To date, few tools for aligning protein-protein interaction networks have been suggested. These tools typically find conserved interaction patterns using various local or global alignment algorithms. However, the improvement of the speed, scalability, simplification, and accuracy of network alignment tools is still the target of new researches. In this paper, we introduce Pin-Align, a new tool for local alignment of protein-protein interaction networks. Pin-Align accuracy is tested on protein interaction networks from IntAct, DIP, and the Stanford Network Database and the results are compared with other well-known algorithms. It is shown that Pin-Align has higher sensitivity and specificity in terms of KEGG Ortholog groups.
1. Introduction
In the last decade, high throughput techniques in experimental biology produced a large amount of biological data which usually can be modeled by large networks such as metabolic networks, gene regulatory networks, and protein-protein interaction (PPI) networks [1]. Analogous to biological sequence comparison [2–4], comparing large biological networks can improve our biological understanding of them. Comparison of PPI networks is a well-studied area due to the important roles of PPI networks. The main idea behind comparison of PPI networks is to find evolutionary conserved interaction modules which describes functional relevance. Exact comparison of large PPI networks is a NP-hard problem and there is no polynomial time algorithm for it. The NP-hardness of this problem is based on the fact that this problem can be reduced to the subgraph isomorphism problem. In this way, the exact comparison of PPI networks is computationally infeasible, and PPI networks comparison is often addressed by heuristic methods [5–7].
Several algorithms have been proposed for biological network alignment. One of the first proposed network alignment algorithms is Path-Blast [8]. This algorithm finds high-score conserved pathways. After that, Sharan et al. designed Network-Blast [9] in order to identify conserved protein complexes in multiple species. Koyuturk et al. devised an evolution-based scoring scheme to detect conserved clusters, called MaWISh (maximum weight induced subgraph) [10]. Flannick et al. proposed Graemlin as the first method to detect conserved subnetworks of arbitrary structures with progressive alignment method [11]. All these mentioned algorithms are called local alignment algorithms owing to the fact that they started to find subnetworks such as pathways and protein complexes and expanded their search results to obtain the feasible alignment. In local network alignment, each protein may be mapped to several proteins. On the other hand, global network alignment algorithm is defined to detect the best overall mapping across all parts of the input networks. IsoRank is the first global network alignment algorithm aiming to maximize the overall match between two networks [12]. Flannick et al. extended Graemlin to global network alignment by using a training set of known network alignments to learn parameters for the scoring function. This novel algorithm is called Graemlin 2.0 and is claimed to have linear time complexity with the number of PPIs [13]. The next version of IsoRank, IsoRankN [14], uses Nibble-Page-Rank algorithm [15] to align input networks locally and globally. Tian and Samatova introduced a connected-components based algorithm, called HopeMap, for pairwise network alignment with the focus on fast identification of maximal conserved patterns [16]. After that, GRAAL [6] and H-GRAAL [1] are presented as the global network alignment algorithms based on seed-extend approaches and Hungarian algorithm, respectively. Both of GRAAL and H-GRAAL algorithms are capable of aligning any kind of network owing to the fact that they are specifically designed based on network structure and do not use sequence similarity data. Observing that the size of true orthologs across species is small comparing to the total number of proteins in all species, Tian and Samatova presented a different approach based on a precompiled list of homologs identified by KO terms [17].
In this paper, Pin-Align (protein interaction network alignment) algorithm is introduced as a novel local network alignment algorithm for aligning two protein-protein interaction networks. In Pin-Align, Hub Clustering and hierarchical clustering algorithms are applied in the heuristic and dynamic programming phases in order to reduce time and space complexity of the problem. Two different types of scoring, Node-Scoring and Structural-Scoring, are used to find the best local alignments. Node-Scoring is simply based on BLAST bit score [18] and Structural-Scoring is based on the importance of Bridges (Bridges are considerable edges, which take part in most of the paths and consequently conserved paths in the networks) in biological networks.
The Pin-Align algorithm is performed on the protein interaction networks of Escherichia coli (E. coli), Salmonella typhimurium (S. typhimurium), Caulobacter crescentus (C. crescentus), human, mouse, yeast, and Drosophila melanogaste (fly) extracted from IntAct [19], DIP [20], and Stanford Network database (SNDB) [21]. The obtained results are compared with other well-known local alignment algorithms NetworkBLAST [9], MaWISh [10], Graemlin [11], and Graemlin 2.0 [13], in a way that Graemlin 2.0 compares its results, because it is the only method that compares results of all local alignment tools against KEGG Ortholog (KO) groups [22].
2. Materials and Methods
Protein-protein interaction (PPI) network is defined as the set of relationships among proteins. Here, a PPI network is modeled by an undirected and weighted graph G=(V,E), where nodes correspond to proteins and each weighted edge specifies the probability that two proteins interact. For two graphs G1=(V1,E1) and G2=(V2,E2) (such as two sample networks A and B in Figure 1), the goal of network alignment is to identify possible mappings which map vertices of G1 into vertices of G2.
Two sample networks A and B.
In addition, for each mapping, the corresponding set of conserved edges is also identified. Mappings may be partial; that is, they do not need to be defined for all the nodes of the graphs. Each mapping implies a common subgraph between the two graphs. When protein a1 from graph G1 is mapped to protein a2 from graph G2, then a1 and a2 refer to the same node in the common subgraph; the edges in the common subgraph correspond to the conserved edges [23].
Pin-Align algorithm is a local network alignment algorithm, mainly designed based on the dynamic programming approach. To explain Pin-Align in more detail, first a formal definition of local network alignment is presented. Given two graphs G1(V1,E1) and G2(V2,E2), local alignment LAG1,G2(Sg1,Sg2,M) is a triplet where Sg1 and Sg2 are the subgraphs of G1 and G2, respectively, and M is defined as the mapping which aligns vertices of Sg1 into vertices of Sg2. Two sample local alignments of networks A and B of Figure 1 are shown in Figure 2(a).
(a) Two local alignments of networks A and B of Figure 1. (b) Result of merging them.
Following, the best local network alignment (BLA) is defined as a function to obtain the best local alignment for each subgraph of G1. The best local alignment of Sg1 as a subgraph of G1 with a subgraph of G2 is shown by BLAG1,G2(Sg1), which represents a local alignment with the highest score among all possible alignments between Sg1 and all possible subgraphs of G2 that can be mapped to Sg1.
Now, a naive dynamic programming algorithm for network alignment is demonstrated in the next subsection, and later Pin-Align algorithm is presented.
2.1. Dynamic Programming Algorithm to Align PPI Networks
In alignment of two weighted graphs G1 and G2 through dynamic programming technique, we generate all LAs of size one (subgraphs with a single vertex) in the first step. For obtaining LAs of larger subgraphs of G1, we merge all possible pairs of LA of smaller subgraphs. In other words, for each subgraph of G1 such as Sg1, there are several pairs of distinct subgraphs (subgraphs with no common vertices) such as (T1,T2) where Sg1=T1∪T2. Figure 2 demonstrated the merge process of two sample local alignments of networks A and B of Figure 1. We produce all possible local alignments for Sg1 such as LAG1,G2(Sg1,Sg2,M) by merging every possible local alignments of T1 such as LAG1,G2(T1,T1′,M′) with every possible local alignment of T2 such as LAG1,G2(T2,T2′,M′′). This process is done iteratively, and finally all local alignments of G1 including BLAG1,G2(G1) are obtained.
Inspired by dynamic programming technique, we propose Simple-Align algorithm in Algorithm 1. This algorithm can find all BLAs; however by considering time and memory complexity, this algorithm is not feasible. To overcome this, we use several heuristics to reduce memory and time complexity and propose a new local alignment algorithm, Pin-Align, in next section.
Algorithm 1: Simple-Align algorithm.
Input: Graph G1(V1, E1), Graph G2(V2, E2) such that |V1|=n
Output: Local alignment
(1) Define S as an array of sets with length n
(2) S[1] = Set of all local alignments of size one
(3) forsize=2 to ndo
(4) fori=1 to size-1do
(5) j=size-i
(6) fork=1 to |S[i]|do
(7) forl=1 to |S[j]|do
(8) LAA = k’th element of S[i]
(9) LAB = l’th element of S[j]
(10) S[size]=S[size]⋃Merge(A,B)
(11) end for
(12) end for
(13) end for
(14) end for
(15) return The LA with highest score among all LAs in S[n]
2.2. Pin-Align Algorithm
As explained, Pin-Align uses dynamic programming approach to solve the local network alignment problem and overcome the deficiencies of dynamic programming such as time and space complexity by applying heuristic approach. The steps of the Pin-Align algorithm are summarized as follows.
Partition the input graph G1 into smaller clusters. These clusters are dense subgraphs of G1. The partitioning is done by using a new clustering algorithm named Hub Clustering. Clusters obtained in this step are named Hub-Clusters.
For each Hub-Clusteri,
create local alignments of size one with BLAST bit score greater than 0, between vertices of Hub-Clusteri and vertices of G2; local alignments obtained by this approach are named Candidates (C=LAG1,G2(Sg1,Sg2,M) such that Sg1 is a subgraph of Hub-Clusteri with size one),
collect all Candidates of subgraph Sg1 of G1 (such as C=LAG1,G2(Sg1,Sg2,M)) in a new set called Candidate-Collection and each Candidate-Collection is represented by its subgraph of G1 such as CC(Sg1); Candidate-Collections obtained for single nodes of G1 are used as seeds in Step 2(c),
merge small Candidate-Collections (seeds) to gain larger ones. Repeat this process continuously to achieve a Candidate-Collection CC(Hub-Clusteri) as the final result of merging process. Size of a Candidate or a Candidate-Collection is defined by the number of vertices of its subgraph of G1 and cardinality of a Candidate-Collection is defined as the number of its Candidates. In this step, each Candidate-Collection can be merged with several Candidate-Collections and create Candidate-Collections with larger size. For determining which Candidate-Collections should be merged together, a new hierarchical clustering is used as a pattern of merging.
The Candidate-Collections of Hub-Clusters are named Final-Candidate-Collections.
Based on Final-Candidate-Collections, find final alignment using similarity graph. This graph is a special weighted bipartite graph where first and second set of vertices of it are shown by V1 and V2 which are the set of vertices of G1 and G2. The weight of each edge is computed based on Final-Candidate-Collections.
Flowchart of Pin-Align algorithm is presented in Figure 3 and its details are explained by an example on two networks A and B given in Figure 1.
Flowchart of Pin-Align algorithm.
In Step 1, in order to reduce the search space of the problem, input graph G1=(V1,E1) is clustered into some dense subgraphs and Step 2 of the algorithm runs for each dense subgraph separately. Generally PPI networks are sparse graphs, but they have dense regions containing high degree vertices named hubs. These regions usually contain protein complexes; therefore conserved modules among different PPIs usually exist in these regions.
To detect dense subgraphs in the graph G1, a novel clustering algorithm, called Hub Clustering, is proposed. The importance of hub nodes and density of subgraphs are two major criteria for this clustering method. First, a portion of highly connected vertices are selected as hub nodes. The vertices degree of these nodes is greater than 95% of other nodes. The hub nodes are considered as the center vertices of dense regions of graph G1 and are supposed to be initial clusters. For other nodes of G1, each node is joint to the nearest cluster (the cluster that contains its nearest hub). For this aim, the length of each edge is defined as the negative logarithm of its weight (from adjacency matrix), and then Dijkstra shortest path algorithm [24] is used to find the nearest cluster. The obtained clusters are named Hub-Clusters.
In Step 2, each of Hub-Clusters of G1 (obtained in Step 1) is aligned to G2. As mentioned, Step 2(a) creates Candidates of size one based on BLAST bit score of vertices of two graphs G1 and G2 (BLAST bit score is a normalized score which is calculated by basic local alignment search tool, based on sequence similarity [18]). Figure 4(a) shows BLAST bit scores of proteins of networks A and B.
(a) BLAST bit scores for proteins of networks A and B shown in Figure 1. (b) All possible Candidates of size one of networks A and B. (c) All possible Candidate-Collections of size one for networks A and B.
After that, all Candidates of each vertex of G1 are collected into a Candidate-Collection with size one as a seed in Step 2(b). In other words, Candidate-Collection of each vertex of G1 such as Sg1 contains all Candidates for Sg1 from G2 which can be aligned with Sg1. For example, Candidates of networks A and B of Figure 1 are shown in Figure 4(b) and those Candidate-Collections are shown in Figure 4(c).
In Step 2(c), two seeds (i.e., Candidate-Collections CC(Sgi) and CC(Sgj)) are merged in order to produce a new seed (new-CC) where its size is sum of sizes of CC(Sgi) and CC(Sgj). Figure 5 illustrates two Candidate-Collections, Candidate-Collection2 and Candidate-Collection3, of Figure 4(c) and the result of merging them.
Two Candidate-Collections and the result of merging them.
The important problem in this step is discovering the feasible combining pattern to find pairs of Candidate-Collections where merging them generates high score Candidate-Collections. In order to find high score Candidates and avoid complex searches, combining pattern for merging Candidate-Collections should produce Candidates with more conserved edges. Hence we use hierarchical clustering as combining patterns [25]. Hierarchical clusters are generally constructed by generating a sequence of partitions where each subcluster belongs to one supercluster in its entity.
In hierarchical clustering we should determine how the distance between two clusters is computed. If Si=(Vi′,Ei′) and Sj=(Vj′,Ej′) are subgraphs of G1=(V1,E1), the distance between two Candidate-Collections CC(Si) and CC(Sj) is computed, using the following formula:
(1)disCCSi,CCSj=Vi′+Vj′∑v∈Vi′∑u∈Vj′(av,u/ev,u).
In this equation, av,u is equal to one if there is an edge between vertices u and v; otherwise it is zero, and ev,u is the negative logarithm of probability of interaction between two proteins v and u in the graph G1. We use the hierarchical clustering obtained based on formula, as the combining pattern for merging Candidate-Collections. Cardinality of new Candidate-Collections obtained in higher levels of hierarchical clustering tree is increased exponentially as a result of merging smaller Candidate-Collections, so we have to save the Candidates with highest score and discard Candidates with low scores. In Step 2(c) we use a multiobjective optimization technique for finding the highest score Candidates [26]. The highest score Candidates are obtained by sorting Candidates based on two criteria, first, Nodes-Scores; next, Structural-Scores. Finally, the Candidates with the highest scores are chosen in each step. Following, we describe this scoring schema.
Node-Score. Node-score of each Candidate C=LAG1,G2(Sg1,Sg2,M) is obtained by using the following equation:
(2)NodeScore(C)=∑i=1Sg1BLASTBitScoreSg1i,MVSg1i,M.
In this equation, i is an iterator on vertices of Sg1 and Sg1i is the ith vertex of Sg1; MV(v,M) indicates a vertex of Sg2 mapped into v∈Sg1 by the mapping M.
Structural-Score. Due to the existence of hubs and small words in biological networks, certainly the network contains bridges. Bridges are considerable edges, which take part in most of the paths (specifically conserved paths) in the networks. Because of the importance of bridges, if Sg1=(V1′,E1′) is a subgraph of G1=(V1,E1) and Sg2=(V2′,E2′) is a subgraph of G2=(V2,E2), the structural scoring function for each Candidate C=LAG1,G2(Sg1,Sg2,M) is defined as follows:
(3)StructuralScore(C)=∑v∈V1′∑u∈V1′fv×fuα×PG1(v,u)PG2MVv,M,MVu,MPG1MVu,M1-α.
In this equation, f(i) shows the number of conserved edges at vertex i in the subgraph Sg1; α is a constant value between 0 and 1 and shows the importance of adjacent vertices degrees; PG(x,y) is the probability of interaction between protein x and protein y in the network G.
As mentioned before, to control the exponential growth of the cardinality of Candidate-Collections in combining process, we use a multiobjective optimization method on node score and structural score, as optimization objectives. In this way, after creating each Candidate-Collection among all Candidates obtained in that, we select only a limited number of Candidates (based on upon scoring functions) and discard others. Obtained Candidate-Collections of Hub-Clusters are named Final-Candidate-Collections which contain Final-Candidates (Candidates of Hub-Clusters).
After creating all Final-Candidate-Collections, in Step 3 of Pin-Align Algorithm, an Initial Similarity Bipartite Graph ISBG(G1,G2)=(V1,V2,E) is created which consists of vertices of G1=(V1,E1) as its first part and vertices of G2=(V2,E2) as its second part, and E is a set of weighted edges between V1 and V2. The weight of each edge represents the similarity between its incident vertices. Let C(vi,uj) be a set of all Final-Candidates which maps vi into uj. In ISBG(G1,G2), the weight of an edge e={vi,uj}, where vi∈V1 and uj∈V2, is equal to zero if |C(vi,uj)|=0; otherwise this weight is defined by the following formula:
(4)Wvi,uj=NodeSimvi,uj+NeighborhoodsSimvi,uj×1|C(vi,uj)|,
where NodeSim(vi,uj) is the similarity between two proteins vi and uj and it is calculated by using the following formula:
(5)NodeSim(vi,uj)=BLASTBitScore(vi,uj)×Cvi,uj,NeighborhoodsSim(vi,uj) is the similarity between conserved neighborhoods of vi and uj and is calculated as follows:
(6)NeighborhoodsSim(vi,uj)=∑c∈C(vi,uj)∑(vi′,uj′)∈CNc(vi,uj)BLASTBitScore(vi′,uj′).
Let NG(v) be the set of all neighbors of v (NG(v)={u∈VG∣uv∈E(G)}); then CNc(vi,uj) for Candidate c=LAG1,G2(Sg1,Sg2,M) is
(7)CNc(vi,uj)={(vi′,MV(vi′,M))∣vi′∈NG1(vi)}.
An Initial Similarity Bipartite Graph for networks A and B in Figure 1 is shown in Figure 6(a). After creating the Initial Similarity Bipartite Graph (ISBG), we can find just a single mapping for each vertex by finding maximum matching in this graph. The maximum matching can be found by applying the Hungarian algorithm (with polynomial time complexity) on the ISBG [27].
(a) The Initial Similarity Bipartite Graph (ISBG) between networks A and B in Figure 1 (the weights of edges are hidden for simplicity). (b) The ISBG, with duplication of first part. (c) Graph M1, obtained from first part of duplicated ISBG by performing Hungarian algorithm.
As mentioned, this approach can find just a single mapping for each vertex, although it cannot support some evolutionary functions such as duplication and diversion. To support duplication and diversion, we should consider matching of a single vertex from one part into two vertices of other parts in the ISBG. This type of matching is named DMatch. In other words, DMatchs are paths with length 2 such as P(vi′,vj,vk′) in the ISBG which aligns vj from one graph into both vi′ and vi′ from another graph. Here the problem is to find a set of Matchs and DMatchs from the ISBG where sum of their weights is maximum. This problem is named Maximum Weighted DMatching. To decrease the size of the solution space of this problem, we decrease the size of edge set of the ISBG by deleting some edges. The above procedure is performed by Algorithms 2 and 3.
Algorithm 2 gives the ISBG as input and creates the Final Similarity Bipartite Graph. In Steps 1–4 of Algorithm 2 all vertices of one part of the ISBG are duplicated and each duplicated vertex is connected to all vertices conjunct to the main vertex with the same weight as shown in Figure 6(b); then the Hungarian algorithm is performed on this graph in Step 5 of Algorithm 2. Obtained bipartite graph is named M1. The result of these steps for ISBG of Figure 6(a) is shown in Figure 6(c). All duplicated vertices of M1 are merged together in Step 6 of Algorithm 2. In Steps 7–12 of Algorithm 2, the same process is implemented on the other part of the obtained ISBG and the matching M2 is obtained. Figures 7(a) and 7(b) illustrate these steps for ISBG of Figure 6(a). Step 13 of Algorithm 2 combines M1 and M2 to create SimG as the Final Similarity Bipartite Braph as shown in Figure 7(c).
(a) The ISBG between networks A and B in Figure 1, with duplication of second part. (b) Graph M2, obtained from second part duplicated ISBG by performing Hungarian algorithm. (c) The Final Similarity Bipartite Graph (simG) for networks A and B.
It can be proven that Maximum Weighted Dmatchings of graph SimG are exactly similar to Maximum Weighted Dmatchings of ISBG. So SimG contains all edges of Maximum Weighted Dmatching and the degree of its vertices is at most 3. For this reason obviously the solution space of the Maximum Weighted DMatching problem on SimG is considerably decreased in comparison with the ISBG.
To find the final alignment using the graph SimG, we use a greedy algorithm with polynomial time complexity as shown in Algorithm 3. Algorithm 3 gives the Final Similarity Bipartite Graph SimG (obtained by Algorithm 2) as input and matches all pairs of adjacent vertices where degree of both of them is one. For other vertices, Algorithm 3 creates all possible Dmatchs and selects them in a greedy manner. Algorithm 3 utilizes the following steps for all vertices of graph SimG.
If the degree of the vertex v is one and degree of adjacent vertex is also one, these two vertices are assigned to each other.
If the degree of the vertex v is two and v is adjacent to vi′ and vj′, the path P(vi′,v,vj′) is created as a potential Dmatch.
If the degree of the vertex v is three and v is adjacent to vi′, vj′, and vk′, potential Dmatchs P(vi′,v,vj′), P(vi′,v,vk′), and P(vj′,v,vk′) are created.
The weight of a given DMatch can be calculated by the following formula:
(8)WPvi′,v,vj′=W(vi,v)+W(v,vj).
After creating all potential DMatchs, we sort them based on their weights; then in a successive rounds we choose the highest weight DMatchs P(vi′,v,vj′) and delete all other edges connected to vi′, v, or vj′. By this way all assignments are found and finally the local alignment of two graphs is obtained in Steps 24–27 of Algorithm 3. Therefore the output of this algorithm is the final alignment between two given graphs G1 and G2. The final alignment between networks A and B in Figure 1 is shown in Figure 8.
Final alignment between networks A and B.
3. Results
In this section, we intend to compare our algorithm to other well-known local alignment algorithms (NetworkBLAST [9], MaWISh [10], Graemlin [11], and Graemlin 2.0 [13]), in a way that Graemlin 2.0 compares its results, because it is the only method that compares results of all local alignment tools against KEGG Ortholog (KO) groups [22] on different data sets. The data sets contain the PPI networks data from SNDB [21] for the organism pairs (E. coli, S. typhimurium) and (E. coli, C. crescentus), PPI networks data from DIP [20] for the organism pair (human, mouse), and data from IntAct [19] database for the organism pair (yeast, fly). The comparison is proceeded based on specificity and sensitivity in terms of KO groups introduced in Graemlin 2.0 [13]. In short—as Graemlin 2.0 defined—to uniquely specify an alignment, the mapping should be transitive; that is, if protein A is aligned with proteins B and C, then protein B must also be aligned with protein C. Mathematically, this means that the mapping is an equivalency relation, so groups of aligned proteins are referred to as equivalence classes. An equivalent class is defined as correct if all protein members in the class are in the same KO group. To measure specificity, Graemlin 2.0 computed two metrics:
the fraction of equivalence classes that were correct (Ceq),
the fraction of nodes that were in correct equivalence classes (Cnode),
and to measure sensitivity, it computed two metrics:
the total number of nodes that were in correct equivalence classes (Cor),
the number of equivalence classes that contained 2 species (Tot).
As we described in Step 1 of the Pin-Align algorithm, Pin-Align chooses some vertices as hubs whose degrees are greater than about 95% of total vertices. For networks S. typhimurium, C. crescentus, human, and yeast, Pin-Align chooses vertices as hubs in a way that their degrees are greater than 264, 126, 16, and 23, respectively. The reason why these degrees are greater than 95 percent of total vertices degrees is clear from cumulative frequency of their vertices degrees which is shown in charts of Figure 9. For example, in the PPI network of S. typhimurium the degrees of about 95% of vertices are less than 264, so in this network we choose vertices with degrees greater than 264 as hubs.
Cumulative frequency of vertices degrees in networks of S. typhimurium, C. crescentus, human, and yeast.
Table 1 shows Pin-Align results with different hub degrees. Column two of this table (percentage column) contains different percentages which are used for hub clustering step (Step 1).
Results of Pin-Align algorithm on four pairs of PPI networks, using different percentages in hub clustering step (Step 1).
Percentage
Ceq
Cnode
Cor
Tot
(E. coli, S. typhimurium)
60%
0.95
0.93
1651
1779
70%
0.95
0.94
1723
1835
80%
0.95
0.94
1773
1890
90%
0.95
0.94
1782
1905
(E. coli, C. crescentus)
60%
0.87
0.84
666
794
70%
0.89
0.86
665
772
80%
0.88
0.85
693
811
90%
0.89
0.87
764
876
(Human, mouse)
60%
0.78
0.72
217
299
70%
0.82
0.78
248
316
80%
0.78
0.73
226
309
90%
0.82
0.78
258
329
(Yeast, fly)
60%
0.87
0.85
258
303
70%
0.89
0.87
280
323
80%
0.91
0.88
350
396
90%
0.91
0.89
416
466
The results of the algorithms are demonstrated in Tables 2 and 3. In these tables NB stands for NetworkBlast, MW for MaWISh, Gr for Graemlin, and Gr 2.0 for Graemlin 2.0. They show that Pin-Align is the most specific and sensitive aligner in comparison with other alignment tools like NetworkBlast, MaWISh, Graemlin, and Graemlin 2.0. Because Graemlin 2.0 for each of its comparisons to other local aligner removed equivalence classes of its output that did not have a node in common with output of another local aligner, we have different results of Graemlin 2.0. Graemlin 2.0 did this due to considerations of local aligners to nodes in conserved modules and global aligners to all nodes.
PPI networks from SNDB for the organism pairs (E. coli, S. typhimurium) and (E. coli, C. crescentus).
(E. coli, S. typhimurium)
(E. coli, C. crescentus)
Ceq
Cnode
Cor
Tot
Ceq
Cnode
Cor
Tot
NB
0.77
0.49
457
1016
0.78
0.50
346
697
Gr2.0
0.95
0.94
627
667
0.79
0.78
447
573
MW
0.84
0.64
1309
2050
0.77
0.54
458
841
Gr2.0
0.97
0.96
1611
1678
0.77
0.76
553
727
Gr
0.80
0.77
985
1286
0.69
0.64
546
847
Gr2.0
0.96
0.95
1157
1217
0.82
0.81
608
750
Pin-Align
0.95
0.94
1782
1905
0.89
0.87
764
876
PPI networks from DIP and IntAct for the organism pairs (human, mouse) and (yeast, fly).
(Human, mouse)
(Yeast, fly)
Ceq
Cnode
Cor
Tot
Ceq
Cnode
Cor
Tot
NB
0.33
0.06
65
1010
0.39
0.14
43
306
Gr2.0
0.83
0.81
228
281
0.58
0.58
155
267
MW
0.59
0.36
87
241
0.45
0.37
10
27
Gr2.0
0.88
0.86
181
210
0.90
0.91
20
30
Gr
0.59
0.53
108
203
0.33
0.29
35
122
Gr2.0
0.86
0.84
151
179
0.61
0.61
57
93
Pin-Align
0.82
0.78
258
329
0.91
0.89
416
466
In Table 2, we use the PPI network data from SNDB for the organism pairs (E. coli, S. typhimurium) and (E. coli, C. crescentus). In Table 3 we use the PPI networks data from DIP and IntAct for the organism pairs (human, mouse) and (yeast, fly).
Figure 10 shows that Pin-Align also finds more correct equivalence classes than NetworkBlast, MaWISh, Graemlin, and Graemlin 2.0. By considering all the above results and comparing Pin-Align to the other algorithms, it is evident that Pin-Align is more accurate in most cases based on the Ceq, Cnode, Cor, and Tot measures.
The number of correct equivalence classes that are found in alignment of organism pairs (E. coli, S. typhimurium), (E. coli, C. crescentus), (human, mouse), and (yeast, fly) by Pin-Align, in comparison with Graemlin 2.0, Graemlin, MaWISh, and NetworkBlast.
4. Conclusions
In this paper, we presented Pin-Align, a pairwise local network alignment to improve accuracy of the alignment. Pin-Align algorithm is mainly designed based on the dynamic programming approach and has specificity and sensitivity comparable with existing tools such as NetworkBlast, Graemlin, Graemlin 2.0, and MaWISh. Our novel scoring system is based on bridges which are considerable edges due to the existence of hubs and small words in biological networks which certainly contain bridges. Bridges take part in most of the paths and consequently conserved paths in the networks. In future work, we plan to extend this approach to multiple network alignment.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
The authors would like to thank Mehdi Sadeghi for valuable comments and suggestions and Graemlin team [13] from Stanford University for sharing the datasets and helpful communication. The authors would like to acknowledge the financial support of University of Tehran for this research under Grant no. 6151184/01/04.
KuchaievO.MilenkovićT.MemiševićV.HayesW.PržuljN.Topological network alignment uncovers biological function and phylogeny20107501341135410.1098/rsif.2010.00632-s2.0-77956576710WangY.HindemittT.MayerK. F. X.Significant sequence similarities in promoters and precursors of Arabidopsis thaliana non-conserved microRNAs200622212585258910.1093/bioinformatics/btl4372-s2.0-33750426743HeP.-A.LiX.-F.YangJ.-L.WangJ.A novel descriptor for protein similarity analysis20116524454582-s2.0-79959600901MR2663722ZhangY.ChenW.A new measure for similarity searching in DNA sequences2011652477488MR26637252-s2.0-79959575331MilenkovićT.PržuljN.Uncovering biological network function via graphlet degree signatures200862572732-s2.0-49649109285MilenkovićT.NgW. L.HayesW.PržuljN.Optimal network alignment with graphlet degree vectors201091211372-s2.0-77955499935NeyshaburB.KhademA.HashemifarS.ArabS. S.NETAL: a new graph-based method for global alignment of protein-protein interaction networks201329131654166210.1093/bioinformatics/btt2022-s2.0-84879967643KelleyB. P.YuanB.LewitterF.SharanR.StockwellB. R.IdekerT.PathBLAST: a tool for alignment of protein interaction networks200432838810.1093/nar/gkh4112-s2.0-3442886313SharanR.SuthramS.KelleyR. M.KuhnT.McCuineS.UetzP.SittlerT.KarpR. M.IdekerT.Conserved patterns of protein interaction in multiple species200510261974197910.1073/pnas.04095221022-s2.0-13844314048KoyuturkM.KimY.TopkaraU.SubramaniamS.SzpankowskiW.GramaA.Pairwise alignment of protein interaction networks200613218219910.1089/cmb.2006.13.182MR22552532-s2.0-33645985807FlannickJ.NovakA.SrinivasanB. S.McAdamsH. H.BatzoglouS.Græmlin: general and robust alignment of multiple large interaction networks20061691169118110.1101/gr.52357062-s2.0-33748257786SinghR.XuJ.BergerB.Global alignment of multiple protein interaction networksProceedings of the 13th Pacific Symposium on Biocomputing (PSB '08)January 2008Kohala Coast, Hawaii, USA3033142-s2.0-40549129437FlannickJ.NovakA.DoC. B.SrinivasanB. S.BatzoglouS.Automatic parameter learning for multiple local network alignment2009168100110222-s2.0-70349215207MR253407410.1089/cmb.2009.0099LiaoC.-S.LuK.BaymM.SinghR.BergerB.IsoRankN: spectral methods for global alignment of multiple protein networks20092512i253i25810.1093/bioinformatics/btp2032-s2.0-66349108229AndersenR.ChungF.LangK.Local graph partitioning using PageRank vectorsProceedings of the 47th Annual IEEE Symposium on Foundations of Computer Science (FOCS '06)October 2006Los Alamitos, Calif, USAIEEE Computer Society47548610.1109/FOCS.2006.442-s2.0-38749085661TianW.SamatovaN. F.Pairwise alignment of interaction networks by fast identification of maximal conserved patterns20091499110TianW.SamatovaN. F.Global alignment of pairwise protein interaction networks for maximal common conserved patterns20132013116706232-s2.0-8487794529810.1155/2013/670623AltschulS. F.MaddenT. L.SchäfferA. A.Gapped BLAST and PSI-BLAST: a new generation of protein database search programs199725173389340210.1093/nar/25.17.33892-s2.0-0030801002KerrienS.Alam-FaruqueY.ArandaB.BancarzI.BridgeA.DerowC.DimmerE.FeuermannM.FriedrichsenA.HuntleyR.KohlerC.KhadakeJ.LeroyC.LibanA.LieftinkC.Montecchi-PalazziL.OrchardS.RisseJ.RobbeK.RoechertB.ThorneycroftD.ZhangY.ApweilerR.HermjakobH.IntAct-Open source resource for molecular interaction data2007351D561D56510.1093/nar/gkl9582-s2.0-33846047770XenariosI.SalwínskiŁ.DuanX. J.HigneyP.KimS.-M.EisenbergD.DIP, the Database of Interacting Proteins: a research tool for studying cellular networks of protein interactions200230130330510.1093/nar/30.1.3032-s2.0-0036088133SrinivasanB. S.NovakA. F.FlannickJ. A.BatzoglouS.McAdamsH. H.Integrated protein interaction networks for 11 microbes20063909114Lecture Notes in Computer ScienceKanehisaM.GotoS.SatoY.KawashimaM.FurumichiM.TanabeM.Data, information, knowledge and principle: back to metabolism in KEGG2014421D199D20510.1093/nar/gkt10762-s2.0-84891760956SinghR.XuJ.BergerB.Pairwise global alignment of protein interaction networks by matching neighborhood topologyProceedings of the 11th Annual International Conference on Research in Computational Molecular Biology (RECOMB '07)April 200716312-s2.0-34547416529CormenT. H.LeisersonC. E.RivestR. L.CliffordS.20093rdCambridge, UKMIT PressMR2572804SchaefferS. E.Graph clustering200711276410.1016/j.cosrev.2007.05.001ZBL063520802-s2.0-35248893285KonakA.CoitD. W.SmithA. E.Multi-objective optimization using genetic algorithms: a tutorial2006919992100710.1016/j.ress.2005.11.0182-s2.0-33745727034KuhnH. W.The Hungarian method for the assignment problem19552839710.1002/nav.3800020109MR0075510