Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Next Article in Journal
Entropy, Statistical Evidence, and Scientific Inference: Evidence Functions in Theory and Applications
Next Article in Special Issue
A 3D Approach Using a Control Algorithm to Minimize the Effects on the Healthy Tissue in the Hyperthermia for Cancer Treatment
Previous Article in Journal
A Novel Color Image Encryption Algorithm Based on 5-D Hyperchaotic System and DNA Sequence
Previous Article in Special Issue
Forecasting COVID-19 Epidemic Trends by Combining a Neural Network with Rt Estimation
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Design and Implementation of a New Local Alignment Algorithm for Multilayer Networks

by
Marianna Milano
*,†,
Pietro Hiram Guzzi
and
Mario Cannataro
Department of Medical and Surgical Sciences, Data Analytics Research Center, University Magna Græcia, 88100 Catanzaro, Italy
*
Author to whom correspondence should be addressed.
These authors contributed equally to this work.
Entropy 2022, 24(9), 1272; https://doi.org/10.3390/e24091272
Submission received: 9 August 2022 / Revised: 6 September 2022 / Accepted: 7 September 2022 / Published: 9 September 2022

Abstract

:
Network alignment (NA) is a popular research field that aims to develop algorithms for comparing networks. Applications of network alignment span many fields, from biology to social network analysis. NA comes in two forms: global network alignment (GNA), which aims to find a global similarity, and LNA, which aims to find local regions of similarity. Recently, there has been an increasing interest in introducing complex network models such as multilayer networks. Multilayer networks are common in many application scenarios, such as modelling of relations among people in a social network or representing the interplay of different molecules in a cell or different cells in the brain. Consequently, the need to introduce algorithms for the comparison of such multilayer networks, i.e., local network alignment, arises. Existing algorithms for LNA do not perform well on multilayer networks since they cannot consider inter-layer edges. Thus, we propose local alignment of multilayer networks (MultiLoAl), a novel algorithm for the local alignment of multilayer networks. We define the local alignment of multilayer networks and propose a heuristic for solving it. We present an extensive assessment indicating the strength of the algorithm. Furthermore, we implemented a synthetic multilayer network generator to build the data for the algorithm’s evaluation.

1. Introduction

Network theory is one of the most important frameworks for a meaningful description and an efficient analysis of many complex systems [1,2,3]. Most popular analysis algorithms comprise the mining of a single network, e.g., using community detection algorithms [4]. In parallel, the comparison of networks has led to the introduction of many algorithms for comparing the structures on both a global and a local scale [5,6] that fall into the class of network alignment (NA) algorithms. The first class of algorithms, also known as global network alignment (GNA) algorithms, aims to find the overall similarity among networks. Differently, algorithms belonging to the second group are called local network alignment (LNA) algorithms and aim to find (relatively) small regions of similarity. The output of LNA algorithms is a set of matched regions (or subgraphs) among two graphs given as the input.
More recently, in many application fields, e.g., mobile and social networks and connectomics and metabolomics studies, the need for introducing models more complex than traditional networks arises [7,8]. In such contexts, nodes may have different classes of interactions among them, and such interactions may also be time-varying. In particular, networks representing multiple different associations among patients can be represented by a multilayer graph comprised of multiple interdependent graphs, where each graph represents an aspect or a set of similar interactions [9,10]. Figure 1 represents a simple multilayer graph with three layers. Each layer is a different graph G. Edges of a multilayer graph can be intra-layer, i.e., connecting nodes of the same layer, and inter-layer, i.e., connecting nodes of two different layers [11,12].
Formally, a multilayer network graph may be described as a tuple G m l = V L , E i n t r a L , E i n t e r L x L , where L = { 0 , 1 , , l } is a set of layers and E i n t e r L x L is a set of edges among layers. For each layer k, we have a graph V k , E i n t r a k (intra-layer edges), and for each pair of layers, k, h we have a set of edges E i n t e r v x k , which is a set of layers connecting nodes of the layers v and k [13].
Examples of multilayer networks come from many different fields, from social network analysis to biological networks. For instance, Figure 2 represents an example of a biological multilayer network representing the interplay among diseases, genes, and drugs.
While many efforts have been made to address challenges related to the analysis of a single network, i.e., community detection in multilayer graphs, there is a need for the formalization and introduction of algorithms to compare multilayer networks. A simple strategy is an adaption or the simple use of existing algorithms for LNA. Unfortunately, this strategy is unsuitable, as previously demonstrated also in heterogeneous networks [14] because the the current algorithms are not able to manage the difference among layers.
Thus, we propose local alignment of multilayer networks (MultiLoAl), a novel algorithm for the local alignment of multilayer networks. We define the local alignment of multilayer networks and propose a heuristic for solving it. MultiLoAl is based on an extension of the previous L-HetNetAligner [14], so it is based on the following steps, as depicted in Figure 3. Our algorithm receives two multilayer networks and a set of similarity relationships among nodes of the same layer in both networks used as the seed to build the alignment.
For instance, considering biological networks, similarity relations are represented by orthologs. The user may find these relations in databases of orthologs (e.g., OrthoMCL, etc.). It produces a set of multilayer graphs representing single local regions of the alignment.
The algorithm merges two input multilayer graphs into a single one, named the multilayer alignment graph, a multilayer graph with the same number of layers of the two inputs, and each layer represents an alignment graph of the same layer of the two input ones. For each node of a layer k, the alignment graph features pairs of nodes of the input ones. After building each alignment graph for each layer separately, we analyse the two input graphs to add inter-layer edges of the multilayer alignment graph. Finally, the algorithm uses a community detection algorithm suitable for multilayer graphs to detect communities representing local regions of similarity, i.e., a single region of the alignment. The result of our algorithm is a list of mappings among a subset of nodes of two networks, i.e., a set of mapped regions among input graphs.
We also realized a preliminary implementation of our algorithm by using the R programming language. We here refined such an implementation even in a high performance computing (HPC) fashion and provided deeper experimentation on a larger dataset. The main contributions of this paper are: (i) the implementation of a novel algorithm for the local alignment of multilayer networks, (ii) the definition of the local alignment of multilayer networks, (iii) the solution of a heuristic for solving it, and (iv) the implementation of a synthetic multilayer network generator to build the data for the algorithm evaluation.
The rest of this paper is organized as follows. Section 2 discusses the background on multilayer networks and multilayer community detection. Section 3 presents the MultiLoAl algorithm. Section 4 presents and discusses the results. Finally, Section 5 concludes the paper.

