Network concepts for analyzing 3D genome structure from chromosomal contact maps
 JeanBaptiste Morlot^{1},
 Julien Mozziconacci^{1}Email author and
 Annick Lesne^{1, 2}Email author
DOI: 10.1140/epjnbp/s4036601600295
© Morlot et al. 2016
Received: 12 February 2016
Accepted: 7 April 2016
Published: 5 May 2016
Abstract
Background
The recent experimental technique of chromosome conformational capture gives an invivo access to pairwise contact frequencies between genomic loci. We present how network analysis can be exploited to extract information from genomewide contact maps.
Methods
We recently proposed to use graph distance for deriving a complete distance matrix from sparse contact maps. Completed with multidimensional scaling (MDS), this networkbased method provided a fast algorithm, ShRec3D, for reconstructing 3D genome structures.
Results
We here develop an extension of this algorithm, by devising a tunable variant of the graph distance and introducing an alternative implementation of multidimensional scaling. This extended algorithm is shown to be more flexible so as to accommodate additional experimental constraints, focus on specific spatial scales, and produce tractable representations of human data.
Conclusions
Network representation of genomic contacts offers a path where physical and systemic approaches are joined to unravel the biological role of the 3D genome structure.
Keywords
3D genome structure Contact map Network analysis Graph distanceBackground
A recent experimental technique, chromosome conformation capture, gives access to pairwise contacts between genomic sites in populations of living cell nuclei [1–3]. Completed with observations obtained by imaging techniques, it provides increasing evidence of the functional importance of the 3D genome structure, e.g. in the regulation of gene expression [4, 5]. Conformation capture data are usually processed into contact maps. We explore the benefits of considering a contact map as the adjacency matrix of an undirected graph, accordingly termed a contact network.
A first interest, reviewed in Section “Contact maps and contact networks”, is to use the concepts developed in statistical physics for complex network analysis [6, 7]. This path has already been explored to characterize the native structure of proteins [8]. In the genomic context, the challenge lies in the large size of the contact maps, their sparseness, and the fluctuating nature of the genome conformation, averaged over cells and time in the experiment.
A second interest, presented in Section “3D genome structure reconstruction”, is to use the contact network representation to compute the graph distance between any pair of genomic sites, including those displaying no (or very few) contact(s). It has been exploited in [9] to devise a fast reconstruction algorithm, named ShRec3D for Shortestpath 3D Reconstruction to underline the importance of taking graph distance as a starting point of multidimensional scaling methods for reconstructing the underlying 3D genome structure.
We propose in Section “Results and discussion: an extension of ShRec3D for human genome” an extension of this reconstruction algorithm, involving a tunable graph distance and two different MDS implementations. In the line of experiments using fluorescence insitu hybridization (FISH) data, which evidenced a powerlaw correlation between contact frequencies and measured distances [2], we explore the relationships between the contact frequencies, the graph distances, and the distances within the reconstructed 3D structures. We dissect the transformations achieved by the different steps of the algorithm and benchmark its possible variants. As a result, we identify a tradeoff between controlling the reconstruction at small scales or at large scales, and propose operational options for exploiting real data, typically human data in normal and pathological situations.
Contact maps and contact networks
Chromosome conformation capture
From contact maps to contact networks
The most standard approach is the direct analysis of contact maps using various statistical tools, e.g. contact density, Principal Component Analysis or motif finding [2, 3, 11]. An alternative approach is to consider a contact map F as the adjacency matrix of an undirected network. There are slightly different ways to implement this general idea, e.g. considering a network with multiple edges or with weighted edges. The simplest case of a binary contact map is presented on Fig. 1 b, using a network drawing minimizing the number of crossings between edges in the plane of the figure. Noticeably, labeling of the nodes in such networks is not purely conventional, like in most complex networks, but prescribed by their linear ordering along the genome.
Contact networks are spatially embodied networks with steric constraints on the node degrees: a node cannot establish a contact with an unlimited number of other nodes, exactly like a city cannot be connected by direct highways to an unlimited number of other cities. As such, they are not expected to satisfy the smallworld property. These constraints are partly alleviated in HiC experiments done on cell populations, where the contact network originates from an average contact map, derived from a huge number of individual conformations.
