1: \documentclass[aps,twocolumn,showpacs,pre]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{graphics}
4: \usepackage{amsmath}
5: \usepackage{amssymb}
6: \usepackage{array}
7: \usepackage{longtable}
8: \usepackage{pstricks}
9: \usepackage{amsfonts}
10: %\usepackage{hyperref}
11: \parindent0ex
12: \clubpenalty = 90000
13: \widowpenalty = 90000
14: \displaywidowpenalty = 90000
15: \setlongtables
16: \setlength{\LTcapwidth}{8.5cm}
17: \begin{document}
18: \title{Spectral Analysis of Protein-Protein Interactions in {\em Drosophila melanogaster}}
19: \author{Christel Kamp}
20: \email{mail@christelkamp.de,
21: c.kamp@imperial.ac.uk}
22: \affiliation{Blackett Laboratory, Imperial College, Prince Consort Road,
23: London SW7 2AZ, United Kingdom}
24: \author{Kim Christensen}
25: \email{k.christensen@imperial.ac.uk}
26: \affiliation{Blackett Laboratory, Imperial College, Prince Consort Road,
27: London SW7 2AZ, United Kingdom}
28: \affiliation{Physics of Geological Processes, University of Oslo, PO Box 1048, Blindern, N-0316 Oslo, Norway}
29:
30: \begin{abstract}
31: Within a case study on the protein-protein interaction network (PIN) of {\em Drosophila melanogaster} we investigate the relation between the network's spectral properties and its structural features such as the prevalence of specific subgraphs or duplicate nodes as a result of its evolutionary history.
32: The discrete part of the spectral density shows fingerprints of the PIN's topological features including a preference for loop structures. Duplicate nodes are another prominent feature of PINs and we discuss their representation in the PIN's spectrum as well as their biological implications.
33: \end{abstract}
34: \pacs{89.75.-k, 89.20.-a, 89.75.Hc, 89.75.Fb, 87.16.Yc, 87.16.-b, 87.10.+e, 02.50.Fz}
35: %more possible pacs
36: %87.16.Ac, 87.23.-n, 02.60.-x, 02.50.Ng
37: \maketitle
38: \section{Introduction}
39: Network structures can be observed in most diverse domains ranging from biological and technological systems to social or economical systems \cite{strogatz:2001}. Genetic regulatory networks, protein-protein interaction networks and metabolic networks support the functions of life in any living organism. Technological networks such as the internet or the World Wide Web have a huge impact on our lives and societies. Networks of acquaintances and the exchange of information within these networks shape social and economical systems.
40: Considering the omnipresence of networks, their
41: investigation has a long tradition in
42: graph theory \cite{bollobas:modernbook,bollobas:randomgraphs}.
43: However, during the last few years high quality data on real-world networks
44: has revealed that they cannot be adequately described by standard models from random graph theory and the topic has attracted growing interest.
45: Still, much attention has been devoted to the derivation of rather
46: specific quantities like degree distributions or
47: clustering coefficients \cite{albert:barabasi:2002} that do not allow for
48: a classification and understanding of network topologies within a broader and self-consistent framework.
49: \newline
50: Making an attempt towards a more comprehensive description
51: spectral graph
52: theory \cite{chung:book,cvetkovic:book,cvetkovic:annals,biggs:book}
53: can be considered as one promising ansatz.
54: A network of $N$ nodes
55: can be described by its adjacency matrix ${\bf A}=(a_{ij})$ with
56: entries
57: \begin{equation}
58: a_{ij}=\left\{
59: \begin{array}{ll}
60: 1& \mbox{if there is a link between node $i$ and $j$}\\
61: 0 & \mbox{otherwise}.
62: \end{array}\right.
63: \end{equation}
64: The adjacency matrix is
65: a symmetric, non-negative matrix in the case of undirected networks
66: and accordingly has real eigenvalues $\{\lambda_j\}$, $j=1,...,N$, being solutions of $\det({\bf A}-\lambda{\bf I})=0$.
67: The relation between features of a network and properties of its spectral
68: density
69: \begin{eqnarray}
70: \rho(\lambda)&=&\frac{1}{N}\sum_{j=1}^N\delta(\lambda-\lambda_j)
71: \end{eqnarray}
72: with respect to its adjacency matrix
73: is a topic of current research. While dense classical random networks
74: exhibit a semi-circular spectral density
75: of the adjacency matrix \cite{bauer:golinelli:2001},
76: networks with a broad or scale free degree distribution give rise to
77: a broader spectrum \cite{farkas:vicsek:2001,dorogovtsev:samukhin:2003,chung:vu:2002,chung:vu:2003,mihail:papadimitriou:2002,goh:kim:2001}.
78: A striking feature of sparse random networks' spectral density
79: is the emergence of peaks at eigenvalues of finite
80: trees \cite{bauer:golinelli:2001}, similar to those found in
81: large random trees \cite{golinelli:2003}, due to the strong prevalence
82: of these subgraphs \footnote{A similar phenomenon has earlier been discussed in the context of a model of quantum percolation \cite{kirkpatrick:eggarter:1972,evangelou:1983}.}.
83: Here we address whether these findings are applicable more generally, that is
84: whether peaks in the spectral density of sparse random networks can be
85: associated to a strong prevalence of specific subgraphs.
86: The search for subgraphs that are statistically
87: overrepresented relative to a null-model, so-called motifs,
88: recently gained much attention \cite{milo:alon:2002,milo:alon:2004}.
89: As a case study on the relation between these two approaches,
90: we investigate the spectral properties of the protein-protein interaction
91: network (PIN) of the fruit fly {\em Drosophila melanogaster}
92: \cite{giot:rothberg:2003}.
93: While no simple correspondence between network motifs and
94: a network's spectral proprieties can be derived, on a more abstract level,
95: we infer from the PIN's spectrum a prevalence of loop structures.
96: Furthermore, some properties specific to
97: a network that has evolved by duplication of nodes are studied and discussed
98: within the context of spectral analysis.
99: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
100: \section{The spectrum of the PIN of {\em Drosophila melanogaster}}
101: For our study we used the PIN
102: of {\em Drosophila melanogaster} as given in \cite{giot:rothberg:2003}
103: and available via the Database of Interacting Proteins \cite{dip}.
104: \begin{figure}[!h]
105: \begin{pspicture}(0,0)(5,6)%\showgrid
106: \rput(2.75,3.2){\scalebox{0.33}{\includegraphics{./Fig1.eps}}}
107: \fontsize{12pt}{12pt}
108: \rput(3,0.0){Minimal confidence value of interaction}
109: \end{pspicture}
110: \caption{The number of nodes divided by 10000 (solid line) and the
111: fraction of nodes (dashed line) in the largest connected component in the
112: PIN as a function of the minimal confidence value of protein-protein
113: interaction. We focus on the PIN defined by a minimal confidence value of $0.5$, see dashed line.
114: } \label{F:Fig1}
115: %\end{figure}
116: %\begin{figure}[!b]
117: \begin{pspicture}(0,0)(5,12.5)
118: \rput(2.75,9.5){\scalebox{0.33}{\includegraphics{./Fig2b.eps}}}
119: \rput(2.75,3.5){\scalebox{0.33}{\includegraphics{./Fig2a.eps}}}
120: \fontsize{12pt}{12pt}
121: \rput(2.75,0.25){Eigenvalue}
122: \rput{90}(-1.65,9.5){Cumulative distribution of eigenvalues}
123: \rput{90}(-1.65,3.5){Frequency of eigenvalue}
124: \rput(-0.1,11.6){(a)}
125: \rput(-0.1,5.6){(b)}
126: \end{pspicture}
127: \caption{Spectral analysis of the protein-protein interaction
128: network of {\em Drosophila melanogaster}. (a) The cumulative
129: spectral density. (b) The discrete
130: frequency spectrum containing 49\% of all eigenvalues.