2. Related Work

2.1. Alignment of Multilayer Networks

The alignment of networks aims to compare two or more networks [5], and existing algorithms may be categorised as local or global based on the approach, despite the existence of other classifications, i.e., algorithms for specific networks such as heterogeneous or temporal networks. Local network alignment (LNA) algorithms aim to find some similar (relatively) small subnetworks, while global network alignment (GNAs) algorithms search for the best superimposition of the whole compared networks. The literature contains many algorithms for other classes of networks (see, for instance, [5]). Unfortunately, the existing algorithms do not perform very well for multilayer networks [11,15].

2.2. Community Detection in Multilayer Networks

Community detection is one of the most popular research areas in various complex systems, such as biology, sociology, medicine, and transportation systems [16,17,18]. The reason is that the community structures, defined as groups of nodes that are more densely connected than the rest of the network, represent significant characteristics for understanding the functionalities and organisations of complex systems modelled as networks [16]. It is expected that the communities play significant roles in the structure–function relationship. For example, in biological networks such as protein–protein interaction (PPI) networks, the communities represent proteins involved in a similar function; in neuroscience, the communities detected in brain networks mean regions of interest (ROI) that are active during tasks; in social networks, communities can be groups of friends or colleagues; in the World Wild Web, the communities represent the web pages sharing the same topic [19]. Thus, the discovery of communities in these systems has become an interesting approach to figuring out how network structure relates to system behaviours.
Discovering a community structure in multilayer networks has became a hot research topic due to the inability of classical community detection methods to deal with the complexity of the multilayer model.
In fact, in multilayer networks, the communities represent groups of well-connected nodes in all layers. Thus, the detection algorithms should take into account the differences among layers. Unfortunately, traditional community detection methods are not able to deal with the complexity of the multilayer networks because (i) they do not enable analysing subsets of the layers and also (ii) they do not depict the diverse layers, and thus, they cannot distinguish between different types of multilayer communities [20]. To overcome these limitations, different community detection algorithms for multilayer networks have been recently proposed. For example, Infomap [21], a multilayer generalization of [22], is a method based on random walks. It considers that an entity randomly following the edges of a network would tend to become captured within communities due to the greater density of edges between nodes within the same community, moving less frequently from one community to another. This algorithm tries to identify a partition of vertices and levels that minimises the equation of the generalised map, which measures the length of the description of a random walk on the partition.
GenLouvain [23] is a multilayer generalisation of the iterative GenLouvain algorithm. This algorithm seeks a partition of the nodes and layers that maximises the multilayer modularity of the network. It searches the global information of the network, finding which are the edges of the network that contribute to the creation of the community structure; then, it applies a novel measure of edge centrality, to classify all the edges of the network concerning their proclivity to propagate information through the network itself.
ABACUS [24] is an algorithm that ensures the mining of multidimensional communities based on the extraction of frequent closed itemsets from monodimensional community memberships. At first, ABACUS considers each dimension independently, and it mines monodimensional communities. After that, it labels each node with a list of pair tags, i.e., the dimension and community the node belongs to in that dimension. Then, ABACUS considers each pair of tags as an item, and it applies a frequent-closed-itemset-mining algorithm. Finally, the multidimensional communities described by the itemsets consist of frequent closed itemsets.
Multilayer clique percolation [25] is a method that extends the popular clique percolation method for simple networks, where dense regions correspond to cliques and adjacency consists of having common nodes. The algorithm extends this step by searching cliques by encompassing multiple layers and reformulating adjacency so that both common nodes and common layers are expected. Multilayer clique percolation communities are combinations of adjacent cliques, so all the edges in these cliques can be considered part of the community.
Multidimensional label propagation (mdlp) [26] is an algorithm based on label propagation. At first, the algorithm assigns a different label to each node, and then, it weights the contribution of each neighbour based on their similarity with the nodes on the different layers. In particular, if two nodes are adjacent on all layers and have the same neighbours, they would have a higher probability of sharing the same label. Finally, the algorithm gives a score for each pair of nodes, referring to how likely a label should be extended from one to the other, bringing a common community.

3. MultiLoAl Algorithm

Initially, the algorithm inputs two multilayer networks and a set of similarities among node pairs of the same layer into the input networks. Then, it builds the alignment by performing two main steps: (i) construction of the multilayer alignment graph and (ii) mining of the multilayer alignment graph.
MultiLoAl analyses separately each corresponding pair of the corresponding layers of the input graphs. Each pair of a network of the same layer builds an alignment graph, as previously shown in L-HetNetAligner [14]. Then, it analyses the inter-layer edges of the input networks to add inter-layer edges to the multilayer alignment graph. Once the alignment graph is built, we use an algorithm for detecting communities in multilayer networks to uncover relevant modules. Figure 4 shows these steps.
MultiLoAl is a novel algorithm for the local alignment of multilayer networks. MultiLoAl builds the alignment on two main steps, as depicted in Figure 3:
  • (i) Construction of the multilayer alignment graph;
  • (ii) Analysis of the alignment graph.
Step (i) may be subdivided into two substeps: (i.a) adding nodes and intra-layer edges; (i.b) adding inter-layer edges.
Let us consider two multilayer input graphs G 1 and G 2 .
Node colours are used to distinguish different types of nodes belonging to two different types of layers. For simplicity, two multilayer input networks have the same number of nodes.