Although such networks are sometimes called interaction networks [12], it should be noted that a contact only reflects a spatial proximity at the time of the experiment. It may result from random thermal motion of DNA, and does not necessarily imply a specific biochemical or physical interaction between the genomic sites. Only a special experimental protocol (chromatin interaction analysis using paired end tags, ChIAPET, [13]), designed to extract the contacts mediated by a given protein, e.g. a polymerase, gives access to chromatin interaction networks [14].
Network representation can be exploited in different ways in the context of genomic studies. We present in the next section a short review of the alternative approaches and points of view. Our exploitation of contact networks for 3D genome reconstruction will appear to be quite novel and unrelated to previous works.
Contact network analysis of HiC maps: a short review
The network view of contact maps already gave systemic insights on the genome organization in the nuclear space.
In [15], the authors chose a network representation in which each observed contact is associated with an edge. Nodes are thus related by multiple edges, as many as the number of reads recorded in the experiment. By implementing a rewiring procedure at fixed degrees, they showed that the human contact network is different from a random graph, in particular as regards the histogram of the number of contacts.
In [12], the authors computed five graphtopological measures of the intrachromosomal contact network: diameter, degree distribution, betweenness centrality, clustering coefficient and Jaccard index (relative number of neighbors shared by a pair of nodes). They actually used scaledependent analogs of the standard notions, related to the diffusion kernel exp[ β(F−K)] (where K is the degree matrix) and presumed to capture network characteristics at different ranges of organization when the parameter β is varied [16]. Identifying each gene with the fragment containing its transcriptional start site, they showed a correlation between coexpression of genes and their 3D colocalization, that was proposed as a prediction tool.
In [17], a network approach of a human contact map at a resolution of 100kb has been developed to analyze the relationship between replication timing and genomic contacts. Replication origins located at the border of replication domains, termed master replication origins, are shown to correspond to nodes of maximal network centrality. This feature is observed for three network centralities (degree, betweenness and eigenvector centralities) in both the unweighted contact network and the network where edges are weighted by the number of reads.
Louvain algorithm devised to detect graph communities has been applied to the contact network of metagenomes, in order to identify the constituting genomes [18].
In [14], using a ChIAPET protocol specifically targeting contacts involving a polymerase, the authors found that 40 % of the total genomic elements involved in chromatin interactions converged to a giant, scalefreelike, hierarchical network organized into chromatin communities, with a negative correlation between the degree and the clustering coefficient. In the context of genomewide association studies, they observed that hubs of this transcriptionassociated interaction network lack diseaseassociated singlenucleotide polymorphisms.
3D genome structure reconstruction
The challenge
Beyond statistical analyses, another direction for exploiting contact maps is to reconstruct the underlying 3D genome structures and visualize the corresponding shapes in the 3D space. An issue is the large size of genomic contact maps, which requires fast reconstruction algorithms. Existing methods for reconstructing the native structure of a protein from its contact map, e.g. by targeted growth [19], are limited to a few hundreds of elements at the very most, hence do not apply to the large HiC contact maps. Standard reconstruction methods for genomic data are based on iterative structure optimization until experimental contacts are matched [20], and they are also limited to a small number of sites.
Another issue lies in the fact that not all the contacts are detected. The absence of reads for a pair of sites does not assess, and should not be handled as, an absence of contact.
In what follows, we consider binary contact maps only for explanation and illustration purposes, as in Fig. 1, and perform all the analyses with continuouslyvalued contact frequencies, so as to avoid the choice of a threshold and fully exploit the quantitative nature of HiC data.
From contact maps to distance matrices
The standard method to derive spatial distances from conformational capture data is to consider that distances are inversely proportional to numbers of contacts and to associate a distance L _{ ij }=1/F _{ ij } to the pair of sites (i,j) [21]. A difficulty arises at high resolution (typically less than 100 kb) due to the sparseness of the contact map F, in which a lot of components vanish. The corresponding distance matrix would thus contain a lot of infinite components. Nonvanishing but very small values of F _{ ij } are also problematic, in giving a very large value 1/F _{ ij } which does not correspond to the average distance between the sites i and j. Moreover, such a definition does not satisfy the triangular inequality, i.e. L is not a distance matrix.
Considering the contact map as the adjacency matrix of a network, we proposed to associate to a pair of sites (i,j) the distance D _{ ij } obtained by computing their graph distance, that is, the minimal number of edges in a path relating the nodes [9, 22]. This definition applies to any pair of sites, including those displaying no significant contact, hence provides a complete distance matrix D. This procedure in particular circumvents experimental limitations preventing to detect all the contacts.
However, the plain graph distance is too rough since it treats equally all the edges of the network, while a high contact frequency F _{ ij } reflects a close proximity of the sites i and j. Accordingly, we have endowed each contactassociated edge with a length L _{ ij }=1/F _{ ij }. The components of L are no longer used as distances, but as auxiliary weights involved in computing the path lengths, instead of simply counting a number of edges. This weighting does not change the fact that the shortestpath distance D _{ ij } is a true distance, satisfying the triangular inequality.
ShRec3D: implementing classical multidimensional scaling on graph distances
The next step is the computation of the socalled metric matrix M, related to D by algebraic relationships (see Methods). In ideal situations, where the distance matrix components are the actual Euclidean distances between the points of 3D structure, M is semidefinite positive of rank equal to the underlying topological dimension, namely 3; a theorem from distance geometry then ensures that it coincides with the Gram matrix G (matrix of scalar products) of the structure [23], which is reversibly related to its 3D coordinates. When starting from experimental data, D is marred by errors, M is not semidefinite positive and the theorem no longer applies. Moreover, D is reconstructed from an average contact map, i.e. from a superposition of structures, which also reflects in the presence of more than 3 nonvanishing eigenvalues. Classical MDS cures both problems in a simple way, by considering the truncation G of rank 3 obtained by keeping the largest three eigenvalues of M. The associated eigenvectors yield the 3D coordinates V (see Methods).
The spectrum of M reflects up to what point the matrix D is close to the Euclidean distance matrix of a single 3D structure. MDS truncation of M enforces the existence of an underlying 3D structure, which is an optimal approximation in the sense that the quadratic error between the experimental distances D and the distances R in the reconstructed structure is minimal [24]. The quality of this approximation can be checked on the spectrum of M, displaying three isolated positive eigenvalues while the remaining part of the spectrum is concentrated around 0, Fig. 2 c. It is essential for the quality of the MDS approximation that D is a true distance matrix, satisfying the triangular inequality. In contrast, it has been checked in [9] that applying MDS to the matrix L (instead of D) gives very poor results, the reconstructed structure being then almost uncorrelated with the actual one.
Since the elements of D take dimensionless values, the 3D structure is obtained up to a scale transformation; only the ratio of the distances is meaningful. The reconstructed distances R could be calibrated with respect to the size of the nucleus. As we focus only on the topology of the 3D genome structure, we kept dimensionless values for the distances, Fig. 2 d.
Results and discussion: an extension of ShRec3D for human genome
A guideline based on fluorescence insitu hybridization(FISH) experiments
FISH protocol associates fluorescent tags to a few specific genomic sites. It allows the accurate measurement in a population of fixed cells of the spatial distances between these sites and their distribution. However, the number of investigated sites is very limited, in contrast to the genomewide coverage permitted by conformational capture techniques. FISH experiments have been used to check that conformational capture actually provides information on invivo distances. They provide the only independent constraint on the 3D reconstruction from HiC maps.
Tunable graph distances
In the line of the powerlaw correlation observed in FISH data, we endow each contactassociated edge with a length \(L_{ij}\sim F_{ij}^{\alpha }\), depending on a tunable parameter α. This extension, proposed for L used as an ansatz for the distances [25, 26], is here integrated in our networkbased computation of the distances. We investigated the influence of the value of α on the properties of the shortestpath distance matrix D and its relationship with F (short blue arrow in Fig. 2 a), with two extreme cases α=0.2 (the rounded value of the exponent observed experimentally in the abovedescribed situation) and α=1 (the value adopted in the original algorithm).
By definition, the shortestpath distance D _{ ij } is always smaller or equal to the edge length L _{ ij }, as can be seen on Fig. 4 b. It is expected —and intended— that D does not rely on low contact frequencies, associated with long edges in the contact network. Figure 4 b shows that the difference between D and L is indeed more marked for smaller contact frequencies, i.e. larger distances. We quantified this feature by the percentage N _{ Sh } of pairs (i,j) with nontrivial shortestpath distance D _{ ij }<L _{ ij }. The pairs of sites contributing to N _{ Sh } are those with low contact frequencies, for which the shortestpath travels through different and shorter connections than the edge (i,j). When α increases, the discrepancy between L and D is observed to increase, as illustrated by the two panels of Fig. 4 b. This trend is assessed by plotting the increase of the percentage N _{ Sh } when α increases, Fig. 4 c. The correlation between the contact frequency F _{ ij } observed for a pair of sites and their shortestpath distance D _{ ij } can be summarized in a scaling law, with an exponent α _{ Sh } (minus the slope of the red lines in Fig. 4 b). The dependence of α _{ Sh } as a function of α is shown on Fig. 4 d. A crossover is observed at a value α≈0.2.
Overall, the improvement brought by using shortestpath distances D as an input to MDS is more important for larger distances and larger values of α. However, choosing a large value of α is not necessarily the best choice: in this regime, the distances D are derived mainly from a few large contact frequencies measured in the HiC experiment while less frequent contacts do not contribute, which filters out noise and unreliable recordings but possibly also relevant information. Also, the scaling of the distances with respect to the contact frequencies is modified by the shortestpath computation, and Fig. 4 d provides a calibration curve for the considered data, allowing one to control α _{ Sh } by a proper choice of α. Further analysis is presented below, with a focus on the extreme values α=0.2 and α=1.
Effect of the multidimensional scaling
We further explored the relationship between the reconstructed distances R and the contact frequencies F (long blue arrow in Fig. 2 a) as a function of α. We moreover compared two versions of MDS, corresponding to different optimization criteria hence different approximations. Classical MDS corresponds to the minimization of \(\sum _{i,j}(D_{ij}R_{ij})^{2}\). The strength of this method is to reduce to the determination of the three first eigenvectors of the metric matrix M, as explained above. Its weakness is the low constraint on small distances, since minimizing the error is achieved mainly by controlling the large distances. This dominance of large distances can be corrected by considering the relative error [25], leading to the socalled (nonclassical) metric MDS (see Methods). Importantly, both classical MDS and metric MDS are applied to the shortestpath distance matrix D. In contrast, MDS applied to L is highly unstable, due to the treatment of infinite or abnormal components of L and the fact that L is not a distance matrix [9]. As regards computational time, nonclassical MDS starts from the classical MDS solution hence takes more time. At larger sizes, their computational performances converge, due to the fact that the (common) limiting step is the computation of shortest paths, see Additional file 1: Figure S1.
The value of α initially taken in the expression of edge lengths L is not recovered in the relationship between the reconstructed distance and the contact frequencies, with exponent α ^{∗}. Part of the difference between the two exponents comes from the shortestpath computation, Fig. 4 d, and part from the MDS dimensional reduction, Fig. 5 c. This latter figure shows that metric MDS has a smaller impact on the exponent α ^{∗} than classical MDS. Using Fig. 4 d, it is possible to choose a value of α to get the desired correlation behavior in the reconstructed structure, with some limitations. Noticeably, the effect of MDS on α ^{∗} is weaker at larger α.
The value α _{ FISH }=0.227 is at the lower boundary of the accessible range for α ^{∗}. However, this exponent has been obtained from experimental data corresponding to large distances. This experimental range is difficult to delineate precisely, so that a partial fit would not be reliable; it is nevertheless apparent on Figs. 5 a–d (dashed black line) that a smaller exponent α ^{∗} would be obtained in the largedistance range, supporting the experimental consistency of the reconstructed structure.
Flexibility of the extended ShRec3D algorithm
It also apparent on the respective 3D reconstructions, Fig. 6 c and Fig. 6 d, that metric MDS reproduces smallscale features (e.g. small loops), while the global shape is more clearly represented with classical MDS.
For small values of α (Fig. 6 c), the reconstructed structure is more compact, closer to the results of imaging experiments. For larger values of α (Fig. 6 d), the reconstructed 3D structure is more extended, which is specially suitable for 3D genome browsers. Tuning the exponent α thus allows one to focus either on short or large scales.
Note that a distortion arises in Fig. 6 c and d due to the 2D projection of the 3D structures on the printed sheet. The alignement of the structures obtained with different MDS implementations have been done without any rescaling, since they are based on the same distance matrix D.
Conclusion
Experimental advances permitted by the HiC protocol pointed to the need of bridging a physical viewpoint, enlightening the functional role of 3D genome structure, with a systemic viewpoint, based on genomewide data and network analysis. A pillar of this bridge is the development of reconstruction algorithms, in which information limited to contacts is sufficient to get a 3D representation of the data. An auxiliary though important step is to transform the contact maps into complete distance matrices.
Our analysis shows that shortestpath distances, inspired by network concepts, is to date the best way to implement this step with human data, making it possible to deal with sparse chromosomal contact maps and match FISH data. The extension of ShRec3D presented here, with a tunable parameter α in the definition of the graph distances and two implementations of MDS, provides a flexible algorithm to accommodate various organisms, conditions and goals.
Methods
Experimental data: We used human HiC data obtained from lymphoblastoids (cell type GM12878) at a resolution of 1kb [3]. In the analyses presented here, we take as a benchmark a fragment of chromosome 1 of size 10 megabases (Mb).
These data have been coarsegrained into bins of 10 kb then unbiased and normalized according to the procedure explained in [10], yielding the contact map F presented in Fig. 4 a. It satisfies \(\sum _{j}F_{ij}=1\) for all sites i.
Contact network: In the binary representation considered for illustration purposes, Fig. 1, the diagonals (i,i±1) are included in the contact map in order to enforce the connectivity of the genome; accordingly the contact network is connected. It is thus possible to compute the shortest path between any pair of nodes. In the extension of our algorithm ShRec3D presented here, we use as an input continuousvalued contact maps F (contact frequencies), and the edge (i,j) is endowed with a length equal to \(F_{ij}^{\alpha }\).
This structure is the best 3D approximation in the sense that it minimizes the quadratic error \(\sum _{i,j}(D_{ij}R_{ij})^{2}\) between D and the distances R in the reconstructed structure. We here keep 3 eigenvectors in a supervised way, since we are looking for a 3D structure. The relevance of this choice can nevertheless be checked on the spectrum of M, which presents exactly 3 isolated eigenvalues, see Fig. 4 c. The same method could apply in any dimension, keeping m eigenvectors for a mdimensional structure.
In contrast with classical MDS, there is no longer an analytical solution relating D with the optimal coordinates. The minimization of the stress is achieved by iterative optimization (MATLAB function mdscale with criterion metricstress). Noticeably, the procedure takes as a starting point the 3D structure provided by classical MDS, in order to reduce the nonconvex optimization problem to a local minimization problem and exploit the efficient dimensional reduction ensured by cMDS. In this way the computational performance remains satisfactory, especially at large sizes for which the duration of the MDS step is anyhow overwhelmed by the computation of the shortestpath distances (see Additional file 1: Figure S1). Other MDS options are possible [25, 26]. Beyond classical and metric MDS, we also investigated the specifications of ShRec3D when implemented with Sammon MDS [28] and nonmetric MDS [29, 30]. Basically these two latter options give results quite similar to metric MDS. Accordingly, we discuss in the main text the results obtained with classical and metric MDS, and present some additional tests comparing the four methods (classical MDS, metric MDS, Sammon MDS and nonmetric MDS) in the Supplementary Materials.
Numerical implementation: The original algorithm ShRec3D [9] has been extended to include the edgelength exponent α as a tunable parameter, and it now implements both classical and nonclassical metric MDS. The MATLAB code is available at: https://sites.google.com/site/julienmozziconacci/#TOCPubliclyavailablesoftwares
Abbreviations
 ChIAPET:

Chromatin interaction analysis using paired end tags
 FISH:

Fluorescence in situ hybridization
 MDS:

MultiDimensional scaling (with two variants, cMDS: classical MDS and mMDS: nonclassical metric MDS)
 ShRec3D:

Shortestpath 3D reconstruction algorithm
Declarations
Acknowledgements
This work has been funded by the French Institut National du Cancer, grant INCa_5960 (AL), the French Agence Nationale de la Recherche, grant ANR13BSH30007 (AL) and grant ANR15CE11002301 (JM), and University Pierre and Marie Curie, Emergence program, grant SU15REMR08 (JM). The authors are grateful to Thierry Forné for his comments on the manuscript, and to Michel Quaggetto for his help with softwares. AL acknowledges the hospitality of Jacobs University, Bremen (Germany) during the WE Heraeus Physics School “The physics behind systems biology" (6–12 July 2015).
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.
Authors’ Affiliations
References
 Dekker J, Rippe K, Dekker M, Kleckner N. Capturing chromosome conformation. Science. 2002; 295:1306–11.View ArticleADSGoogle Scholar
 LiebermanAiden E, van Berkum NL, Williams L, Imakaev M, Ragoczy T, Telling A, Amit I, Lajoie BR, Sabo PJ, Dorschner MO, et al. Comprehensive mapping of longrange interactions reveals folding principles of the human genome. Science. 2009; 326(5950):289–93.View ArticleADSGoogle Scholar
 Rao SS, Huntley MH, Durand NC, Stamenova EK, Bochkov ID, Robinson JT, Sanborn AL, Machol I, Omer AD, Lander ES, et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell. 2014; 159(7):1665–80.View ArticleGoogle Scholar
 Dekker J, Misteli T. Longrange chromatin interactions. Cold Spring Harbor Perspect Biol. 2015; 7(10):019356.View ArticleGoogle Scholar
 Ea V, Baudement MO, Lesne A, Forné T. Contribution of topological domains and loop formation to 3D chromatin organization. Genes. 2015; 6(3):734–50.View ArticleGoogle Scholar
 Newman ME. The structure and function of complex networks. SIAM Rev. 2003; 45(2):167–256.MathSciNetView ArticleMATHADSGoogle Scholar
 Boccaletti S, Latora V, Moreno Y, Chavez M, Hwang DU. Complex networks: Structure and dynamics. Phys Reports. 2006; 424(4):175–308.MathSciNetView ArticleADSGoogle Scholar
 Di Paola L, De Ruvo M, Paci P, Santoni D, Giuliani A. Protein contact networks: an emerging paradigm in chemistry. Chem Rev. 2012; 113(3):1598–613.View ArticleGoogle Scholar
 Lesne A, Riposo J, Roger P, Cournac A, Mozziconacci J. 3D genome reconstruction from chromosomal contacts. Nat Methods. 2014; 11(11):1141–3.View ArticleGoogle Scholar
 Cournac A, MarieNelly H, Marbouty M, Koszul R, Mozziconacci J. Normalization of a chromosomal contact map. BMC Genomics. 2012; 13:436.View ArticleGoogle Scholar
 Dixon JR, Selvaraj S, Yue F, Kim A, Li Y, Shen Y, Hu M, Liu JS, Ren B. Topological domains in mammalian genomes identified by analysis of chromatin interactions. Nature. 2012; 485(7398):376–80.View ArticleADSGoogle Scholar
 Babaei S, Mahfouz A, Hulsman M, Lelieveldt BP, de Ridder J, Reinders M. HiC chromatin interaction networks predict coexpression in the mouse cortex. PLoS Comput Biol. 2015; 11(5):1004221.View ArticleGoogle Scholar
 Singh Sandhu K, Li G, Sung WK, Ruan Y. Chromatin interaction networks and higher order architectures of eukaryotic genomes. J Cell Biochem. 2011; 112(9):2218–21.View ArticleGoogle Scholar
 Sandhu KS, Li G, Poh HM, Quek YLK, Sia YY, Peh SQ, Mulawadi FH, Lim J, Sikic M, Menghi F, Thalamuthu A, Sung WK, Ruan X, Fulwood MJ, Liu E, Csermely P, Ruan Y. Largescale functional organization of longrange chromatin interaction networks. Cell Reports. 2012; 2(5):1207–19.View ArticleGoogle Scholar
 Botta M, Haider S, Leung IX, Lio P, Mozziconacci J. Intraand interchromosomal interactions correlate with CTCF binding genome wide. Mol Syst Biol. 2010; 6(1):426.Google Scholar
 Hulsman M, Dimitrakopoulos C, de Ridder J. Scalespace measures for graph topology link protein network architecture to function. Bioinformatics. 2014; 30(12):237–45.View ArticleGoogle Scholar
 Boulos R, Arneodo A, Jensen P, Audit B. Revealing longrange interconnected hubs in human chromatin interaction data using graph theory. Phys Rev Lett. 2013; 111(11):118102.View ArticleADSGoogle Scholar
 Marbouty M, Cournac A, Flot JF, MarieNelly H, Mozziconacci J, Koszul R. Metagenomic chromosome conformation capture (meta3c) unveils the diversity of chromosome organization in microorganisms. Elife. 2014; 3:03318.View ArticleGoogle Scholar
 Vendruscolo M, Kussell E, Domany E. Recovery of protein structure from contact maps. Folding Design. 1997; 2(5):295–306.View ArticleGoogle Scholar
 Serra F, Di Stefano M, Spill YG, Cuartero Y, Goodstadt M, Baù D, MartiRenom MA. Restraintbased threedimensional modeling of genomes and genomic domains. FEBS Lett. 2015; 589(20):2987–95.View ArticleGoogle Scholar
 Fraser J, Rousseau M, Shenker S, Ferraiuolo MA, Hayashizaki Y, Blanchette M, Dostie J. Chromatin conformation signatures of cellular differentiation. Genome Biol. 2009; 10:37.View ArticleGoogle Scholar
 Hirata Y, Horai S, Aihara K. Reproduction of distance matrices and original time series from recurrence plots and their applications. Eur Phys J Special Topics. 2008; 164(1):13–22.View ArticleADSGoogle Scholar
 Havel TF, Kuntz I, Crippen GM. The theory and practice of distance geometry. Bull Math Biol. 1983; 45:665–720.MathSciNetView ArticleMATHGoogle Scholar
 Torgerson WS. Multidimensional scaling: I. theory and method. Psychometrika. 1952; 17(4):401–19.MathSciNetView ArticleMATHGoogle Scholar
 Zhang Z, Li G, Toh KC, Sung WK. 3D chromosome modeling with semidefinite programming and hic data. J Comput Biol. 2013; 20(11):831–46.MathSciNetView ArticleGoogle Scholar
 Varoquaux N, Ay F, Noble W, Vert JP. A statistical approach for inferring the threedimensional structure of the genome. Bioinformatics. 2014; 30:26–33.View ArticleGoogle Scholar
 Wickelmaier F, Vol. 46. An introduction to MDS. Denmark: Sound Quality Research Unit, Aalborg University; 2003.Google Scholar
 Sammon JW. A nonlinear mapping for data structure analysis. IEEE Trans Comput. 1969; 18(5):401–409.View ArticleADSGoogle Scholar
 Kruskal JB. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika. 1964; 29(1):1–27.MathSciNetView ArticleMATHGoogle Scholar
 Bécavin C, Tchitchek N, MintsaEya C, Lesne A, Benecke A. Improving the efficiency of multidimensional scaling in the analysis of highdimensional data using singular value decomposition. Bioinformatics. 2011; 27(10):1413–21.View ArticleGoogle Scholar