131: } \label{F:Fig2}
132: \end{figure}
133: The protein-protein interactions have been derived using the
134: two-hybrid method which, however, is known to generate many false positives.
135: Therefore each interaction in the network is classified by a confidence
136: value between zero and one defining a hierarchy of networks with
137: increasing minimal confidence value for the protein-protein interactions.
138: In Fig. \ref{F:Fig1} the size of the largest connected component in a network
139: with a given minimal confidence value of interactions is shown.
140: \newline
141: For our further analysis we choose a network with a minimal confidence
142: value of $0.5$ which contains $4681$ proteins and $4794$ interactions
143: corresponding to an average degree $\langle k\rangle=2.05$.
144: The network is enriched with biologically meaningful
145: interactions while it still shows a strong largest connected component (i.e. a giant component) containing
146: about $2/3$ of its nodes.
147: \newline
148: We determined the eigenvalues of the adjacency matrix corresponding to
149: this PIN. The cumulative
150: spectral density in Fig. \ref{F:Fig2} (a) exhibits
151: jumps at various eigenvalues which are represented by the discrete
152: spectrum in Fig. \ref{F:Fig2} (b) \footnote{An eigenvalue is considered
153: to belong to the discrete spectrum if there exists at least one
154: other eigenvalue that does not deviate more than $10^{-12}$.}.
155: Since about $2/3$ of the network's nodes belong to its
156: giant component and $49\%$ of the eigenvalues in the network's spectrum
157: are in the discrete spectrum, the emergence of spectral peaks cannot
158: be explained by small isolated clusters alone.
159: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
160: \subsection{The discrete spectrum and network motifs}\label{motifsec}
161: To get a better understanding of the emergence of spectral peaks
162: we compare the discrete spectrum with the corresponding spectra
163: of two reference networks. First, we look at a network of the same size
164: and degree
165: sequence but randomized links following the procedure of
166: \cite{maslov:sneppen:2002} (a randomized PIN). Second,
167: we consider a classical random network of the same size
168: and average degree
169: $\langle k\rangle=2.05$ (a random network), that is a
170: network with a probability $p=0.000438$ for a link between any two nodes.
171: In Fig. \ref{F:Fig3} the discrete spectrum of the adjacency matrix of the
172: protein-protein interaction network of {\em Drosophila melanogaster} as well
173: as of the two reference networks are shown, the latter being averages over $10$ reference networks.
174: \begin{figure}[!h]
175: \begin{pspicture}(0,0)(5,18.5)%\showgrid
176: \rput(2.75,15.5){\scalebox{0.33}{\includegraphics{./Fig3a.eps}}}
177: \rput(2.75,9.5){\scalebox{0.33}{\includegraphics{./Fig3b.eps}}}
178: \rput(2.75,3.5){\scalebox{0.33}{\includegraphics{./Fig3c.eps}}}
179: \fontsize{12pt}{12pt}
180: \rput(2.75,0.25){Eigenvalue}
181: \rput{90}(-1.65,15.5){Frequency of eigenvalue}
182: \rput{90}(-1.65,9.5){Frequency of eigenvalue}
183: \rput{90}(-1.65,3.6){Frequency of eigenvalue}
184: \rput(-0.2,17.7){(a)}
185: \rput(-0.2,11.7){(b)}
186: \rput(-0.2,5.7){(c)}
187: \end{pspicture}
188: \caption{The discrete frequency spectrum of (a) the PIN
189: of {\em Drosophila melanogaster} containing 49\% of all eigenvalues,
190: (b) a randomized PIN with identical degrees at each node containing 43\% of all eigenvalues,
191: and (c) a classical random network of identical size and average degree $\langle k \rangle = 2.05$
192: containing 27\% of all eigenvalues.
193: } \label{F:Fig3}
194: \end{figure}
195:
196: \begin{figure}[b]
197: \centerline{ \includegraphics[width=6.6cm,angle=0]{graphtab4.eps}}
198: \caption{\label{graphs1} Connected subgraphs with up to 4 nodes: A bullet ($\bullet$) in the three middle columns denotes that the eigenvalues of this graph can be found in the spectrum of the original network (PIN), the randomized network (Rand. PIN) or the random network (Rand. network), respectively. The rightmost column shows whether the subgraph is a motif according to the {\tt mfinder} software (default settings, \cite{milo:alon:2002,mfinder}). White bullets ($\circ$) correspond to single eigenvalue occurrences. }
199: \end{figure}
200: \begin{figure}[!h]
201: \centerline{ \includegraphics[width=6.6cm,angle=0]{graphtab5.eps}}
202: \caption{\label{graphs2}Connected subgraphs with 5 nodes: A bullet ($\bullet$) in the three middle columns denotes that the eigenvalues of this graph can be found in the spectrum of the original network (PIN), the randomized network (Rand. PIN) or the random network (Rand. network), respectively. The rightmost column shows whether the subgraph is a motif according to the {\tt mfinder} software (default settings, \cite{milo:alon:2002,mfinder}). White bullets ($\circ$) correspond to single eigenvalue occurrences. }
203: \end{figure}
204:
205: To get more reliable results, we concentrate our further analysis only on
206: eigenvalues that can be found more than twice in the spectrum of the original
207: network.
208: Qualitatively, we see that while the classical random network shows only a few peaks of that size
209: corresponding to the eigenvalues of simple tree-graphs (2.1, 3.1, 4.1, 4.2. in
210: Fig. \ref{graphs1}), additional eigenvalues appear in the discrete spectrum
211: of the randomized PIN with the same degree sequence as the original network
212: and eventually the original network (see Fig. \ref{F:Fig3}).
213: This change in the spectral properties indicates some
214: differences in the structural organization of the underlying networks.
215: In the following paragraphs, we will discuss how the observed hierarchy of spectral peaks reflects the networks' topologies and relates to other concepts like the search for motifs.
216: \newline
217: Following the arguments of \cite{golinelli:2003}, we suggest that
218: the prevalence of specific peaks in the discrete spectrum of a network
219: corresponds to a strong representation of certain subgraphs.
220: It has recently been shown that networks from different contexts show
221: characteristic overrepresentation of
222: specific subnetworks which are usually referred to as
223: motifs \cite{milo:alon:2002,milo:alon:2004}.
224: Although motifs can be expected to leave marks in a network's spectrum,
225: there is seemingly no simple correspondence between the eigenvalues of
226: small subgraphs and spectral peaks. First, subgraphs are not generally
227: represented by their eigenvalues in the spectrum of the whole network.
228: Second, isospectral graphs are not necessarily isomorphic
229: \cite{spectraofgraphs:book}.
230: Nevertheless,
231: a thorough comparative study of the discrete spectrum can provide some
232: insight into the networks' structure. In Figs. \ref{graphs1} and \ref{graphs2} we show the
233: connected subgraphs up to size $5$ with the full set of eigenvalues present
234: in the discrete spectrum of the whole network. It shows that the
235: spectrum of the PIN is more
236: consistent with loop-structures (cf. graphs 3.2, 4.3, 5.4, 5.5, 5.6,
237: 5.8) than any randomized version
238: which might hint to regulatory functionality supplied by this network.