3.1. Step (1.a): Adding Nodes and Intra-Layer Edges to the Alignment Graph

In the first step, the algorithm considers each pair of corresponding layers separately see Figure 5. For each layer, it builds an alignment graph following the approach proposed in L-HetNetAligner [14], adapted to the case of one-colour networks, as reported in Figure 6.
At this stage, the algorithm, starting from an initial list of seed nodes, builds the alignment graph by initially constructing two intermediate alignment graphs, which we call alignment graph layer 1 and alignment graph layer 2, for two networks belonging to layer 1 and two networks belonging to layer 2. Therefore, we define the alignment graph G a l = ( V a l , E a l ) as a graph constructed by two initial input graphs G 1 = ( V 1 , E 1 ) and G 2 = ( V 2 , E 2 ) . Each node v a l V a l represents the matching of nodes of the input graphs, so V a l V 1 × V 2 . The selection of node pairs is guided by the input similarity relationships. Therefore, each node is matched with the most similar node of the other network through the use of the input similarity relationships, i.e., seed nodes; each node of the alignment graph represents a pair of similarities among nodes from the input networks; see Figure 7.
Once all nodes have been added to the graph, the algorithm builds the edges of the alignment graphs. For each pair of nodes, the algorithm examines the two input graphs, and it inserts and weights the edges considering three conditions: match, mismatch, and gap. Let us consider the nodes of the alignment graphs; in particular, let us consider the pair of nodes ( G 1 G 1 ) and ( G 2 G 2 ) in Figure 6. To determine the presence of an edge, we consider the edge ( G 1 , G 2 ) G 1 network and ( G 1 , G 2 ) G 2 network. If G 1 and G 2 contain these nodes and the nodes are adjacent, there is a match, which we call, for convenience, a homogeneous match, since the nodes of the two networks are of the same type (see Figure 8a).
Let us consider Δ = 2 as the node distance, i.e., the length of the shortest connecting path threshold to discriminate between gaps and mismatches. If G 1 and G 2 contain these nodes and the nodes are adjacent only in a single network, there is a mismatch, which we call a homogeneous mismatch (Figure 8b).
If G 1 and G 2 contain these nodes, the nodes are adjacent only in a single network, and they are at a distance less than Δ (gap threshold) in the other network, there is a gap, which we call a homogeneous gap (Figure 8c). After the edges of the alignment, graphs are added, and a weight is assigned to each edge by applying an ad hoc scoring function F and the gap threshold Δ . The function assigns a high score to the matches than to the mismatches and gaps. The kind of scoring function has a large significance for the resulting alignment graph and on the alignment itself. The algorithm enables the user to choose other values to optimize the quality of the results. In this work, we set the weight assigned to each edge as follows: homogeneous match equal to 1, homogeneous mismatch equal to 0.5, homogeneous gap equal to 0.2.

3.2. Step (1.b): Adding Inter-Layer Edges

The algorithm adds the inter-layer edges among multilayer alignment graph layer 1 and alignment graph layer 2. For each pair of nodes in the multilayer alignment graphs, the algorithm examines the corresponding layers of the input graphs. Let us consider the pair of nodes ( G 1 ) and ( D 4 ) in Figure 8. To determine the presence of an edge, we consider the edge ( G 1 , D 4 ) G 1 network and ( G 1 , D 4 ) G 2 network. The initial graph contains both edges connecting their internal nodes, and if the nodes are adjacent, there is a match, which we call, for convenience, a heterogeneous match, since the nodes of the two networks are of different types; see Figure 9a.
Let us consider the pair of nodes ( G 5 ) and ( D 2 ) in Figure 8b. To determine the presence of an edge, we consider the edge ( G 5 , D 2 ) G 1 network and ( G 5 , D 2 ) G 2 network. G 1 contains the edge ( G 5 , D 2 ) , while nodes G 5 and D 2 are disconnected in G 2 If the initial graph contains both edges connecting their internal nodes and the nodes are adjacent, there is a match, which we call, for convenience, a heterogeneous match, since the nodes of the two networks are of different types; see Figure 9a. Therefore, there is a heterogeneous mismatch (Figure 9b). Then, we set the weight assigned to each edge as follows: heterogeneous match equal to 0.9, heterogeneous mismatch equal to 0.4.

3.3. Step 2: Detection of Communities on the Alignment Graph

Finally, the final alignment graph is then mined to discover communities by applying a community detection algorithm by using existing algorithms for multilayer networks [27,28,29,30], see Figure 10. Our methodology presents a general design, so it is possible to mine the final alignment graph by applying a different mining method.
In the current version of MultiLoAl, we applied the Infomap algorithm to mine the communities on the alignment graph. However, the user can choose which community detection algorithm to select among Generalized Louvain, ABACUS, clique percolation, and mdlp. The output consists of a file that contains the extracted communities as a list of nodes, the weight of the edge, and the string in which it is reported if there is a homogeneous/heterogeneous match, homogeneous/mismatch, or homogeneous gap (see an example of the output at https://github.com/mmilano87/MultiLoAl (accessed on 12 August 2022)).

3.4. MultiLoAl vs. L-HetNetAligner

MultiLoAl, despite being based on the previous L-HetNetAligner, presents many different characteristics. First, the algorithms have different scopes: MultiLoAl is a local aligner of multilayer networks, while L-HetNetAligner works only on heterogeneous networks. In detail, by analysing the building of local alignment, MultiLoAl and L-HetNetAligner have two main general steps: (i) construction of the alignment graph; (ii) mining of the alignment graph. The building of the alignment graph is the first main difference among the two algorithms. In fact, MultiLoAl builds a multilayer alignment graph through two substeps: (i) by adding nodes and intra-layer edges, following the approach proposed in L-HetNetAligner adapted to the case of one-colour networks; (ii) by adding inter-layer edges. This last step represents the main novelty compared to L-HetNetAligner, because MultiLoAl analyses and adds the edges among different layers of input networks. Otherwise, L-HetNetAligner builds a heterogeneous alignment graph. Initially, L-HetNetAligner defines the nodes of the alignment graph as composite nodes representing pairs of nodes matched by the similarity considerations. The algorithm inserts and weights the edges in the alignment graph to the nodes for which the edge links have the same colour and according to their distance in the input network. Finally, once the alignment graph is built, both algorithms mine the alignment graph to discover modules that represent local alignment. MultiLoAl applies a community detection algorithm, Infomap, to mine the final alignment. The result consists of the extracted communities as a list of nodes, the weight of the edge, and the string in which it is reported if there is a homogeneous/heterogeneous match, homogeneous/mismatch, or homogeneous gap. Otherwise, L-HetNetAligner uses the Markov clustering (MCL) algorithm to cluster the graph. Each extracted module represents a single region of the alignment. The result of our algorithm is a list of mappings among a subset of nodes of two networks, i.e., a set of mapped regions among input graphs.

4. Results and Discussion

4.1. Evaluation of the Quality of the Alignment

The evaluation of the quality of the alignment of network is still a matter of debate for simple networks [5,31,32]. There exist many measures able to evaluate both the correctness of the alignment, as well as the quality of the obtained alignment [33]. On the other side, there is no gold standard to benchmark the alignment. Moreover, all the existing measures need to be extended in the multilayer case. Thus, we first introduce novel measures of correctness in the multilayer case (to the best of our knowledge, there are not any other available measures), then we perform an assessment of our methods. We first designed a proof of concept to show the ability of our algorithm to map correct nodes and edges by aligning a synthetic network with itself and with some randomised versions.
The correctness of an alignment is usually evaluated by means of the analysis of its topological quality, i.e., the ability to reconstruct the underlying true node mapping well (when such a mapping is known) and if it conserves many edges. For simple networks, F NC (F-score node correctness) is a measure of node correctness, and it is a combination of two measures: P NC and R NC . P NC is calculated as M N M , and R NC is defined as M N N , where M is the set of node pairs that are mapped under the true node mapping and N is the set of node pairs that are aligned under an alignment f.
We here extended such a measure in the multilayer case. We first considered in a separate way each layer, and we calculated the F NC for each layer . Then, we computed the multilayer F NC as the average of all F NC .
Similarly, the edge correctness in the simple case can be measured by considering NCV-G S 3 , which is a combination of two measures: high node coverage (NCV) and generalized S 3 (G S 3 ). NCV is the percentage of nodes from G 1 and G 2 that are also in G 1 and G 2 , and G S 3 measures how well edges are conserved between G 1 and G 2 , where G 1 and G 2 are two graphs and G 1 and G 2 are subgraphs of G 1 and G 2 that are induced by the mapping.
We used NCV-G S 3 to measure the edge correctness of each layer , then we averaged the measures of such values for all the layers, and we obtained the multilayer NCV-G S 3 .
Finally, we should consider the edge correctness for the inter-layer edges. Without loss of information, we considered all the inter-layer edges as a whole, and we calculated the correctness of all the inter-layer edges as NCV GS \ .

4.2. Proof of Concept

As a proof of principle, we present the use of the MultiLoAl dataset consisting of ten multilayer synthetic networks that we built with the graph generator, implemented ad hoc in the R code. An example of the multilayer network and R function are available on the web site of the project (https://github.com/mmilano87/MultiLoAl (accessed on 12 August 2022)).
All the multilayer networks have 30 nodes and 2 layers, whereas the edges are distributed as depicted in Table 1.
First, we aligned each network with respect to itself to show the ability to find known regions of similarity; second, we considered the alignment of the network with respect to an altered version of the network obtained by adding different levels of noise (5%, 10%, 15%, 20%, and 25%) by randomly removing edges from the network. The aim of the test was to demonstrate that the alignment algorithms are capable of producing high-quality alignments with an edge conservation of about 90%. Then, we implemented different versions of the MultiLoAl algorithm by varying the strategy applied to mine the community on the alignment graph. We executed the experiments on an Intel Core i5 Processor, 2.9 Ghz, with 4 Gbytes of main memory running the Ubuntu OS ver 18.04. MultiLoAl built 60 alignments, and it completed the whole process of alignment in ten seconds.
To measure the performance of the alignments built with different versions of MultiLoAl, we evaluated the quality of the results by considering the topological aspects of alignments and the number of communities found. At first, the results were evaluated by the topological quality.
We computed the NCV-G S 3 and F-NC measures for all alignment networks by considering the intra-layer and inter-layer edges. Table 2, Table 3, Table 4 and Table 5 report the results. Table 6, Table 7, Table 8 and Table 9 report the mean and standard deviation values of the NCV-G S 3 and F-NC measures for each synthetic network aligned with its noisy counterpart.
The results show that the quality of the alignment was greater when Infomap was applied to mine the community. Furthermore, increasing the noise level from 5% to 25% in the original networks caused NCV-G S 3 and F-NC to decrease.

5. Conclusions

Recently, the applications of multilayer networks in social network analysis, in finance, and in biology have been increasing. Multilayer networks can be seen as a set of networks (each network is a distinct layer) connected by inter-layer links. We here focused on the problem of comparing two multilayer networks, highlighting small local regions of similarity. Since existing algorithms for simple networks do not perform well on multilayer networks, we proposed Local Alignment of Multilayer Networks (MultiLoAl), a novel algorithm for the local alignment of multilayer networks. We proposed a heuristic for solving it. Furthermore, we performed an extensive evaluation to reveal the strength of the algorithm. Since we presented the use of MultiLoAl on multilayer synthetic networks, we plan to extend the application to real biological networks.

Author Contributions

Methodology, P.H.G. and M.M.; software, M.M.; data curation, M.M.; writing—original draft preparation, P.H.G. and M.C.; writing—review and editing, P.H.G., M.M. and M.C. funding acquisition, M.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

https://github.com/mmilano87/MultiLoAl (accessed on 6 September 2022).

Acknowledgments

This work was partially funded by the following research project funded by PON “Ricerca e Innovazione” 2014-2020 (PON R&I) Ricercatori a Tempo Determinato di tipo A (RTDA) (DM 1062 del 10/08/2021) Asse IV—Istruzione e ricerca per il recupero—REACT-EU.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Cannataro, M.; Guzzi, P.H.; Veltri, P. Protein-to-protein interactions: Technologies, databases, and algorithms. ACM Comput. Surv. (CSUR) 2010, 43, 1–36. [Google Scholar] [CrossRef]
  2. Barabasi, A.L.; Oltvai, Z.N. Network biology: Understanding the cell’s functional organization. Nat. Rev. Genet. 2004, 5, 101–113. [Google Scholar] [CrossRef] [PubMed]
  3. Gu, S.; Jiang, M.; Guzzi, P.H.; Milenković, T. Modeling multi-scale data via a network of networks. Bioinformatics 2022, 38, 2544–2553. [Google Scholar] [CrossRef] [PubMed]
  4. Fortunato, S. Community detection in graphs. Phys. Rep. Rev. Sec. Phys. Lett. 2010, 486, 75–174. [Google Scholar] [CrossRef]
  5. Guzzi, P.H.; Milenković, T. Survey of local and global biological network alignment: The need to reconcile the two sides of the same coin. Brief. Bioinform. 2018, 19, 472–481. [Google Scholar] [CrossRef]
  6. Hiram Guzzi, P.; Petrizzelli, F.; Mazza, T. Disease spreading modeling and analysis: A survey. Brief. Bioinform. 2022, 23, bbac230. [Google Scholar] [CrossRef]
  7. Boccaletti, S.; Bianconi, G.; Criado, R.; Del Genio, C.I.; Gómez-Gardenes, J.; Romance, M.; Sendina-Nadal, I.; Wang, Z.; Zanin, M. The structure and dynamics of multilayer networks. Phys. Rep. 2014, 544, 1–122. [Google Scholar] [CrossRef]
  8. Bianconi, G. Multilayer Networks: Structure and Function; Oxford University Press: Oxford, UK, 2018. [Google Scholar]
  9. Ortuso, F.; Mercatelli, D.; Guzzi, P.H.; Giorgi, F.M. Structural genetics of circulating variants affecting the SARS-CoV-2 spike/human ACE2 complex. J. Biomol. Struct. Dyn. 2021, 40, 1–11. [Google Scholar] [CrossRef]
  10. Gallo Cantafio, M.E.; Grillone, K.; Caracciolo, D.; Scionti, F.; Arbitrio, M.; Barbieri, V.; Pensabene, L.; Guzzi, P.H.; Di Martino, M.T. From single level analysis to multi-omics integrative approaches: A powerful strategy towards the precision oncology. High Throughput 2018, 7, 33. [Google Scholar] [CrossRef]
  11. Tagarelli, A.; Amelio, A.; Gullo, F. Ensemble-based community detection in multilayer networks. Data Min. Knowl. Discov. 2017, 31, 1506–1543. [Google Scholar] [CrossRef]
  12. Hammoud, Z.; Kramer, F. Multilayer networks: Aspects, implementations, and application in biomedicine. Big Data Anal. 2020, 5, 1–18. [Google Scholar] [CrossRef]
  13. Kivelä, M.; Arenas, A.; Barthelemy, M.; Gleeson, J.P.; Moreno, Y.; Porter, M.A. Multilayer networks. J. Complex Netw. 2014, 2, 203–271. [Google Scholar] [CrossRef]
  14. Milano, M.; Milenković, T.; Cannataro, M.; Guzzi, P.H. L-hetnetaligner: A novel algorithm for local alignment of heterogeneous biological networks. Sci. Rep. 2020, 10, 3901. [Google Scholar] [CrossRef]
  15. Ren, Y.; Sarkar, A.; Veltri, P.; Ay, A.; Dobra, A.; Kahveci, T. Pattern discovery in multilayer networks. IEEE ACM Trans. Comput. Biol. Bioinform. 2021, 19, 741–752. [Google Scholar] [CrossRef] [PubMed]
  16. Fortunato, S.; Hric, D. Community detection in networks: A user guide. Phys. Rep. 2016, 659, 1–44. [Google Scholar] [CrossRef]
  17. Lancichinetti, A.; Fortunato, S. Consensus clustering in complex networks. Sci. Rep. 2012, 2, 336. [Google Scholar] [CrossRef] [PubMed]
  18. Gligorijević, V.; Panagakis, Y.; Zafeiriou, S. Fusion and community detection in multi-layer graphs. In Proceedings of the 2016 23rd International Conference on Pattern Recognition (ICPR), Cancun, Mexico, 4–8 December 2016; pp. 1327–1332. [Google Scholar]
  19. Wang, Z.; Li, Z.; Yuan, G.; Sun, Y.; Rui, X.; Xiang, X. Tracking the evolution of overlapping communities in dynamic social networks. Knowl. Based Syst. 2018, 157, 81–97. [Google Scholar] [CrossRef]
  20. Magnani, M.; Hanteer, O.; Interdonato, R.; Rossi, L.; Tagarelli, A. Community detection in multiplex networks. ACM Comput. Surv. (CSUR) 2021, 54, 1–35. [Google Scholar] [CrossRef]
  21. De Domenico, M.; Lancichinetti, A.; Arenas, A.; Rosvall, M. Identifying modular flows on multilayer networks reveals highly overlapping organization in interconnected systems. Phys. Rev. X 2015, 5, 011027. [Google Scholar] [CrossRef]
  22. Rosvall, M.; Bergstrom, C.T. Maps of random walks on complex networks reveal community structure. Proc. Natl. Acad. Sci. USA 2008, 105, 1118–1123. [Google Scholar] [CrossRef] [Green Version]
  23. Jutla, I.S.; Jeub, L.G.; Mucha, P.J. A Generalized Louvain Method for Community Detection Implemented in MATLAB. 2011. Available online: http://netwiki.amath.unc.edu/GenLouvain (accessed on 7 July 2022).
  24. Berlingerio, M.; Pinelli, F.; Calabrese, F. ABACUS: Frequent pattern mining-based community discovery in multidimensional networks. Data Min. Knowl. Discov. 2013, 27, 294–320. [Google Scholar] [CrossRef]
  25. Afsarmanesh Tehrani, N.; Magnani, M. Partial and overlapping community detection in multiplex social networks. In Proceedings of the International Conference on Social Informatics; Springer: Cham, Switzerland, 2018; pp. 15–28. [Google Scholar]
  26. Boutemine, O.; Bouguessa, M. Mining community structures in multidimensional networks. ACM Trans. Knowl. Discov. Data (TKDD) 2017, 11, 1–36. [Google Scholar] [CrossRef]
  27. Kim, J.; Lee, J.G. Community detection in multi-layer graphs: A survey. ACM SIGMOD Rec. 2015, 44, 37–48. [Google Scholar] [CrossRef]
  28. Paul, S.; Chen, Y. Spectral and matrix factorization methods for consistent community detection in multi-layer networks. Ann. Stat. 2020, 48, 230–250. [Google Scholar] [CrossRef]
  29. Guzzi, P.H.; Salerno, E.; Tradigo, G.; Veltri, P. Extracting dense and connected communities in dual networks: An alignment based algorithm. IEEE Access 2020, 8, 162279–162289. [Google Scholar] [CrossRef]
  30. Huang, X.; Chen, D.; Ren, T.; Wang, D. A survey of community detection methods in multilayer networks. Data Min. Knowl. Discov. 2021, 35, 1–45. [Google Scholar] [CrossRef]
  31. Crawford, J.; Sun, Y.; Milenković, T. Fair evaluation of global network aligners. Algorithms Mol. Biol. 2015, 10, 1–17. [Google Scholar] [CrossRef]
  32. Hayes, W.B.; Mamano, N. SANA NetGO: A combinatorial approach to using Gene Ontology (GO) terms to score network alignments. Bioinformatics 2018, 34, 1345–1352. [Google Scholar] [CrossRef]
  33. Mina, M.; Guzzi, P.H. Improving the robustness of local network alignment: Design and extensive assessmentof a markov clustering-based approach. IEEE ACM Trans. Comput. Biol. Bioinform. 2014, 11, 561–572. [Google Scholar] [CrossRef]
Figure 1. Example of a multilayer network. The figure represents a simple multilayer graph with three layers. Each layer is a different graph. Edges of a multilayer graph can be intra-layer, i.e., connecting nodes of the same layer, and inter-layer, i.e., connecting nodes of two different layers.
Figure 1. Example of a multilayer network. The figure represents a simple multilayer graph with three layers. Each layer is a different graph. Edges of a multilayer graph can be intra-layer, i.e., connecting nodes of the same layer, and inter-layer, i.e., connecting nodes of two different layers.
Entropy 24 01272 g001
Figure 2. Example of a biological multilayer network.
Figure 2. Example of a biological multilayer network.
Entropy 24 01272 g002
Figure 3. Local alignment of multilayer networks.
Figure 3. Local alignment of multilayer networks.
Entropy 24 01272 g003
Figure 4. Workflow of the proposed algorithm.
Figure 4. Workflow of the proposed algorithm.
Entropy 24 01272 g004
Figure 5. The algorithm separates the input networks according to the layer type.
Figure 5. The algorithm separates the input networks according to the layer type.
Entropy 24 01272 g005
Figure 6. Example of a multilayer network. The nodes represents a set of genes and a set of diseases that belong to the gene layer and the disease layer.
Figure 6. Example of a multilayer network. The nodes represents a set of genes and a set of diseases that belong to the gene layer and the disease layer.
Entropy 24 01272 g006
Figure 7. Building of the alignment graph: node definition. The algorithm takes the two networks and a subset of node pairs matched according to a similarity function and starts to build the alignment graph. In this step, the algorithm defines the nodes of the alignment graph represented by the pair of matched nodes.
Figure 7. Building of the alignment graph: node definition. The algorithm takes the two networks and a subset of node pairs matched according to a similarity function and starts to build the alignment graph. In this step, the algorithm defines the nodes of the alignment graph represented by the pair of matched nodes.
Entropy 24 01272 g007
Figure 8. Example of homogeneous match, homogeneous mismatch, and homogeneous gap and building of alignment graph.
Figure 8. Example of homogeneous match, homogeneous mismatch, and homogeneous gap and building of alignment graph.
Entropy 24 01272 g008
Figure 9. Example of a homogeneous match, mismatch, and gap and building of inter-layer alignment graph.
Figure 9. Example of a homogeneous match, mismatch, and gap and building of inter-layer alignment graph.
Entropy 24 01272 g009
Figure 10. Example of community detection extraction on inter-layer alignment graph.
Figure 10. Example of community detection extraction on inter-layer alignment graph.
Entropy 24 01272 g010
Table 1. Characteristics of the synthetic multilayer networks.
Table 1. Characteristics of the synthetic multilayer networks.
NetworkLayersNodesEdges
N123090
N223096
N323084
N423078
N523095
N623088
N723093
N823083
N923094
N1023096
Table 2. NCV-G S 3 values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 2. NCV-G S 3 values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkNoiseNCV-G S 3 with InfomapNCV-G S 3 with Generalized LouvainNCV-G S 3 with ABACUSNCV-G S 3 with Clique PercolationNCV-G S 3 with mdlp
N100.990.9990.9750.9660.955
50.9760.9680.9470.9660.954
100.960.920.9250.9420.95
150.9260.9070.9180.8870.911
200.9140.8910.9030.8870.906
250.9030.8790.9010.8770.896
N200.9950.9960.9510.9660.976
50.9640.9920.8430.9120.829
100.9430.980.7880.8720.783
150.9410.9530.7730.8620.708
200.890.9450.7670.8020.666
250.8810.8940.7660.7920.615
N300.9950.9890.9770.9140.845
50.9720.9820.9520.8810.839
100.9610.9560.8830.8150.836
150.9370.9490.8550.8140.771
200.9330.9230.8010.730.754
250.9290.8810.7820.7260.657
N400.9790.9990.9760.8540.809
50.9790.970.9590.8490.744
100.9620.9380.8920.8270.742
150.9150.9140.8440.820.698
200.8870.8860.7970.8150.694
250.8810.8850.7880.7210.691
N500.990.9920.9320.9210.839
50.9680.970.9210.8960.822
100.9570.9630.8810.890.756
150.9520.9370.8660.8810.718
200.9350.920.8450.8170.717
250.9010.9030.7950.7410.683
N600.9940.9660.9690.8720.968
50.9680.9430.9680.810.913
100.9420.9410.9380.7960.884
150.9320.9360.8360.7650.834
200.9160.9170.8350.7510.808
250.8910.9090.760.7410.665
N700.9660.9890.9380.860.98
50.9660.9870.9320.8160.978
100.9560.9590.8550.8070.9
150.9430.9420.8460.7950.892
200.9410.8980.8410.7650.87
250.8870.8970.840.7280.683
N800.9970.960.9690.9540.969
50.980.9570.9550.8390.812
100.9660.9530.8920.8360.803
150.9560.9510.8370.8250.791
200.9530.9430.7820.790.765
250.9220.8950.7720.7260.764
N900.9970.9870.970.9560.95
50.9360.9670.9690.9370.916
100.9230.9670.950.930.786
150.9090.9340.9150.9230.726
200.8820.9280.80.8740.635
250.8790.9050.7670.7430.616
N1000.9910.9860.9480.9720.839
50.9820.9580.8720.9590.825
100.9440.9570.8530.9360.75
150.9380.9430.8390.8710.689
200.9050.9380.8380.8430.623
250.9040.9110.8110.7570.603
Table 3. NCV-G S 3 values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 3. NCV-G S 3 values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkNoiseNCV-G S 3 with InfomapNCV-G S 3 with Generalized LouvainNCV-G S 3 with ABACUSNCV-G S 3 with Clique PercolationNCV-G S 3 with mdlp
N100.760.760.7450.7110.576
50.7270.7270.6910.7060.571
100.680.680.6760.7010.551
150.6410.6410.6640.6780.537
200.6240.6240.5980.6440.513
250.6170.6170.5670.5990.506
N200.7570.7820.7250.7330.579
50.7160.7280.720.6650.578
100.7090.7110.7140.6620.555
150.6860.6390.7050.6130.551
200.6720.630.7010.5980.505
250.6390.5960.5950.5530.492
N300.7350.7710.7270.7070.6
50.7340.6930.7250.7050.592
100.7170.6620.6270.6780.592
150.6760.6560.610.6380.532
200.6630.6020.5940.6110.516
250.6180.5890.5780.5950.504
N400.6660.7880.6520.710.591
50.6540.6790.6420.7030.579
100.6450.5880.6320.6830.528
150.6430.5870.6160.6520.527
200.6380.5860.60.5830.509
250.6050.5840.5880.5760.497
N500.7910.7810.7130.7110.591
50.7590.7780.7030.7110.587
100.7210.6640.670.6660.57
150.7210.6070.6430.6260.529
200.6360.5880.6340.5880.524
250.6360.5850.5730.5640.518
N600.7990.760.6850.7390.596
50.7940.7540.6710.7080.558
100.790.7290.6050.6980.538
150.7280.6850.6040.6910.53
200.7190.6510.5710.5520.518
250.6590.6360.5580.5510.51
N700.7540.730.7030.6290.59
50.7280.6970.6750.6130.571
100.7120.6810.6540.6070.565
150.7060.6510.6310.570.56
200.6740.6170.5720.5650.502
250.6390.5920.5630.560.499
N800.7890.7830.7380.7220.584
50.7170.6590.6660.6850.548
100.6840.6230.5890.6390.54
150.6660.6220.5830.630.515
200.6180.6210.5760.6270.509
250.6150.5860.5670.620.505
N900.7670.770.7270.7280.588
50.750.7490.7220.6740.584
100.6970.7480.6030.6460.574
150.670.7290.5750.6370.547
200.6620.6690.5620.6160.518
250.6090.5840.5530.5640.504
N1000.7830.7750.7450.7370.597
50.7440.7610.6780.7250.586
100.7210.7410.6660.6540.574
150.7030.6660.6640.6530.541
200.6490.6540.6070.590.528
250.6430.6480.5980.570.508
Table 4. F-NC values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 4. F-NC values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkNoiseF-NC with InfomapF-NC with Generalized LouvainF-NC with ABACUSF-NC with Clique PercolationF-NC with mdlp
N100.6260.5750.5910.530.568
50.620.5660.5730.5180.54
100.6080.5560.5580.5110.528
150.6010.5490.5280.5020.512
200.5990.5280.5050.5010.504
250.5680.5240.5010.4920.479
N200.6430.6050.5710.5550.552
50.6430.6030.5660.550.506
100.6190.5980.5650.5370.5
150.610.5960.5420.5280.483
200.6050.5810.5340.5210.461
250.5870.5230.530.510.452
N300.6480.6090.580.5760.546
50.6090.6030.5420.5560.532
100.5990.5980.5410.5530.532
150.580.5920.5160.5160.502
200.5780.590.5160.5160.473
250.5770.570.5070.5070.47
N400.6280.6080.5840.5760.552
50.6260.5920.5670.5730.54
100.6120.5810.5580.5530.537
150.570.5570.5570.5440.528
200.5650.530.4970.5360.509
250.5590.5170.4950.5210.473
N500.6490.6010.60.5590.531
50.6480.5570.5920.5260.516
100.6370.5550.5780.4830.511
150.6350.5420.550.4820.485
200.6260.5410.5470.4720.474
250.6180.5390.5030.4710.451
N600.6360.5780.5870.5660.525
50.6330.5770.5720.560.516
100.5870.5460.5390.5530.499
150.580.5320.5270.5450.482
200.5680.5280.5190.5070.468
250.5610.5140.490.4830.467
N700.6420.6080.5720.5670.542
50.6340.6060.5680.5470.537
100.5970.5730.5430.5020.536
150.5970.5680.5210.4910.507
200.5650.5110.5050.490.501
250.5540.510.4990.4880.47
N800.6250.590.5850.5710.551
50.6010.5550.5650.5040.537
100.5910.5540.5470.4920.515
150.5830.5410.5450.4860.514
200.5520.5220.540.4830.508
250.5510.520.5330.4720.466
N900.6390.6050.5940.5680.563
50.6320.5960.5830.560.523
100.6070.5740.5790.5540.513
150.5820.5610.5550.530.498
200.5750.5420.540.5120.456
250.5740.5230.5070.5090.455
N1000.6410.5970.5810.570.505
50.6140.5790.5540.520.492
100.6030.5580.5220.510.478
150.5570.5260.5190.4940.47
200.5520.5260.5090.4910.469
250.5510.5130.4910.4840.461
Table 5. F-NC values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 5. F-NC values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkNoiseF-NC with InfomapF-NC with Generalized LouvainF-NC with ABACUSF-NC with Clique PercolationF-NC with mdlp
N100.5660.5770.5550.5480.465
50.550.5730.5090.5430.462
100.5450.5390.4960.5330.46
150.5330.5190.4740.4930.448
200.5310.5160.4720.4730.441
250.5010.4820.4670.4710.429
N200.5910.5660.5480.5390.485
50.5680.5450.5480.4980.478
100.5220.5320.5240.4970.461
150.5120.5170.5040.4770.452
200.5060.5060.4830.4710.402
250.5030.4990.4710.4550.401
N300.5610.5760.5530.5480.476
50.5590.5480.5110.5480.469
100.5530.5470.5050.5110.447
150.550.5430.4890.4960.446
200.5310.4960.4830.4670.425
250.5020.4850.4830.4630.425
N400.5960.5770.5340.5430.467
50.5610.5740.5290.5160.465
100.5470.5680.5130.510.446
150.530.5310.5110.490.442
200.5280.5050.5050.4820.44
250.5120.4820.4620.4580.402
N500.5880.5690.5380.520.47
50.5850.5240.5270.520.45
100.5850.50.5240.5120.412
150.5650.4860.4990.4720.403
200.5160.4850.4920.4630.403
250.5070.4850.4750.4550.401
N600.5780.5640.5430.5390.481
50.5650.5540.5410.5330.471
100.5540.5420.5360.4980.466
150.5520.5390.5250.4930.457
200.5360.5330.5130.4920.412
250.5180.5180.4960.4810.41
N700.5870.5250.5450.5480.488
50.5730.5130.5020.5220.479
100.5550.510.4950.520.476
150.5290.5070.4830.5140.465
200.5290.4960.480.4910.462
250.5080.4820.4710.4540.419
N800.5980.5790.5530.520.486
50.590.5630.5320.5110.473
100.5780.5630.5170.4940.439
150.560.5610.5120.4810.437
200.5350.5490.4780.480.413
250.5240.480.4610.4780.41
N900.5970.560.5380.550.479
50.590.540.5220.5370.472
100.5780.5380.4970.4970.469
150.5450.5350.4960.4760.461
200.540.5090.4790.4660.456
250.5150.5010.4640.4650.41
N1000.5980.5410.5390.5320.475
50.5660.540.5050.5310.473
100.5610.5270.4710.530.471
150.560.5090.470.4750.448
200.550.4950.4680.4560.418
250.5460.4870.4660.4510.414
Table 6. NCV-G S 3 mean and standard deviation values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 6. NCV-G S 3 mean and standard deviation values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkMeasureNCV-G S 3 with InfomapNCV-G S 3 with Generalized LouvainNCV-G S 3 with ABACUSNCV-G S 3 with Clique PercolationNCV-G S 3 with mdlp
N1mean0.9450.9270.9280.9210.929
sd0.0350.0470.0280.0420.027
N2mean0.9360.9600.8150.8680.763
sd0.0440.0380.0730.0660.130
N3mean0.9550.9470.8750.8130.784
sd0.0260.0400.0790.0770.073
N4mean0.9340.9320.8760.8140.730
sd0.0450.0460.0800.0480.046
N5mean0.9510.9480.8730.8580.756
sd0.0300.0330.0510.0670.063
N6mean0.9410.9350.8840.7890.845
sd0.0370.0200.0860.0480.105
N7mean0.9430.9450.8750.7950.884
sd0.0300.0410.0470.0450.109
N8mean0.9620.9430.8680.8280.817
sd0.0260.0240.0850.0750.077
N9mean0.9210.9480.8950.8940.772
sd0.0430.0310.0890.0790.140
N10mean0.9440.9490.8600.8900.722
sd0.0370.0250.0470.0820.100
Table 7. NCV-G S 3 mean and standard deviation values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 7. NCV-G S 3 mean and standard deviation values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkMeasureNCV-G S 3 with InfomapNCV-G S 3 with Generalized LouvainNCV-G S 3 with ABACUSNCV-G S 3 with Clique PercolationNCV-G S 3 with mdlp
N1mean0.6750.6750.6570.6730.542
sd0.0580.0580.0650.0440.029
N2mean0.6970.6810.6930.6370.543
sd0.0410.0710.0490.0630.037
N3mean0.6910.6620.6440.6560.556
sd0.0460.0660.0660.0480.043
N4mean0.6420.6350.6220.6510.539
sd0.0210.0830.0250.0590.038
N5mean0.7110.6670.6560.6440.553
sd0.0630.0920.0510.0620.033
N6mean0.7480.7030.6160.6570.542
sd0.0560.0530.0520.0830.031
N7mean0.7020.6610.6330.5910.548
sd0.0410.0510.0560.0290.038
N8mean0.6820.6490.6200.6540.534
sd0.0660.0700.0680.0410.030
N9mean0.6930.7080.6240.6440.553
sd0.0590.0700.0800.0550.035
N10mean0.7070.7080.6600.6550.556
sd0.0540.0580.0530.0680.035
Table 8. F-NC mean and standard deviation values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 8. F-NC mean and standard deviation values computed on intra-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkMeasureF-NC with InfomapF-NC with Generalized LouvainF-NC with ABACUSF-NC with Clique PercolationF-NC with mdlp
N1mean0.6040.5500.5430.5090.522
sd0.0200.0200.0370.0140.031
N2mean0.6180.5840.5510.5340.492
sd0.0220.0310.0180.0170.036
N3mean0.5990.5940.5340.5370.509
sd0.0280.0140.0270.0280.033
N4mean0.5930.5640.5430.5510.523
sd0.0320.0360.0380.0210.028
N5mean0.6360.5560.5620.4990.495
sd0.0120.0230.0360.0360.030
N6mean0.5940.5460.5390.5360.493
sd0.0330.0270.0360.0330.025
N7mean0.5980.5630.5350.5140.516
sd0.0350.0440.0310.0340.028
N8mean0.5840.5470.5530.5010.515
sd0.0290.0260.0190.0360.029
N9mean0.6020.5670.5600.5390.501
sd0.0290.0310.0320.0250.042
N10mean0.5860.5500.5290.5120.479
sd0.0380.0340.0330.0320.016
Table 9. F-NC mean and standard deviation values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
Table 9. F-NC mean and standard deviation values computed on inter-layer edges for all the alignments by applying Infomap, Generalized Louvain, ABACUS, clique percolation, and mdlp community detection algorithms.
NetworkMeasureF-NC with InfomapF-NC with Generalized LouvainF-NC with ABACUSF-NC with Clique PercolationF-NC with mdlp
N1mean0.5380.5340.4960.5100.451
sd0.0220.0360.0330.0350.014
N2mean0.5340.5280.5130.4900.447
sd0.0370.0250.0330.0290.037
N3mean0.5430.5330.5040.5060.448
sd0.0230.0350.0270.0370.021
N4mean0.5460.5400.5090.5000.444
sd0.0300.0400.0260.0300.023
N5mean0.5430.5330.5040.5060.448
sd0.0230.0350.0270.0370.021
N6mean0.5580.5080.5090.4900.423
sd0.0370.0330.0240.0300.029
N7mean0.5470.5060.4960.5080.465
sd0.0300.0150.0260.0320.024
N8mean0.5640.5490.5090.4940.443
sd0.0300.0350.0340.0180.031
N9mean0.5610.5310.4990.4990.458
sd0.0320.0220.0270.0370.025
N10mean0.5640.5170.4870.4960.450
sd0.0180.0230.0300.0390.028
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Milano, M.; Guzzi, P.H.; Cannataro, M. Design and Implementation of a New Local Alignment Algorithm for Multilayer Networks. Entropy 2022, 24, 1272. https://doi.org/10.3390/e24091272

AMA Style

Milano M, Guzzi PH, Cannataro M. Design and Implementation of a New Local Alignment Algorithm for Multilayer Networks. Entropy. 2022; 24(9):1272. https://doi.org/10.3390/e24091272

Chicago/Turabian Style

Milano, Marianna, Pietro Hiram Guzzi, and Mario Cannataro. 2022. "Design and Implementation of a New Local Alignment Algorithm for Multilayer Networks" Entropy 24, no. 9: 1272. https://doi.org/10.3390/e24091272

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop