting from a small seed. The model incorporates geometric
rules that provide suitable nesting spaces for the new group and the chemical affinity ν of the
system to acceping an excess number of particles. For varying chemical affinity, we grow different
classes of assemblies by binding the cliques of distributed sizes. Furthermore, to characterise the
emergent large-scale structures, we use the metrics of graph theory and algebraic topology of
graphs, and 4-point test for the intrinsic hyperbolicity of the networks. Our results show that
higher Q-connectedness of the appearing simplicial complexes can arise due to only geometrical
factors, i.e., for ν = 0, and that it can be effectively modulated by changing the chemical potential
and the polydispersity of the size of binding simplexes. For certain parameters in the model we
obtain networks of mono-dispersed clicks, triangles and tetrahedrons, which represent the geo-
metrical descriptors that are relevant in quantum physics and frequently occurring chemical clusters.
PACS numbers:
FIG. 1: (a) Examples of geometrical shapes identified as cliques of the order qmax = 1, 2, 3, 4, from left to right. (b) Addition
of a tetrahedron (qmax = 3) to the system of blue nodes can be nested in three different ways, i.e., by its face of the dimension
q = 0, for instance, including the node “1”, q = 1, the link “3-6”, and q = 2, the triangle “4-5-6”. The corresponding number
of new particles na = qmax + 1 − (q + 1) = qmax − q are shown by red nodes. (c) The number of simplexes Σ(t) as function
of time for aggregation of poly-disperse cliques at different parameter ν. Lower panel shows the corresponding number nσ (t)
of added simplexes per time step. Inset: Average growth rate RΣ ≡ hdΣ/dti vs. ν. Bottom panels: Networks of aggregated
cliques of sizes s ∈ [2, 10] for the varied chemical potential ν = −9, ν = 0, and ν = +9, left to right. Different colours of nodes
indicate the network’s community structure.
shown by the blue nodes. Considering Eq. (1), the case of new nodes, and list of all nodes which belong to that
ν = 0 describes the probability of attachment by geomet- clique.
rical factor alone. In this case, the population of docking
sites of the order q determines the likelihood that a new 24 42 729 2 1 14 42
clique will attach by its q-face. On the other hand, the 25 45 785 6 3 28 29 32 43 44 45
number of docking sites of a given size depends on the 26 46 789 3 1 16 28 46
actual structure of the network. Note that, by adding 27 47 791 2 1 12 47
a particular clique of the order qmax to the system, all 28 48 795 3 1 17 36 48
its unshared faces also appear as new cliques of lower or- 29 49 797 2 1 36 49
ders. Thus, the number of simplexes fluctuates in time 30 55 1805 10 6 17 28 30 31 50 51 52 53 54 55
depending not only on the dimension of the clique which For varied chemical potential ν, despite the statistically
is formed in the docking event but also on the size of similar population of cliques appearing in the process
the actual docking site. It should be noted that while (taken from the same distribution), the network growth
the simplicial complexes grow through the attachment of speed and the average rate of the addition of simplexes
new cliques via shared faces, the process can not generate RΣ ≡ hdΣ(t)/dti are different being dependent on the
holes and cliques of the order larger than the cut-off size docking probability. Fig. 1c displays the evolution of the
smax of the original distribution g(s). In the simulations, total number of simplexes Σ(t) and the number nσ (t) of
we keep track of details constituting each event. For ex- the added simplexes per time step for different networks
ample, the small segment of the output file shown below until they exceed the targeted size of N = 1000 nodes for
indicates the time step, current network size, the num- the first time.
ber of simplexes, order of the added clique, the number Specifically, a fast growth of the network is observed
4
for the negative values of the parameter ν while much each other, i.e., for ν < 0, their common faces will be
slower growth rates characterize the assembly process the lowest orders, such as single nodes and links and,
for ν ≥ 0. In fact, for ν < 0 the system “likes” ad- less often, triangles or higher structures. The opposite
dition of new particles, which represent the non-nested situation typically occurs for ν > 0 where the simplexes
parts of the new clique. Hence, the cliques effectively have a high affinity towards sharing nodes; cf. structure
repel each other resulting in a sparse structure and fast in Fig. 1. Due to shared faces, for instance, of the order
growth of the network size and also the addition of new q, the number of distinguishable simplexes of that order
simplexes. In contrast, when ν > 0 the cliques are prefer- is smaller than the number of faces Cq of a free added
ably nested along their larger faces, thus reducing the clique. Therefore, the topological response function fq
number of the newly added nodes. This powerful at- of the network33 can be determined as the number of
traction among cliques leads to dense network structure different simplexes at the topology level q; it provides
and a small number of added nodes and unshared faces a good measure of the combinatorial complexity of the
per time step. This situation results in a slower growth of assembly in response to the varying external parameters
the network and reduced simplex addition rate, as shown ν, smax . In Fig. 2, we show how the function fq varies
in Fig. 1c. In contrast, the case with strictly geometri- along the topology levels q depending on the parameter ν
cal assembly, ν = 0, has no preference for any size of a and the range of the distribution of the attaching cliques.
docking site; the probability is strictly determined by the The peak of the distribution shifts towards higher val-
number of locations of a given size. Accordingly, these ues when larger dimension cliques appear, whereas the
details of the process have an impact onto the topology of height depends on the way that they interconnect at each
the evolving assemblies, which we study in the following. the topology level. Further, Q-analysis based on the
For illustration, three examples of the networks contain-
ing the number N ≥ 1000 nodes for varied parameter 10
4.3
12
ν and the same distribution of the incident cliques are i=-9 i=0 i=+9 10
4.2 8
10
shown in bottom panels in Fig. 1. 6
4
fq 104.0
3.7
10
B. Combinatorial topology of aggregates with
poly-disperse cliques 0.0
2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12
q+1 q+1 q+1
200
clustering coefficient, which can be related to the original 100
population of cliques. On the other hand, for the posi- 1000 i=-9
tive values of ν, the cliques firmly attach to each other, 800 i=-5
i=-1
SSV
600 i=0
resulting in a gradually smaller number of the simultane- 400 i=1
i=5
1
a large core of densely packed higher-order cliques while 0.8
low-order structures remain at the periphery. An im-
TSV
0.6
0.4
pressive network architecture with well-separated com- 0.2
munities appears for ν = 0, assembled under geometrical 0
0 2 4 6 8
constraints alone. As described below, the graph prop- q
FIG. 4: Adjacency matrix of the network’s nodes which participate in structures that are “visible” at the topology level q=5.
left to right: ν =-1, 0, and +1. Different colors identify clusters or communities.
Pc(k)
-(x/12)
hierarchical structure of various emergent networks. In 2
10
y=5800e
2 qmax=6
of cliques of all dimensions (taken from the same distri- 10 qmax=4
bution) in all studied networks. However, their inherent 1
o=2
10
structure is significantly different, which is expressed by
the components of SSV and TSV for various q (see Meth- 0
10 0 1 2 3
10 10 10 10
ods). Notably, the third structure vector in networks for k
ν < 0 has non-zero components only at lowest topol-
ogy levels; this implies that different higher-order cliques FIG. 5: Cumulative distributions of the degree in networks of
present in the graph will be separated from each other aggregated poly-disperse cliques s ∈ [2, 12] and varied chem-
by removing the structures of the order q = 1 (link) be- ical potential ν (top panel) and for purely geometrical ag-
tween them. The situation is much different in the assem- gregation (ν = 0) and varied size of the largest added clique
blies grown when ν > 0 where the simplicial complexes smax (lower panel). Each distribution is averaged over several
containing the higher-order cliques remain strongly inter- samples of the networks with the number of nodes N ≥ 5000.
connected until the before-last level qmax − 1 =8. These
findings agree with the impact of the chemical potential
favoring the cliques attraction for ν > 0 and repulsion
for ν < 0. In this context, it is interesting to note that The node’s participation in building various simplexes
structure that was grown solely under the geometrical also manifests in the global statistical features of the
rules (ν = 0) already possesses a sizable hierarchical or- network. The cumulative degree distribution for sev-
ganization of simplicial complexes; although the degree eral studied aggregates is given in Fig. 5. It is averaged
of connectivity is systematically lower than in the case over several realizations of the systems containing over
ν = +1, the structure holds together until the level q = 7. 5000, where smax ∈ [2, 12]. Although a broad distribu-
(See Table I for the exact values). As the Fig. 3 shows, tion of the node’s degree occurs in each case, it strongly
this hierarchical architecture of the assembled networks varies with the parameter ν. It is interesting to note
gradually builds with increasing values of the parameter that, in the networks grown by geometrical constraints
ν. To illustrate the differences in the hierarchical orga- with ν = 0, we obtain the distribution with a power-
nization of the systems for ν = -1, 0, +1, in Fig. 4 we law decay τ + 1 ≈ 3 (within the numerical error bars);
display those parts of their structure that are still vis- its cut-off appears to depend on the size of the largest
ible at the topology level q = 5. Precisely, the nodes clique. In contrast, the exponential decay is observed
participating in the simplexes of order q ≤ 5 which are for ν < 0 while a structure containing many nodes of
not faces of the cliques of the order q > 5, are removed. a large degree is present in the case of clique attraction
The connections among the remaining nodes are shown for ν > 0, which is separated from the low-degree nodes.
according to the network’s adjacency matrix. Other graph theoretic measures also vary accordingly.
6
C. δ-Hyperbolicity of the emergent networks building blocks. Some compelling examples are the ag-
gregates of tetrahedra and triangles. Tetrahedral forms
For network structures, δ-hyperbolicity is a general- are ubiquitous minimum-energy clusters of covalently
ization of negative curvature in the large36 . Here, we bonded materials12 . We also study the impact of the
consider the aggregates of cliques, which are known 0- chemical potential in the event of aggregation of trian-
hyperbolic graphs; therefore, these structures are ex- gles. The importance of triangular geometry was recently
pected to exhibit this intrinsic property at a larger scale. pointed in the context of quantum networks43 . Some ex-
Following the procedure described in36 , we investigate amples of these structures grown by the aggregation rules
the 4-point Gromov hyperbolicity of different emergent of our model are shown in Fig. 7.
networks. Specifically, we determine the average hyper-
bolicity hδi in comparison to the graph’s diameter for
ν = -5, -1, 0,+1, and +5, by a sampling of 109 sets of
four nodes, as described in Methods. Considering three
different realizations of the network for each ν, we find
numerically that δ can take the values {0, 1/2, 1}; hence,
the maximum value δmax = 1 suggests that these as-
semblies are 1-hyperbolic. In Fig. 6 (bottom panel) we
plot the average hyperbolicity hδi against the minimal
distance dmin of the involved pair in the smallest sum
S, see Methods. Notably, for all network types hδi re-
mains bounded at small values. In particular, we find
that hδi = 0 for the tree graph of cliques corresponding
to ν = −5. Whereas, the hyperbolicity parameter is close
to zero in the sparse network of cliques for ν = −1, and
slightly increases in the more compact structures corre-
sponding to ν = 0 and ν > 0. Note that due to a small
number of pairs of nodes having the largest distance in
the graph we observe the fluctuation of hδi ∈ [0, 0.5]. The
histograms of distances between all pairs of nodes in the
considered networks are also shown in Fig. 6 (top panel).
4.6
10
4.5
i=-1
10 i=0
i=1
P(d)
4.3
10 i=5
4.0
10
0.0
2 4 6 8 10 12 14
0
d
10
-1
10
<b>
-2
10
FIG. 7: Aggregates of tetrahedra (top) and triangles (bottom)
-3
10 for ν=0.0.
-4
10
2 4 6 8 10
dmin Since the aggregation process does not alter the size
of the largest clique, these networks have only few topol-
ogy levels. Specifically, in the aggregates of tetrahedra
FIG. 6: Histograms of shortest distances d between pairs of
qmax = 3, and they can share nodes, links, and trian-
nodes (top) and average hyperbolicity hδi vs. dmin (bottom)
in three samples of networks for ν = -1, 0, +1, and +5. Net- gles as faces of lower orders; for triangles, qmax = 2
work size is above N = 500 nodes. and shared faces are links and nodes. Therefore, their
structure vectors are rather short. However, they pos-
sess a captivating structure of simplicial complexes, de-
pending on the chemical potential and geometry con-
straints. Consequently, the degree distributions are al-
D. Aggregation of monodisperse cliques tered by changing ν, as shown in Fig. 8. Notably, the
appearance of some scale-invariant structures is favored
In this section, we briefly consider the structures grown by the mutual attraction of cliques for ν > 0. The ag-
with the same aggregation rules but with mono-disperse gregation of tetrahedra more efficiently builds such struc-
7
4
10
4
an intrinsic global negative curvature; these features are
10
10
3
important for their practical use and functionality. Our
model with graph-based representation provides a better
Pc(k)
2
10
3
10 1
insight into the mechanisms that drive the assembly of
10
0
hierarchically organised networks with higher topological
10 1 2
complexity, which is a growing demand for technological
Pc(k)
i=-5 10 10
i=-1 k
2
10 i=0 applications.
i=1
i=5
1.28
1.76
1 2.39
10
3.51
5.74 IV. METHODS
1.90
0
10 1 2 3
10
k
10 10
Program flow for clique aggregation 39
FIG. 8: Cumulative distributions of the node’s degree in net- Algorithm 1 Program Flow: Growth of the graph by
works of aggregated mon-disperse cliques (main panel) tetra- attaching simplexes
hedra, and (inset) triangles, for different values of the chem- 1: INPUT: ν, smin , smax , Nmax
ical potential ν. Sample averaging and the number of nodes 2: initialise graph G as a simplex of size s taken from the
N ≥ 5000 applies. Thick broken and full lines indicate the distribution p(s) = As−2 , which is defined in the range
range where the slopes given in the legend are measured [smin , smax ]
(within the maximal error bars ±0.07) 3: while N < Nmax do
4: select new simplex size snew ∈ [smin , smax ] from the dis-
tribution p(s)
tures as compared with triangles. Whereas, the scale-free 5: for all simplexes σ ∈ G whose order qσ < snew − 1 do
range is limited with the exponential cut-offs in the case 6: compute pσ (snew , qσ ) = exp −ν(snew − qσ − 1)
of triangles unless ν is sufficiently large. Further analysis 7: end for P
8: normalise the probability such that σ pσ (snew , qσ ) = 1
of these and other networks of mono-disperse simplexes
9: select the docking site σ ∈ pσ (snew , qσ )
is left for future work. 10: form a new simplex σnew by attaching snew − qσ − 1 new
nodes to the qσ + 1 nodes of the docking simplex σ
11: sampling the data of interest;
III. CONCLUSIONS 12: end while
13: END
bined in three ways and ordered. For instance, first compute the matrix of distances between all pairs of
nodes; then, by sampling a large number of sets of nodes
d(A, B)+d(C, D) ≤ d(A, C)+d(B, D) ≤ d(A, D)+d(B, C). for the 4-point condition (2) we determine and plot the
average hδi against the corresponding distance dmin .
We denote the largest value L = d(A, D) + d(B, C), the Graphs visualisation We used gephi.org for graph
middle M = d(A, C) + d(B, D), smallest S = d(A, B) + presentation and community structure detection by max-
d(C, D), and the smallest pair distance of S as dmin = imum modularity method45 .
min{d(A, B), d(C, D)}. Then the graph is δ-hyperbolic
if there is a fixed value δ for which any four nodes of the
graph satisfy the 4-point condition: Acknowledgments
L−M
≤ δ. (2) The authors acknowledge the financial support from
2
the Slovenian Research Agency under the program P1-
There is a trivial upper bound (L−M)/2 ≤ dmin . Hence, 0044 and from the Ministry of Education, Science and
by plotting (L−M)/2 against dmin we can investigate the Technological Development of the Republic of Serbia, un-
worst case growth of the function. For a given graph, we der the projects OI 171037, III 41011 and OI 174014.
1
Pelaz, B., Jaber, S., Jimenez de Aberasturi, D., magnesium alloy clusters: a detailed study. Phys. Chem.
Wulf, V., Aida, T., de la Fuente, J.M. et al. The Chem. Phys., 18:26177–26183, 2016.
13
state of nanoparticle-based nanoscience and biotechnol- Krivovichev, S.V. Combinatorial topology of salts of inor-
ogy: Progress, promises, and challenges. ACS Nano, ganic oxoacids: zero-, one- and two-dimensional units with
6(10):8468–8483, 2012. PMID: 23016700. corner-sharing between coordination polyhedra. Crystal-
2
Whitesides, G.M. & Grzybowski, B. Self-assembly at all lography Reviews, 10(3):185–232, 2004.
14
scales. Science, 295(5564):2418–2421, 2002. Senyuk, B., Liu, Q., Bililign, E., Nystrom, P.D. & Sma-
3
Boles, M.A., Engel, M. & Talapin, D.V. Self-assembly of lyukh, I.I. Geometry-guided colloidal interactions and self-
colloidal nanocrystals: From intricate structures to func- tiling of elastic dipoles formed by truncated pyramid parti-
tional materials. Chemical Reviews, 116(18):11220–11289, cles in liquid crystals. Phys. Rev. E, 91:040501, Apr 2015.
15
2016. PMID: 27552640. Ikeda, S. & Kotani, M. Materials inspired by mathematics.
4
Meledandri, C.J., Stolarczyk, J.K. & Brougham, D.F. Hi- Science and Technology of Advanced Materials, 17(1):253–
erarchical gold-decorated magnetic nanoparticle clusters 259, 2016.
16
with controlled size. ACS Nano, 5(3):1747–1755, 2011. Živković, J. & Tadić, B. Nanonetworks: The graph the-
PMID: 21309572. ory framework for modeling nanoscale systems. Nanoscale
5
Wang, L.. Xu, L., Kuang, H., Xu, C. & Kotov, N.A. Dy- Systems MMTA, 2, 2013.
17
namic nanoparticle assemblies. Accounts of Chemical Re- Blunt, M.O., Šuvakov, M., Pulizzi, F., Martin, C.P.,
search, 45(11):1916–1926, 2012. PMID: 22449243. Pauliac-Vaujour, E., Stannard, A., Rushforth, A.W.,
6
Luo, D., Yan, C. & Wang, T. Interparticle forces underly- Tadić, B. & Moriarty, P. Charge transport in cellu-
ing nanoparticle self-assemblies. Small, 11(45):5984–6008, lar nanoparticle networks: meandering through nanoscale
2015. mazes. Nano Letters, 7(4):855–860, 2007. PMID:
7
Liu, S. & Yu, J. Cooperative self-construction and en- 17335264.
18
hanced optical absorption of nanoplates-assembled hierar- Šuvakov, M. & Tadić, B. Modeling collective charge trans-
chical bi2wo6 flowers. Journal of Solid State Chemistry, port in nanoparticle assemblies. Journal of Physics: Con-
181(5):1048 – 1055, 2008. densed Matter, 22(16):163201, 2010.
8 19
Wang, R., Sha, T., Dreyfus, R., Leunissen, M.E., Maas, Tadić, B., Andjelković, M. & Šuvakov, M. The influence of
C., Pine, D., Chaikin, P.M., & Seeman, N.C.. Self- architecture of nanoparticle networks on collective charge
replication of information-bearing nanoscale patterns. Na- transport revealed by the fractal time series and topology
ture, 478(7365):225–229, 2011. of phase space manifolds. Journal of Coupled Systems and
9
Rossi, L., Sacanna, S., Irvine, W.T.M., Chaikin, P.M., Multiscale Dynamics, 4(1), 30-42, 2016.
20
Pine, D.J. & Philipse, A.P. Cubic crystals from cubic col- Hirata, A., Kang, L. J., Fujita, T., Klumov, B., Matsue,
loids. Soft Matter, 7:4139–4142, 2011. K., Kotani, M., Yavari, A. R. & Chen, M. W. Geomet-
10
Gu, Y., Burtovyy, R., Townsend, J., Owens, J.R., Luzinov, ric frustration of icosahedron in metallic glasses. Science,
I. & Kornev, K.G. Collective alignment of nanorods in thin 10.1126/science.1232450, 2013.
21
newtonian films. Soft Matter, 9:8532–8539, 2013. Kotani M. & Ikeda S. A new direction in mathematics for
11
Toulemon, D., Rastei, M.V., Schmool, D., Garitaonandia, materials science. in Springer Briefs in the Mathematics of
J.S., Lezama, L., Cattoen, X., Begin-Collin, S. & Pichon, Materials, Vol.I, 2015, Springer, Tokyo
22
B.P.. Enhanced collective magnetic properties induced Šuvakov, M. & Tadić, B. Topology of Cell-Aggregated Pla-
by the controlled assembly of iron oxide nanoparticles in nar Graphs, pages 1098–1105. Springer Berlin Heidelberg,
chains. Adv. Funct. Mat., 26(15):1616–3028, 2016. Berlin, Heidelberg, 2006.
12 23
Xing, X., Wang, J., Kuang, X., Xia, X., Lu, C. & Maroulis, Tesoro, S., Göpfrich, K., Kartanas, T., Keyser, U. F. &
G. Probing the low-energy structures of aluminum- Ahnert,S. E. Nondeterministic self-assembly with asym-
9
37
metric interactions. Phys. Rev. E, 94:022404, 2016. Albert, R., DasGupta, B. & Mobasheri, N. Topological
24
Bianconi, G. & Rahmede, C. Emergent hyperbolic network implications of negative curvature for biological and social
geometry. Sci. Rep., 7:41974, 2017. networks. Phys. Rev. E, 89:032811, 2014.
25 38
Krioukov, D., Papadopoulos, F., Kitsak, M., Vahdat, A. Narayan, O. & Saniee, I. Large-scale curvature of networks.
& Boguñá, M. Hyperbolic geometry of complex networks. Phys. Rev. E, 84:066108, 2011.
39
Phys. Rev. E, 82:036106, Sep 2010. Šuvakov, M., Andjelković, M. & Tadić, B. Applet: Simplex
26
Šuvakov, M. & Tadić, B. Transport processes on homo- aggregated growing graph (http://suki.ipb.rs/ggraph/).
geneous planar graphs with scale-free loops. Physica A: 2017.
40
Statistical Mechanics and its Applications, 372(2):354 – Atkin, R.H. An algebra for patterns on a complex, II.
361, 2006. Nonlinearity, Nonequilibrium and Complexity: International Journal of Man-Machine Studies, 8(5):483 –
Questions and Perspectives in Statistical Physics. 498, 1976.
27 41
Delphine Toulemon, D., Mircea V. Rastei, M.V., David Johnson, J. Some structures and notation of q-analysis. in
Schmool, D., Jos Siz Garitaonandia, J.S., Lezama, L., Environment and Planning B: Planning and Design, 8:73–
Cattoën, X., Bégin-Colin, S. & Pichon, B.P. Enhanced 86, 1981.
42
collective magnetic properties induced by the controlled Beaumont, J.R. & Gatrell, A.C. An Introduction to Q-
assembly of iron oxide nanoparticles in chains. Advanced Analysis. Geo Abstracts, Norwich-Printed by Edmund
Functional Materials, 26(15):2454–2462, 2016. Nome Press, Norwich, 1982.
28 43
Bollobás, B. Modern Graph Theory. Springer, New York, Bianconi, G., Rahmede, C. & Wu, Z. Complex quan-
inc., 1998. tum network geometries: Evolution and phase transitions.
29
Dorogovtsev, S.N. Lectures on Complex Networks. Oxford Phys. Rev. E, 92:022815, 2015.
44
Master Series in Statistical, Computational, and Theoret- Bron, C. & Kerbosch, J. Finding all cliques of an undi-
ical Physics, Oxford University Press, 2010. rected graph. Comm ACM, 16, 575-577, 1973.
30 45
Kozlov, D. Combinatorial Algebraic Topology. Springer Blondel, V.D., Guillaume, J.L., Lambiotte, R. & Lefeb-
Series Algorithms and Computation in Mathematics , Vol. vre, E. Fast unfolding of communities in large networks.
21, Springer-Verlag Berlin Heidelberg, 2008. Journal of Statistical Mechanics: Theory and Experiment,
31
Jonsson, J. Simplicial Complexes of Graphs. Lecture Notes 2008(10):P10008 (12pp), 2008.
in Mathematics, Springer-Verlag, Berlin, 2008.
32
Miroslav Andjelković, Bosiljka Tadić, Slobodan Maletić,
and Milan Rajković. Hierarchical sequencing of online so- ν = −1 ν=0 ν = +1
cial graphs. Physica A: Statistical Mechanics and its Ap- q Qq nq Q̂q Qq nq Q̂q Qq nq Q̂q
plications, 436:582 – 595, 2015. 0 1 453 0.997 1 632 0.998 1 815 0.999
33
Andjelković, M., Gupte, N. & Tadić, B. Hidden geometry 1 308 453 0.320 367 632 0.419 411 815 0.495
of traffic jamming. Phys. Rev. E, 91:052817, 2015. 2 227 259 0.124 203 330 0.385 166 442 0.624
34
Tadić, B., Andjelković, M., Boshkoska, B.M. & Levnajić, 3 149 157 0.051 124 194 0.361 126 300 0.580
Z. Algebraic topology of multi-brain connectivity networks 4 110 111 0.009 98 134 0.269 83 196 0.576
reveals dissimilarity in functional patterns during spoken 5 76 76 0.000 84 94 0.106 65 125 0.480
communications. PLOS ONE, 11(11):1–25, 11 2016. 6 63 63 0.000 56 59 0.051 58 86 0.325
35
Andjelković, M., Tadić, B., Mitrović Dankulov, M., Ra- 7 36 36 0.000 34 35 0.029 43 54 0.204
jković, M. & Melnik, R. Topology of innovation spaces in 8 20 20 0.000 22 22 0.000 30 32 0.063
the knowledge networks emerging through questions-and- 9 11 11 0.000 11 11 0.000 17 17 0.000
answers. PLOS ONE, 11(5):1–17, 05 2016.
36
Kennedy, W. S., Saniee, I. & Narayan, O. On the hyper- TABLE I: The components of the three structure vectors for
bolicity of large-scale networks and its estimation. In 2016 the networks generated at different chemical potential ν.
IEEE International Conference on Big Data (Big Data),
pages 3344–3351, Dec 2016.