239: The eigenvalues behind these structures might
240: correspond to eigenvalues of trees, e.g., the eigenvalues of a triangle
241: (graph 3.2) or the box (graph 4.3, often also referred to as bi-fan structure)
242: might well be explained by graphs 2.1 and 5.3. However, to represent
243: the eigenvalues of graphs 5.5 and 5.6 one has to consider trees of minimum
244: size 8 and 7, respectively. The eigenvalues of graph 5.8 cannot be found
245: among trees of size up to 10. Considering that the frequency of a given tree of size $n$ in a sparse network decreases exponentially with $n$
246: \cite{bauer:golinelli:2001}
247: and relating the findings in the PIN
248: to those in the randomized reference networks we hypothesize that
249: the spectral peculiarities reflect the loop structure in the original network.
250: \newline
251: To quantify the correspondence between the number of specific subgraphs in the PIN and the PIN's discrete spectrum we tried to decompose the spectrum into the contributions of connected subgraphs up to size $5$. This, however, was not feasible indicating that higher order contributions, though being individually small, cannot be neglected as a whole.
252: \newline
253: Although we have to ascertain that there is no simple correspondence between
254: subgraphs of a network and the prevalence of their eigenvalues in the discrete
255: spectrum of the whole network we want to discuss the relation of spectral properties to the notion of motifs. According to the definition introduced in Ref. \cite{milo:alon:2002}, a motif is a subnetwork that shows strong prevalence within the network relative to a randomized network. For our analysis we refer to the default requirements implemented in the {\tt mfinder} software \cite{milo:alon:2002,mfinder}, that is a motif is a subgraph that occurs
256: at least by two standard deviations more than in 100 randomized networks with
257: the same degree sequence.
258: In Figs. \ref{graphs1} and \ref{graphs2} the rightmost columns show
259: which connected subgraphs up to size $5$ are motifs in the PIN according to these criteria.
260: There exist a lot of highly connected motifs
261: while the spectrum reflects more the tree-structures in the network.
262: However, the fingerprints of trees in the spectrum of both the
263: original and the randomized PIN are consistent with the
264: fact that they do not show up as motifs according to the above definition.
265: One might further speculate
266: whether some motifs are hidden for spectral analysis because they are
267: in fact building blocks of larger units. For example graph 5.5 as well as
268: its subgraph 4.4 is a motif according to \cite{milo:alon:2002,mfinder}.
269: But only the eigenvalues of 5.5 can be found in the spectrum of the PIN.
270: Moreover highly connected motifs do not occur in high (absolute) numbers
271: and might accordingly be drowned in spectral analysis.
272: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
273: \subsection{The circuitry of the PIN}
274: In section \ref{motifsec} we have shown that the discrete spectrum of the PIN of {\em Drosophila melanogaster} favors the eigenvalues of loopy subgraphs.
275: This observation derived from the investigation of distinct local structures and their eigenvalue representations can be confirmed by an assessment of the whole set of eigenvalues.
276: Evaluating the trace of the matrix ${\bf A}^k$
277: \begin{equation}\label{trace}
278: Tr({\bf A}^k)=\sum_{i=1}^N \lambda_i^k
279: \end{equation}
280: yields the number of directed loops of length $k$
281: in the underlying network \cite{cvetkovic:book,farkas:vicsek:2001} as shown in Fig. \ref{F:Fig6}, though neglecting details of the graphs underlying the loops. Note, that even loops might be trivial going back and forth in a tree while odd loops are non-trivial.
282: \begin{figure}
283: \begin{pspicture}(0,0)(5,12.5)
284: \rput(2.75,9.5){\scalebox{0.33}{\includegraphics{./Fig6a.eps}}}
285: \rput(2.75,3.5){\scalebox{0.33}{\includegraphics{./Fig6b.eps}}}
286: \fontsize{12pt}{12pt}
287: %\rput(2.75,6.4){Loop size {\small $k$}}
288: \rput(2.75,0.25){Loop size {\small $k$}}
289: \rput{90}(-1.65,9.6){Frequency of loop}
290: \rput{90}(-1.65,3.6){Relative frequency of loop}
291: \rput(-0.2,11.7){(a)}
292: \rput(-0.2,5.7){(b)}
293: \end{pspicture}
294: \caption{(a) The frequency of loops of size $k$ in the PIN of {\em Drosophila melanogaster} (solid line),
295: a randomized PIN with identical degrees at each node (dashed line) and
296: a classical random network of identical size and average degree $\langle k \rangle = 2.05$ (dotted line).
297: Odd cycles represents non-trivial loops, that is, deviations from a tree-like structure
298: in the networks.
299: (b) The relative frequency of loops of size $k$ in the PIN of {\em Drosophila melanogaster}
300: with respect to the randomized PIN (dashed line) and the classical random network (dotted line).
301: } \label{F:Fig6}
302: \end{figure}
303: The difference in growth rates of the numbers of loops of growing size between the original network and the classical random network is likely due to the strong fragmentation of the latter one (many isolated nodes). However, the strong relative prevalence of loops of odd length in the original network is more remarkable with respect to the networks' topologies. This becomes more obvious from
304: Fig. \ref{F:Fig6} (b) showing the number of loops of a given size in the original PIN normalized to the numbers in the two reference networks. While tree graphs only have trivial loops of even length, loops of odd length indicate non-trivial loops which confirms the results derived from the evaluation of the discrete spectrum on the basis of eigenvalue representations of small subgraphs.
305: \newline
306: The analysis of a network's discrete spectrum could reveal some structural information about the network as a whole. This information is less specific than an analysis in terms of motifs, that is only conclusions about more general properties like the prevalence of loops are possible instead of exact motif counts. However, it should be emphasized that spectral analysis is not hampered by an a priori {\em bias} towards predefined
307: quantities like motifs of a given size.
308: It is a challenging question of future research to investigate the relationship between a network's spectrum and its topological features, e.g. in terms of motifs, in more detail to get a more rigorous and {\em unbiased} characterization of a network's topological features.
309: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
310: \section{Fingerprints of duplication}\label{dup}
311: The evolution of many biological networks and specifically PINs
312: is assumed to be strongly driven by duplication (and diversification)
313: of nodes in the network \cite{ohno:book,teichmann:babu:2004}.
314: The genomes underlying the PIN of many organisms have undergone a few whole genome duplications complemented by many single-gene duplications \cite{amoutzias:bauer:2004}. After duplication, one of the duplicates usually diverges from its original appearance, possibly providing new functionality.
315: The concept of duplication has similarly been recognized to be important for
316: functional roles in a network motif \cite{kashtan:alon:2004}.
317: The search for fingerprints of the evolutionary history of a PIN naturally has to include an assessment of duplicate nodes, that is those that share the same interaction partners. Each set of duplicate nodes represents an equivalence class also referred to as an orbit. The reduced network is a network in which all nodes of an orbit are reduced to one node.
318: \begin{table}[h]
319: \begin{tabular}{l|l|l}
320: & Duplicate nodes & Duplicate links\\
321: \hline
322: PIN {\em D. melanogaster} & $686$ & $728$\\
323: Randomized PIN & $626.0\pm 22.0$ & $629.1\pm 22.0$\\
324: Random network & $151.0\pm14.2$ & $151.0 \pm 14.2$\\
325: \end{tabular}
326: \caption{\label{duptab} The table shows the number of nodes
327: that are duplicates (duplicate nodes) and the
328: number of neighbors associated to these nodes (duplicate links) in the original network and the two reference networks, that is the difference in the number of nodes and links between the original network and the reduced network. Isolated nodes have been neglected.}
329: \end{table}
330:
331: Tab. \ref{duptab}
332: shows that the PIN has more duplicate
333: nodes (with associated links) than the reference networks.
334: Fig. \ref{F:Fig8} shows the
335: frequency of orbits of a given size in the original as well as in
336: the reference networks. The distribution of orbit sizes in the original
337: network is very close to the one found in the randomized
338: PIN with the same degree sequence, but much broader than
339: that of the classical random network.
340: \begin{figure}[!h]
341: \begin{pspicture}(0,0)(5,6.5)%\showgrid
342: \rput(2.75,3.5){\scalebox{0.33}{\includegraphics{./Fig8.eps}}}
343: \fontsize{12pt}{12pt}
344: \rput(3,0.25){Orbit size}
345: \rput{90}(-1.65,3.6){Frequency of orbit}
346: \end{pspicture}
347: \caption{The frequency of orbit size in the PIN of {\em Drosophila melanogaster} (solid line),
348: a randomized PIN with identical degrees at each node (dashed line) and
349: a classical random network of identical size and average degree $\langle k \rangle = 2.05$ (dotted line).
350: Isolated nodes have been neglected.
351: } \label{F:Fig8}
352: \end{figure}
353:
354: Again, spectral analysis offers a complementary approach to the topic.
355: Using the results in Appendix A, we determined the eigenvalues
356: of those graphs that arise from duplication of the two simplest reduced
357: graphs: a line and a triangle (graphs 2.1 and 3.2 in Fig. \ref{graphs1}).
358: We allowed for up to ten duplications of each
359: node of the reduced network and searched for the eigenvalues of the
360: resulting subgraphs. However, spectral analysis is only consistent
361: with the emergence of star graphs and the original triangle as
362: well as the box or bi-fan structure (graph 4.3 in Fig. \ref{graphs1}).
363: \newline
364: Considering the representation of star graphs in the
365: spectrum of the PIN one might guess
366: that the high frequency of large orbits mainly reflects nodes with many leaves.
367: A look at the joint distribution of the size of an orbit and the degree
368: of its nodes in the original and the reference networks supports this
369: hypothesis. In 100 reference networks (of both kind)
370: the nodes in an orbit larger than one have degree one,
371: that is only nodes with degree one have duplicates.
372: Only in extremely rare cases do nodes with degree two have a single duplicate.
373: \begin{figure}[t]
374: \includegraphics[width=8cm,angle=0]{dupgraphs.eps}
375: \caption{\label{dupgraphs}Subgraphs of nodes that form orbits of
376: size $\geq 2$ and that have
377: degree $\geq 2$ (black), orbits of size 2 of nodes with degree 2 have been omitted. The white nodes are the neighbors of duplicate nodes that may have
378: more neighbors than shown. The nodes' labels are their identifiers from the Database of Interacting Proteins \cite{dip}, cf. Appendix B for more details.}
379: \end{figure}
380: \newline
381: This matches the global situation in the original network, however, there are
382: some remarkable exceptions with nodes of high degree
383: in large orbits
384: shown in Fig. \ref{dupgraphs} that cannot be found in the reference networks.
385: This is also well in accordance with the values found in Tab. \ref{duptab}.
386: Different from the original network, in both reference networks the number of duplicate links is practically the same as the number of duplicate nodes.
387: \newline
388: From the Database of Interacting Proteins \cite{dip} and FlyBase \cite{flybase} we derived names and descriptions (if available) for the proteins in Fig. \ref{dupgraphs} as shown in Appendix B.
389: We find that duplicate proteins are likely to have similar functionality
390: in accordance with results in the yeast PIN \cite{samanta:liang:2003}.
391: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
392: \section{Summary and conclusions}
393: Recent developments in the research on complex networks have brought up a better understanding of a network's topology and its connection to functionality. However, a comprehensive theory of networks incorporating classical graph theory as well as recent findings into a self-consistent framework has still to be worked out. Considering spectral graph theory to be a promising ansatz for this attempt, we have done a case study on the PIN of {\em Drosophila melanogaster}. The eigenvalues of a network's adjacency matrix (and of related matrices) provide information about a network's structural properties like the number of connected components, its diameter or characteristics of its degree distribution. Here, we have put special emphasis on the investigation of the discrete spectrum of a sparse network relating it to prevalent substructures. Although it will probably not be possible to derive the densities of specific subgraphs from the spectrum of a network we could show that structural prevalences on a more abstract level are reflected in the (discrete) spectrum of the PIN under investigation. While we here focused on the appearance of loops in subgraphs as well as the whole PIN future analysis might reveal further topological features.
394: \newline
395: Considering the evolutionary history of PINs we also discussed the appearance of proteins that share their neighbors together with the fingerprints of these structures that can be found in the network's spectrum. Studying structures of duplicate proteins in more details we find that they often have close functional relationships in accordance with earlier findings in yeast.
396: \newline
397: The requirement applied here for the members of an orbit
398: to show exactly the same
399: neighborhood is very restrictive, though required to allow for
400: transitivity. This might be generalized by the definition of a
401: similarity measure that quantifies the overlap of the neighborhoods
402: of two nodes. This similarity measure can be defined
403: as a distance measure between nodes and the application of
404: a clustering algorithm in the associated metric space might
405: give further insight into local structures.
406: \newline
407: This case study shows that a more systematic assessment of the relation between a network's spectral and topological properties has to be a topic of future research. It is a challenging task, however, it can bring important insight into a network's structure in a less biased and more systematic way than currently available.
408: \vspace*{0.3cm}
409: \newline
410: {\bf Acknowledgments:} C. Kamp would like to thank A. Bunten for providing computing facilities and software and for many helpful discussions on technical problems. I am as grateful to N. Farid, S.A. Teichmann and J. Leal for the discussions and many helpful comments on the manuscript. Also, we would like to acknowledge the hospitality of the department of physics of the university of Oslo during the period of finalizing this manuscript. This work was supported by a fellowship within the Postdoc-Programme of the German Academic Exchange Service (DAAD).
411: \section*{Appendix A}
412: Let {$\bf A$} be a $N\times N$ matrix representing an undirected graph,
413: i.e. a symmetric matrix with entries $a_{ij}\in\{0,1\}$ and $a_{ii}=0$.
414: Let {$\bf D$ } be the matrix that is
415: obtained after $m$ perfect duplications of nodes or in other words
416: by $m$ duplications of rows and columns, respectively.
417: Let $i_1$,...,$i_l$, $l\in\{1,...,N\}$ the number (identifier)
418: of (mutually different) nodes that have
419: been duplicated and $m_{i_1}$,...,$m_{i_l}$ be the corresponding
420: number of duplications per node with
421: $m=\sum_{j=1}^l m_{i_j}$.
422: Let ${\bf A}_{i_j}$ the matrix ${\bf A}$ but with the element $a_{i_j i_j}$
423: replaced by $a_{i_j i_j}+\lambda$. Analogously, the matrix
424: ${\bf A}_{i_1...i_l}$ corresponds to the matrix ${\bf A}$ but
425: with $a_{i_1 i_1}$,...,$a_{i_l i_l}$ replaced by
426: $a_{i_1 i_1}+\lambda$,... $a_{i_l i_l}+\lambda$.
427: Let furthermore ${\bf I}$ be the identity matrix.
428: Then the following equation holds
429: \begin{eqnarray}\label{dupform}
430: &&\det({\bf D}-\lambda {\bf I})=\\
431: && \!\!(-\lambda)^m \det({\bf A}-\lambda{\bf I})\nonumber\\
432: &+&\!\!(-\lambda)^m \sum_{r\leq l} m_{i_r} \det({\bf A}_{i_r}\!-\lambda {\bf I})\nonumber\\
433: &+&\!\!(-\lambda)^m\!\sum_{r\leq l-1}\sum_{r<j\leq l}
434: \!m_{i_r}m_{i_j}\det({\bf A}_{i_ri_j}\!-\lambda {\bf I})\nonumber\\
435: &+&\!\!(-\lambda)^m\!\sum_{r\leq l-2}\sum_{r<j\leq l-1}\sum_{j<s\leq
436: l}\!\!\!m_{i_r}m_{i_j}m_{i_s}\det({\bf A}_{i_ri_ji_s}\!-\lambda {\bf I})\nonumber\\
437: &\vdots&\nonumber\\
438: &+&\!\!(-\lambda)^m\!\sum_{r\leq 1}\sum_{r<j\leq 2}...\sum_{x<y\leq
439: l}\!m_{i_r}...m_{i_y}\det({\bf A}_{i_r...i_y}\!-\lambda{\bf I}).\nonumber
440: \end{eqnarray}
441: Note that the last term is equivalent to
442: $m_{i_1}...m_{i_l}\det({\bf A}_{i_1...i_l}-\lambda{\bf I})$.
443: It gets obvious from this formula that perfect duplication of nodes
444: only adds zeros to the spectrum of the graph.
445: \newline
446: Equation (\ref{dupform}) can be proven by induction.
447: Considering a graph with adjacency matrix ${\bf A}$
448: in which an arbitrary node $i$ is duplicated
449: $m$ times leading a duplication matrix ${\bf D}$ one can show that
450: \begin{equation}\label{step}
451: \det({\bf D}-\lambda {\bf I})=(-\lambda)^m[\det({\bf A}-\lambda {\bf I})
452: +m\det({\bf A}_{i}-\lambda {\bf I})].
453: \end{equation}
454: After validating the case of $l=0$, $m=0$ of equation (\ref{dupform})
455: we do the induction by evaluation of the
456: adjacency matrix $\tilde{{\bf D}}$
457: of a graph generated from the duplication graph represented by ${\bf D}$
458: by duplicating (a non-duplicate) node $i_{l+1}$ $m_{i_{l+1}}$ times.
459: Therefore, we apply (\ref{step})
460: \begin{eqnarray*}
461: &&\det({\tilde{{\bf D}}}-\lambda {\bf I})\\
462: &=&
463: (-\lambda)^m_{i_{l+1}}[\det({\bf D}-\lambda {\bf I})
464: +m_{i_{l+1}}\det({\bf D}_{i_{l+1}}-\lambda {\bf I})]
465: \end{eqnarray*}
466: and derive $\det({\bf D}-\lambda {\bf I})$ and
467: $\det({\bf D}_{i_{l+1}}-\lambda {\bf I})$ using the assumption
468: (\ref{dupform}) yielding the formula (\ref{dupform}) for $m+m_{i_{l+1}}$
469: duplications of $l+1$ mutually different nodes.
470: \newline
471: As an example, this formula is applied to the $2\times2$-matrix ${\bf A}$
472: corresponding to two connected nodes. Then, $i_{1}=1$, ${i_2=2}$ and
473: one gets for the matrix $D$ after $m=m_1+m_2$ duplications:
474: \begin{eqnarray*}
475: \det({\bf D}-\lambda {\bf I})&=&(-\lambda)^m(\lambda^2-1-m_1-m_2-m_1m_2)\\
476: \lambda_{1;2}&=&\pm\sqrt{1+m+m_1m_2}.
477: \end{eqnarray*}
478: Translating this into the number of nodes per orbit $n_i=m_i+1$ leads
479: to the eigenvalues
480: \begin{eqnarray*}
481: \lambda_{1;2}=\pm\sqrt{n_1n_2}.
482: \end{eqnarray*}
483: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
484: \section*{Appendix B}
485: The following tables contain the information on the proteins shown in Fig. \ref{dupgraphs} extracted from the Database of Interacting Proteins \cite{dip} and FlyBase \cite{flybase}.
486: %\begin{table}[!h]
487: \begin{longtable}[!t]{ll}
488: DIP ID & Protein name/description\\
489: \hline
490: {\bf DIP:17489N} & CG11719-PA open reading frame, Mst98Ca,\\
491: & (Male-specific RNA 98Ca) \\
492: {\bf DIP:17490N} & CG18396-PA open reading frame, Mst98Cb,\\
493: &(Male-specific RNA 98Cb) \\
494: DIP:17144N &CG4015-PA open reading frame, Fcp3C,\\
495: & (Follicle cell protein 3C)\\
496: %& Polypeptides: Fcp3C-P1, Transcript: Fcp3C-RA\\
497: DIP:18312N & CG17777-PA open reading frame\\
498: DIP:17051N & CG17666-PA open reading frame\\
499: DIP:17123N & CG15781-PA open reading frame\\
500: DIP:17121N & CG15032-PA open reading frame\\
501: DIP:17122N & CG15489-PA open reading frame\\
502: DIP:20125N & CG1981-PA open reading frame, Thd1,\\
503: &G/T-mismatch-specific-thymine-DNA-\\
504: & glycosylase, double-stranded DNA-binding,\\
505: & mismatch repair\\
506: DIP:20084N & CG13363-PA open reading frame\\
507: DIP:19658N & CG12212-PA open reading frame, peb,\\
508: &(pebbled),\\
509: & transcription factor activity\\
510: %& (Polypeptides: peb-P1, Transcripts: peb-RA)\\
511: DIP:17485N &CG10154-PA open reading frame, \\
512: &structural constituent of peritrophic\\
513: & membrane, (sensu Insecta)\\
514: &\\
515: \caption{Proteins found in the 2-orbit with nodes of degree 10,
516: both duplicates (bold) are male specific RNA (with corresponding polypeptides). }\label{tab210}
517: \end{longtable}
518: %\end{table}
519: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
520: \begin{longtable}[!h]{ll}
521: DIP ID & Protein name/description\\
522: \hline
523: {\bf DIP:18704N} & CG2789-PA open reading frame,\\
524: & bonzodiazepine receptor activity, \\
525: & transporter activity,\\
526: & metabolism and transport\\
527: {\bf DIP:18703N} & CG1341-PA open reading frame, Rpt1,\\
528: &endopeptidase activity, ATPase activity,\\
529: & proteolysis and peptidolysis\\
530: DIP:21261N & CG3173-PA open reading frame\\
531: %&Domains: ARM repeat\\
532: DIP:20398N & CG12096-PA open reading frame\\
533: %&Domains: ARM repeat\\
534: DIP:17864N & CG10694-PA open reading frame\\
535: &damaged DNA-binding, base excision repair\\
536: %&Domains: UNA domain, ubiquities like\\
537: \hline
538: {\bf DIP:20457N} & CG12405-PA open reading frame, Prx2540-1,\\
539: &(Peroxiredoxin 2540),\\
540: &peroxidase, antioxidant activity,\\
541: & defense response, oxygen species metabolism\\
542: {\bf DIP:20458N} & CG12896-PA open reading frame,\\
543: &peroxidase activity\\
544: & defense response, oxygen species metabolism\\
545: DIP:20623N & CG9624-PA open reading frame\\
546: DIP:17811N & CG5576-PA open reading frame, imd,\\
547: &(immune deficiency),\\
548: &antimicrobial humoral response,\\
549: &(sensu Invertebrata)\\
550: DIP:17346N &CG12470-PA open reading frame\\
551: \hline
552: {\bf DIP:18389N} & CG18779-PA open reading frame\\
553: {\bf DIP:18387N} & open reading frame CG-10530/4-PA,\\
554: & Lcp65Ag1/Lcp65Ag2 protein,\\
555: & (larval cuticle protein),\\
556: & structural constituent of larval cuticle,\\
557: & (sensu Insecta),\\
558: & larval cuticle biosynthesis,\\
559: & (sensu Insecta)\\
560: DIP:18392N & CG2082-PA open reading frame,\\
561: & signal transduction\\
562: DIP:18391N & CG16978-PA open reading frame\\
563: DIP:18390N & CG12907-PA open reading frame\\
564: &\\
565: \caption{Proteins found in a 2-orbit with nodes of degree 3, lines
566: separate different orbits, bold proteins are duplicates.}\label{tab23}
567: \end{longtable}
568: %\end{table}
569: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
570: %\begin{table}[!h]
571: \begin{longtable}[!h]{ll}
572: DIP ID & Protein name/description\\
573: \hline
574: {\bf DIP:20847N} & CG8284-PA open reading frame, UbcD4,\\
575: &(Ubiquitin conjugating enzyme 4),\\
576: & ubiquitin conjugating enzyme activity,\\
577: & ligase activity,\\
578: & protein metabolism, ubiquitin cycle\\
579: {\bf DIP:23198N} & CG30344-PA open reading frame\\
580: {\bf DIP:18933N} & CG10862-PA open reading frame,\\
581: & ubiquitin conjugating enzyme activity, \\
582: & ligase activity,\\
583: & protein metabolism\\
584: DIP:18935N & CG8974-PA open reading frame,\\
585: & transcription regulatory activity, \\
586: & ``nucleo-metabolism'', transcription,\\
587: DIP:18934N & CG32581-PA open reading frame,\\
588: & transcription regulatory activity, \\
589: & ``nucleo-metabolism'', transcription\\
590: \hline
591: {\bf DIP:20125N} & CG1981-PA open reading frame, Thd1,\\
592: &G/T-mismatch-specific-thymine-DNA-\\
593: & glycosylase, double-stranded DNA-binding,\\
594: & mismatch repair\\
595: {\bf DIP:19658N} & CG12212-PA open reading frame, peb,\\
596: &(pebbled),\\
597: & transcription factor activity\\
598: %& (Polypeptides: peb-P1, Transcripts: peb-RA)\\
599: {\bf DIP:20084N} &CG13363-PA open reading frame\\
600: DIP:17489N & CG11719-PA open reading frame, Mst98Ca,\\
601: & (Male-specific RNA 98Ca) \\
602: DIP:17490N & CG18396-PA open reading frame, Mst98Cb,\\
603: &(Male-specific RNA 98Cb) \\
604: &\\
605: \caption{{Proteins found in a 3-orbit with nodes of degree 2, lines separate different orbits, bold proteins are duplicates. Note, that Mst98Ca and Mst98Cb form a 2-orbit of degree 10, too (cf. Tab. \ref{tab210}).}}\label{tab32}
606: \end{longtable}
607: %\end{table}
608: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
609: %\begin{table}[!h]
610: \begin{longtable}[!h]{ll}
611: DIP ID & Protein name/description\\
612: \hline
613: {\bf DIP:19548N} &CG31366/18743-PA open reading frame,\\
614: &Hsp70A, (Heat shock protein 70A),\\
615: & heat, defense response, \\
616: &protein complex assembly and folding\\
617: {\bf DIP:19549N} &CG31449/31359/6489-PA open reading frame,\\
618: & Hsp70B, (heat shock protein 70B),\\
619: & heat, defense response,\\
620: & protein complex assembly and folding\\
621: {\bf DIP:19551N} & open reading frame CG31449-PA, Hsp70Ba,\\
622: & (heat shock protein 70Ba),\\
623: & heat, defense response,\\
624: & protein complex assembly and folding\\
625: {\bf DIP:19552N} & CG5834-PA open reading frame, Hsp70Bbb,\\
626: &(heat shock protein 70Bbb)\\
627: DIP:18493N & CG7945-PA open reading frame,\\
628: & chaperone activity\\
629: DIP:20272N & CG5203-PA open reading frame, CHIP,\\
630: %&(Polypeptides: CHIP-P1, Transcripts: CHIP-RA)\\
631: &chaperone activity, protein folding \\
632: & and metabolism\\
633: DIP:20308N & CG32130-PA open reading frame\\
634: DIP:18578N & CG13165-PA open reading frame\\
635: %(DIP:19158N & Heat shock protein cognate 4, \\
636: %&Hsc70-4 (CG4264-PA))\\
637: %& chaperone activity, heat shock protein activity...\\
638: %& defense response, protein complex assembly,\\
639: %& protein folding...\\
640: %(DIP:23702N & CG7970-PA open reading frame)
641: &\\
642: \caption{Proteins found in the 4-orbit with nodes of degree 4,
643: all duplicates (bold) are
644: heat shock proteins (Hsp), released after heat shock or other stress.}\label{tab44}
645: \end{longtable}
646: %\end{table}
647: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
648: %\begin{table}[!h]
649: \begin{longtable}[!h]{ll}
650: DIP ID & Protein name/description\\
651: \hline
652: {\bf DIP:19748N} & CG1252-PA open reading frame, Ccp84Ab,\\
653: &(cuticle cluster 7),\\
654: & structural constituent of larval cuticle,\\
655: & (sensu Insecta)\\
656: {\bf DIP:19749N} & CG2360-PA open reading frame, Ccp84Aa,\\
657: &(cuticle cluster 8),\\
658: & structural constituent of larval cuticle,\\
659: &(sensu Insecta)\\
660: DIP:17392N & CG9949-PA open reading frame, sina,\\
661: & (seven in absentia),\\
662: & sensory organ development\\
663: %& glutathione transferase activity,\\
664: %& defense response, response to toxins\\
665: DIP:18536N & CG6615-PA open reading frame, scaf6,\\
666: & RNA binding, nuclear mRNA splicing\\
667: & via spliceosome, spliceosome complex\\
668: DIP:20225N & CG2341-PA open reading frame, Ccp84Ad,\\
669: &(cuticle cluster 5),\\
670: & structural constituent of larval cuticle,\\
671: & (sensu Insecta)\\
672: DIP:17713N & CG15422-PA open reading frame\\
673: DIP:17492N & CG12723-PA open reading frame\\
674: DIP:17076N & CG6945-PA open reading frame\\
675: DIP:17488N & CG11505-PB open reading frame\\
676: &\\
677: \caption{Proteins found in the 2-orbit with nodes of degree 7, duplicates (bold) are constituents of the larval cuticle. Note, that peb, Thd1, and CG13363-PA also form a 3-orbit with respect to Mst98Ca and Mst98Cb (cf. Tab. \ref{tab32}).}\label{tab27}
678: \end{longtable}
679: %\end{table}
680:
681: \begin{thebibliography}{31}
682: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi
683: \expandafter\ifx\csname bibnamefont\endcsname\relax
684: \def\bibnamefont#1{#1}\fi
685: \expandafter\ifx\csname bibfnamefont\endcsname\relax
686: \def\bibfnamefont#1{#1}\fi
687: \expandafter\ifx\csname citenamefont\endcsname\relax
688: \def\citenamefont#1{#1}\fi
689: \expandafter\ifx\csname url\endcsname\relax
690: \def\url#1{\texttt{#1}}\fi
691: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
692: \providecommand{\bibinfo}[2]{#2}
693: \providecommand{\eprint}[2][]{\url{#2}}
694:
695: \bibitem[{\citenamefont{Strogatz}(2001)}]{strogatz:2001}
696: \bibinfo{author}{\bibfnamefont{S.~H.} \bibnamefont{Strogatz}},
697: \bibinfo{journal}{Nature} \textbf{\bibinfo{volume}{410}},
698: \bibinfo{pages}{268} (\bibinfo{year}{2001}).
699:
700: \bibitem[{\citenamefont{Bollob{\'a}s}(1998)}]{bollobas:modernbook}
701: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Bollob{\'a}s}},
702: \emph{\bibinfo{title}{Modern Graph Theory}} (\bibinfo{publisher}{Springer},
703: \bibinfo{address}{New York}, \bibinfo{year}{1998}).
704:
705: \bibitem[{\citenamefont{Bollob{\'a}s}(2001)}]{bollobas:randomgraphs}
706: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Bollob{\'a}s}},
707: \emph{\bibinfo{title}{Random Graphs}} (\bibinfo{publisher}{Cambridge
708: University Press}, \bibinfo{address}{Cambridge}, \bibinfo{year}{2001}).
709:
710: \bibitem[{\citenamefont{Albert and
711: Barab{´\'{a}}si}(2002)}]{albert:barabasi:2002}
712: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Albert}} \bibnamefont{and}
713: \bibinfo{author}{\bibfnamefont{A.-L.} \bibnamefont{Barab{´\'{a}}si}},
714: \bibinfo{journal}{Rev. Mod. Phys.} \textbf{\bibinfo{volume}{74}},
715: \bibinfo{pages}{47} (\bibinfo{year}{2002}).
716:
717: \bibitem[{\citenamefont{Chung}(1994)}]{chung:book}
718: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Chung}},
719: \emph{\bibinfo{title}{Spectral Graph Theory}}, vol.~\bibinfo{volume}{92} of
720: \emph{\bibinfo{series}{Regional Conference Series in Mathematics}}
721: (\bibinfo{publisher}{American Mathematical Society Providence},
722: \bibinfo{address}{Rhode Island}, \bibinfo{year}{1994}).
723:
724: \bibitem[{\citenamefont{Cvetkovi{\'c} et~al.}(1980)\citenamefont{Cvetkovi{\'c},
725: Doob, and Sachs}}]{cvetkovic:book}
726: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Cvetkovi{\'c}}},
727: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Doob}}, \bibnamefont{and}
728: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Sachs}},
729: \emph{\bibinfo{title}{Spectra of graphs}}, Pure and applied mathematics
730: (\bibinfo{publisher}{Academic Press}, \bibinfo{year}{1980}).
731:
732: \bibitem[{\citenamefont{Cvetkovi{\'c} et~al.}(1988)\citenamefont{Cvetkovi{\'c},
733: Doob, Gutman, and Torga{\v{s}}ev}}]{cvetkovic:annals}
734: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Cvetkovi{\'c}}},
735: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Doob}},
736: \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Gutman}}, \bibnamefont{and}
737: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Torga{\v{s}}ev}},
738: \emph{\bibinfo{title}{Recent results in the theory of graph spectra}},
739: no.~\bibinfo{number}{36} in \bibinfo{series}{Annals of discrete mathematics}
740: (\bibinfo{publisher}{North Holland}, \bibinfo{year}{1988}).
741:
742: \bibitem[{\citenamefont{Biggs}(1996)}]{biggs:book}
743: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Biggs}},
744: \emph{\bibinfo{title}{Algebraic graph theory}} (\bibinfo{publisher}{Cambridge
745: University Press}, \bibinfo{address}{Cambridge}, \bibinfo{year}{1996}).
746:
747: \bibitem[{\citenamefont{Bauer and Golinelli}(2001)}]{bauer:golinelli:2001}
748: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Bauer}} \bibnamefont{and}
749: \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{Golinelli}},
750: \bibinfo{journal}{J. Stat. Phys.} \textbf{\bibinfo{volume}{103}},
751: \bibinfo{pages}{301} (\bibinfo{year}{2001}).
752:
753: \bibitem[{\citenamefont{Farkas et~al.}(2001)\citenamefont{Farkas, Der\'{e}nyi,
754: Barab\'{a}si, and Vicsek}}]{farkas:vicsek:2001}
755: \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Farkas}},
756: \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Der\'{e}nyi}},
757: \bibinfo{author}{\bibfnamefont{A.-L.} \bibnamefont{Barab\'{a}si}},
758: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Vicsek}},
759: \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{64}},
760: \bibinfo{pages}{026704} (\bibinfo{year}{2001}).
761:
762: \bibitem[{\citenamefont{Dorogovtsev et~al.}(2003)\citenamefont{Dorogovtsev,
763: Goltsev, Mendes, and Samukhin}}]{dorogovtsev:samukhin:2003}
764: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Dorogovtsev}},
765: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Goltsev}},
766: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Mendes}}, \bibnamefont{and}
767: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Samukhin}},
768: \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{68}},
769: \bibinfo{pages}{046109} (\bibinfo{year}{2003}).
770: % \urlprefix\url{http://arxiv.org/abs/cond-mat/0306340}.
771:
772: \bibitem[{\citenamefont{Chung et~al.}(2003{\natexlab{a}})\citenamefont{Chung,
773: Lu, and Vu}}]{chung:vu:2002}
774: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Chung}},
775: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Lu}}, \bibnamefont{and}
776: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Vu}},
777: \bibinfo{journal}{Annals of Combinatorics} \textbf{\bibinfo{volume}{7}},
778: \bibinfo{pages}{21} (\bibinfo{year}{2003}{\natexlab{a}}).
779: % \bibinfo{note}{preprint from http://www.math.ucsd.edu/\~fan}.
780:
781: \bibitem[{\citenamefont{Chung et~al.}(2003{\natexlab{b}})\citenamefont{Chung,
782: Lu, and Vu}}]{chung:vu:2003}
783: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Chung}},
784: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Lu}}, \bibnamefont{and}
785: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Vu}}, \bibinfo{journal}{Proc.
786: Natl. Acad. Sci. USA} \textbf{\bibinfo{volume}{100}}, \bibinfo{pages}{6313}
787: (\bibinfo{year}{2003}{\natexlab{b}}).
788:
789:
790: \bibitem[{\citenamefont{Mihail and
791: Papadimitriou}(2002)}]{mihail:papadimitriou:2002}
792: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Mihail}} \bibnamefont{and}
793: \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Papadimitriou}}, in
794: \emph{\bibinfo{booktitle}{Proceedings of the 6th International Workshop on
795: Randomization and Approximation Techniques}} (\bibinfo{year}{2002}), pp.
796: \bibinfo{pages}{254--262}.
797:
798: \bibitem[{\citenamefont{Goh et~al.}(2001)\citenamefont{Goh, Kahng, and
799: Kim}}]{goh:kim:2001}
800: \bibinfo{author}{\bibfnamefont{K.-I.} \bibnamefont{Goh}},
801: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Kahng}}, \bibnamefont{and}
802: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Kim}},
803: \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{64}},
804: \bibinfo{pages}{051903} (\bibinfo{year}{2001}).
805:
806: \bibitem[{\citenamefont{Golinelli}(2003)}]{golinelli:2003}
807: \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{Golinelli}},
808: \bibinfo{journal}{http://arxiv.org/abs/cond-mat/0301437v1}
809: (\bibinfo{year}{2003}).
810:
811: \bibitem[{\citenamefont{Milo et~al.}(2002)\citenamefont{Milo, Shen-Orr,
812: Itzkovitz, Kashtan, Chklovskii, and Alon}}]{milo:alon:2002}
813: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Milo}},
814: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Shen-Orr}},
815: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Itzkovitz}},
816: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Kashtan}},
817: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Chklovskii}},
818: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Alon}},
819: \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{298}},
820: \bibinfo{pages}{824} (\bibinfo{year}{2002}).
821:
822: \bibitem[{\citenamefont{Milo et~al.}(2004)\citenamefont{Milo, Kashtan, Levitt,
823: Shen-Orr, Ayzenshtat, Sheffer, and Alon}}]{milo:alon:2004}
824: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Milo},
825: \bibfnamefont{R.~Itzkovitz}},
826: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Kashtan}},
827: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Levitt}},
828: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Shen-Orr}},
829: \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Ayzenshtat}},
830: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Sheffer}}, \bibnamefont{and}
831: \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Alon}},
832: \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{303}},
833: \bibinfo{pages}{1538} (\bibinfo{year}{2004}).
834:
835: \bibitem[{\citenamefont{Giot et~al.}(2003)\citenamefont{Giot, Bader, Brouwer,
836: Chaudhuri, Kuang, Li, Hao, Ooi, Godwin, Vitols et~al.}}]{giot:rothberg:2003}
837: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Giot}},
838: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Bader}},
839: \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Brouwer}},
840: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Chaudhuri}},
841: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Kuang}},
842: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Li}},
843: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Hao}},
844: \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Ooi}},
845: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Godwin}},
846: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Vitols}},
847: \bibnamefont{et~al.}, \bibinfo{journal}{Science}
848: \textbf{\bibinfo{volume}{302}}, \bibinfo{pages}{1727} (\bibinfo{year}{2003}).
849:
850: \bibitem[{dip()}]{dip}
851: \emph{\bibinfo{title}{Database of Interacting Proteins ({DIP})}},
852: \urlprefix\url{http://dip.doe-mbi.ucla.edu/}.
853:
854: \bibitem[{\citenamefont{Maslov and Sneppen}(2002)}]{maslov:sneppen:2002}
855: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Maslov}} \bibnamefont{and}
856: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Sneppen}},
857: \bibinfo{journal}{Science} \textbf{\bibinfo{volume}{296}},
858: \bibinfo{pages}{910} (\bibinfo{year}{2002}).
859:
860: \bibitem[{\citenamefont{Milo et~al.}()\citenamefont{Milo, Shen-Orr, Itzkovitz,
861: Kashtan, Chklovskii, and Alon}}]{mfinder}
862: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Milo}},
863: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Shen-Orr}},
864: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Itzkovitz}},
865: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Kashtan}},
866: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Chklovskii}},
867: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Alon}},
868: \emph{\bibinfo{title}{{\tt Mfinder} Software}},
869: %\bibinfo{howpublished}{http://www.weizmann.ac.il/mcb/UriAlon/groupDownloadab%
870: %leData.html},
871: \urlprefix\url{http://www.weizmann.ac.il/mcb/UriAlon/groupDownloadableData.html}.
872:
873: \bibitem[{\citenamefont{Cvetkovi{\'c} et~al.}(1979)\citenamefont{Cvetkovi{\'c},
874: Doob, and Sachs}}]{spectraofgraphs:book}
875: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Cvetkovi{\'c}}},
876: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Doob}}, \bibnamefont{and}
877: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Sachs}},
878: \emph{\bibinfo{title}{Spectra of Graphs}}, Pure and Applied Mathematics
879: (\bibinfo{publisher}{Academic Press}, \bibinfo{year}{1979}).
880:
881: \bibitem[{\citenamefont{Ohno}(1970)}]{ohno:book}
882: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Ohno}},
883: \emph{\bibinfo{title}{Evolution by gene duplication}}
884: (\bibinfo{publisher}{Springer}, \bibinfo{address}{Berlin},
885: \bibinfo{year}{1970}).
886:
887: \bibitem[{\citenamefont{Teichmann and Babu}(2004)}]{teichmann:babu:2004}
888: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Teichmann}} \bibnamefont{and}
889: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Babu}},
890: \bibinfo{journal}{Nature Genetics} \textbf{\bibinfo{volume}{36}},
891: \bibinfo{pages}{492} (\bibinfo{year}{2004}).
892:
893: \bibitem[{\citenamefont{Amoutzias et~al.}(2004)\citenamefont{Amoutzias,
894: Robertson, Oliver, and Bornberg-Bauer}}]{amoutzias:bauer:2004}
895: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Amoutzias}},
896: \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Robertson}},
897: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Oliver}}, \bibnamefont{and}
898: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Bornberg-Bauer}},
899: \bibinfo{journal}{{EMBO}} \textbf{\bibinfo{volume}{5}}, \bibinfo{pages}{1}
900: (\bibinfo{year}{2004}).
901:
902: \bibitem[{\citenamefont{Kashtan et~al.}(2004)\citenamefont{Kashtan, Itzkovitz,
903: Milo, and Alon}}]{kashtan:alon:2004}
904: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Kashtan}},
905: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Itzkovitz}},
906: \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Milo}}, \bibnamefont{and}
907: \bibinfo{author}{\bibfnamefont{U.}~\bibnamefont{Alon}},
908: \bibinfo{journal}{Bioinformatics} (\bibinfo{year}{2004}), \bibinfo{note}{in
909: press}.
910:
911: \bibitem[{fly()}]{flybase}
912: \emph{\bibinfo{title}{FlyBase, A Database of the Drosophila Genome}},
913: \urlprefix\url{http://flybase.bio.indiana.edu/}.
914:
915: \bibitem[{\citenamefont{Samanta and Liang}(2003)}]{samanta:liang:2003}
916: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Samanta}} \bibnamefont{and}
917: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Liang}},
918: \bibinfo{journal}{Proc. Natl. Acad. Sci. USA} \textbf{\bibinfo{volume}{100}},
919: \bibinfo{pages}{12579} (\bibinfo{year}{2003}).
920:
921: \bibitem[{\citenamefont{Kirkpatrick and
922: Eggarter}(1972)}]{kirkpatrick:eggarter:1972}
923: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Kirkpatrick}} \bibnamefont{and}
924: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Eggarter}},
925: \bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{6}},
926: \bibinfo{pages}{3598} (\bibinfo{year}{1972}).
927:
928: \bibitem[{\citenamefont{Evangelou}(1983)}]{evangelou:1983}
929: \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Evangelou}},
930: \bibinfo{journal}{Phys. Rev. B} \textbf{\bibinfo{volume}{27}},
931: \bibinfo{pages}{1397} (\bibinfo{year}{1983}).
932:
933: \end{thebibliography}
934:
935: %\bibliographystyle{apsrev}
936: %\input{literatur_resub}
937: %\bibliography{/mn/tid/pgp-e1/ckamp/literature/lit}
938: \end{document}
939:
940:
941: