Volume 17 Supplement 14
Proceedings of the 14th Annual Research in Computational Molecular Biology (RECOMB) Comparative Genomics Satellite Workshop: bioinformatics
Comparative genomics meets topology: a novel view on genome median and halving problems
 Nikita Alexeev†^{1}Email author,
 Pavel Avdeyev†^{1} and
 Max A. Alekseyev^{1}
https://doi.org/10.1186/s1285901612637
© The Author(s) 2016
Published: 11 November 2016
Abstract
Background
Genome median and genome halving are combinatorial optimization problems that aim at reconstruction of ancestral genomes by minimizing the number of evolutionary events between them and genomes of the extant species. While these problems have been widely studied in past decades, their solutions are often either not efficient or not biologically adequate. These shortcomings have been recently addressed by restricting the problems solution space.
Results
We show that the restricted variants of genome median and halving problems are, in fact, closely related. We demonstrate that these problems have a neat topological interpretation in terms of embedded graphs and polygon gluings. We illustrate how such interpretation can lead to solutions to these problems in particular cases.
Conclusions
This study provides an unexpected link between comparative genomics and topology, and demonstrates advantages of solving genome median and halving problems within the topological framework.
Keywords
Median problem Halving problem Breakpoint graphs Embedded graphsIntroduction
One of the key computational problems in comparative genomics is the reconstruction of ancestral genomes based on gene^{1} orders in the extant species [1–4]. Since most dramatic changes in genomic architectures are caused by genome rearrangements (such as reversals, translocations, fusions, and fissions), this problem is often posed as minimization of the total distance (i.e., the number of genome rearrangements) between extant and ancestral genomes along the branches of the evolutionary tree. The basic case of three given genomes represents the genome median problem (GMP), which asks for reconstruction of a single ancestral genome, called median genome.
Since genome rearrangements preserve the gene content, it must be restricted to genes present in all input genomes with the same multiplicity. To account for genes appearing different number of times in different genomes, one need to consider other types of evolutionary events. One of important sources of duplicated genes in genomes are the whole genome duplication (WGD) events that simultaneously duplicate each chromosome of a genome. WGD events are known to happen in evolution of yeasts [5], fishes [6], plants [7], and even mammalian species [8], which inspires the problem of reconstruction of doubled genomes, i.e., genomes immediately resulted from a WGD in the course of evolution. This problem is often posed for input genomes that have all genes present either in a single copy (ordinary genomes) or in two copies (allduplicated genomes). In the simplest form, it is known as the genome halving problem (GHP), which asks for an ordinary genome for a given allduplicated genome such that the distance between them is minimized. In the case of a given allduplicated genome and an ordinary genome, the problem, called the guided genome halving problem (GGHP), asks for an ordinary genome at the minimal total distance from both given genomes.
While the GHP admits a polynomial solution [9–11], its solution space is enormously large, which makes it impractical to obtain biologically adequate doubled genomes. The GGHP improves biological relevance by using an additional ordinary genome. Similarly, solutions for the GMP are not always biologically adequate [12–14]. Furthermore, the GGHP and GMP are known to be NPcomplete in many models of genome rearrangements. This obstacles inspire researchers to study restricted variants of the GGHP and GMP.
A recently introduced variant of the GMP, called the intermediate genome median problem (IGMP), restricts its solutions to the intermediate genomes, i.e., genomes appearing in a shortest rearrangement scenario between two of the three given genomes [13]. Similarly, for the GGHP, there exists a variant (we called it the restricted guided genome halving problem, RGGHP) that restricts the constructed doubled genomes to the GHP solution space [15]. It is worth to mention that the proposed heuristic solutions [13, 15] to the IGMP and RGGHP are based on similar ideas. We also remark that the computational complexity of these problems remain an open question.
In this study, we show that the IGMP and RGGHP are, in fact, closely related, and put them into the framework of embedded graphs and polygon gluings [16]. This framework is traditionally studied in mathematical physics and has applications in fields such as random matrices [17] and moduli space of curves [18]. It is also studied in computational geometry with applications in computer graphics and related fields [19, 20]. More recently, it has been also applied in computational biology for analysis of RNA secondary structure [21, 22]. We show that the topological reformulation of the IGMP and RGGHP leads to solving these problems in some particular cases. As a byproduct, we also determine the cardinality of the GHP solution space.
Background
Genome rearrangements and breakpoint graphs
A DoubleCutandJoin (DCJ) (also called a 2break) operation breaks a genome at two positions and glue the resulting fragments in a new order, which model common types of genome rearrangements [23, 24]. A DCJ in genome P corresponds in \(\mathfrak {G}(P)\) to the replacement of a pair of Pedges with a different pair of Pedges^{2} on the same set of four vertices.
A DCJ scenario between genomes P and Q is a sequence of DCJs transforming P into Q. A shortest such scenario has the following property:
Lemma 1
In a shortest DCJ scenario between genomes P and Q, each DCJ splits some PQcycle in their breakpoint graph into two and thus increases the number of PQcycles by one.
From Lemma 1, one can immediately get a formula for the DCJ distance (i.e., the length of a shortest DCJ scenario) between two genomes:
Theorem 2
Whole genome duplications and contracted breakpoint graphs
The definition of breakpoint graph based on edge gluing can be easily extended to genomes with duplicated genes as follows. Let A be an allduplicated genome and \(\mathfrak {G}(A)\) be the corresponding genome graph. By the definition of an allduplicated genome, the directed edges in the genome graph \(\mathfrak {G}(A)\) come in pairs that are identically labeled (Fig. 1 a). By gluing edges in these pairs, we obtain the contracted genome graph \(\hat {\mathfrak {G}}(A)\), where Aedges form cycles (since each vertex is incident to two Aedges), called Acycles. For a doubled genome 2R resulted from a WGD^{3} of an ordinary genome R, the contracted genome graph \(\hat {\mathfrak {G}}(2R)\) contains pairs of parallel Redges, called 2Redges. It is clear that 2Redges form a matching in \(\hat {\mathfrak {G}}(2R)\). Replacing 2Redges with Redges in \(\hat {\mathfrak {G}}(2R)\) transforms it into the (contracted) breakpoint graph \(\hat {\mathfrak {G}}(R) = \mathfrak {G}(R)\).
For an allduplicated genome A and an ordinary genome R composed of the same genes, the contracted breakpoint graph \(\hat {\mathfrak {G}}(A,R)\) (resp. \(\hat {\mathfrak {G}}(A,2R)\)) is defined as the superposition of \(\hat {\mathfrak {G}}(A)\) and \(\hat {\mathfrak {G}}(R)\) (resp. \(\hat {\mathfrak {G}}(2R)\)), and can be constructed in the same way as breakpoint graphs [9] (Fig. 1 b). The Aedges and Redges in \(\hat {\mathfrak {G}}(A,R)\) form Acycles and Rmatching, respectively.
The graph \(\hat {\mathfrak {G}}(A,2R)\) can be decomposed into a collection of ARcycles, called an ARcycle decomposition. We remark that there exists an exponential number of ARcycle decompositions of \(\hat {\mathfrak {G}}(A,2R)\). Below, we describe two special types of ARcycle decompositions. One is maximal ARcycle decompositions, which have the maximum possible number of ARcycles, denoted \(c_{max}(\hat {\mathfrak {G}}(A,2R))\) (Fig. 1 c). Another type of ARcycle decompositions is constructed as follows. For each Acycle in \(\hat {\mathfrak {G}}(A,2R)\), we fix some orientation. Then each Aedge becomes a directed edge. We decompose \(\hat {\mathfrak {G}}(A,2R)\) into a collection of ARcycles such that each Redge in an ARcycle connects the head of one Aedge and the tail of another. We call such ARcycle decomposition an htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\).
GHP and RGGHP
Let us recall the formulation of the GHP and discuss the structure of its solutions.
Problem
For a given allduplicated genome A, find an ordinary genome R minimizing d _{ DCJ }(A,2R).
In other words, the GHP asks for an ordinary genome R maximizing \(c_{max}(\hat {\mathfrak {G}}(A,2R))\). Existence of such genome is guaranteed by the following theorem:
Theorem 3
It was shown in [9] that the maximum of \(c_{max}(\hat {\mathfrak {G}}(A,2R))\) is achieved on genomes R such that \(\hat {\mathfrak {G}}(A,R)\) is Rnoncrossing as defined below.
For the graph \(\hat {\mathfrak {G}}(A,R)\), an Redge connecting vertices of distinct Acycles is called Rinteredge. An Redge connecting vertices of same Acycles is called Rintraedge. We represent vertices and edges of each Acycle in \(\hat {\mathfrak {G}}(A,R)\) as points and arcs on a circle, and draw all Rintraedges as straight chords inside these circles.
Definition 4

a single even Acycle (i.e., Acycle of even size) and noncrossing Rintraedges (as chords within the corresponding circle); or

a pair of odd Acycles (i.e., Acycles of odd size) with single Rinteredge and noncrossing Rintraedges.
While the condition of the graph \(\hat {\mathfrak {G}}(A,R)\) being Rnoncrossing guarantees that the genome R yields a solution to the GHP for an alldoubled genome A, this condition is not necessary, and there exist other genomes R solving the GHP (i.e., maximizing \(c_{max}(\hat {\mathfrak {G}}(A,2R))\) as in Theorem 3). Namely, while in an Rnoncrossing \(\hat {\mathfrak {G}}(A,R)\) connected components with two odd Acycles contain a single Rinteredge, other solutions may have more than one Rinteredge connecting such Acycles. The following lemma establishes a correspondence between the GHP solutions and htdecompositions of \(\hat {\mathfrak {G}}(A,2R)\).
Lemma 5
Let an ordinary genome R be a solution to the GHP for an allduplicated genome A. Then there exists an orientation of Acycles such that the htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\) is maximal.
The proof of Lemma 5 that requires the notions of nonorientable surfaces and gluings will be published elsewhere.
We remark that the maximal decomposition of an Rnoncrossing graph \(\hat {\mathfrak {G}}(A,R)\) proposed in [9] represents the htdecomposition for the clockwise orientation of Acycles (Fig. 1 c). More generally, Lemma 5 provides an important step towards a complete characterization and enumeration of the solutions to the GHP.
Since the solution space of the GHP is enormously large, one may restrict it by taking into account an additional genome and posing the following restricted problem:
Problem
Given an allduplicated genome A and an ordinary genome B, find an ordinary genome R that is a solution to the GHP for A and minimizes d _{ DCJ }(B,R).
Connection between IGMP and RGGHP
We recall the definition of an intermediate genome from [13] (Fig. 2):
Definition 6
An intermediate genome between two genomes is any genome appearing in a shortest DCJ scenario between them. In other words, a genome I is intermediate between genomes P and Q iff d _{ DCJ }(P,I)+d _{ DCJ }(I,Q)=d _{ DCJ }(P,Q).
Similarly to Rnoncrossing contracted breakpoint graphs, for ordinary genomes P, Q, I, the breakpoint graph \(\mathfrak {G}(P,Q,I)\) is called Inoncrossing if every its connected component is formed by a single PQcycle and noncrossing Iintraedges (as chords inside each PQcycle) (Fig. 2). The following theorem describes an important properties of intermediate genomes:
Theorem 7
 (1)
a genome I is intermediate between genomes P and Q,
 (2)
\(\mathfrak {G}(P,Q,I)\) is Inoncrossing,
 (3)
the total number of PI and QIcycles in \(\mathfrak {G}(P,Q,I)\) equals n+c(P,Q).
Similarly to the GHP, one can restrict the solution space of the GMP to intermediate genomes and pose the following problem:
Problem
Given genomes P, Q, and an outgroup genome R, find an intermediate genome I between genomes P and Q that minimizes d _{ DCJ }(R,I).
From Theorem 7, one can observe that the intermediate genome I plays in the IGMP a similar role to those of the ordinary genome R in the GHP. Indeed, let PQ be an artificial allduplicated genome formed by the union of genomes P and Q. Then the breakpoint graph \(\mathfrak {G}(P,Q,I)\) can be viewed as the contracted breakpoint graph \(\hat {\mathfrak {G}}(PQ,I)\), which has no odd PQcycles. If \(\mathfrak {G}(P,Q,I)\) is Inoncrossing, then \(\hat {\mathfrak {G}}(PQ,I)\) is also Inoncrossing, and \(c_{max}(\mathfrak {G}(PQ,I)) = n + k\), where k=c(P,Q) is the number of cycles in \(\hat {\mathfrak {G}}(PQ,I)\). More generally, the IGMP asks for a shortest DCJ scenario transforming the breakpoint graph \(\mathfrak {G}(P,Q,R)\) into the breakpoint graph \(\mathfrak {G}(P,Q,I)\) for some genome I such that \(\mathfrak {G}(P,Q,I)\) is Inoncrossing. Thus, the IGMP can be viewed as a particular case of the RGGHP, where all cycles are even. We remark that Lemma 5 for the IGMP can be refined as follows: the htdecomposition with respect to any orientation of PQcycles in \(\mathfrak {G}(PQ,I)\) is maximal (since all PQcycles are even), and each cycle in this decomposition is either a PIcycle or a QIcycle.
Below we will show that both RGGHP and IGMP can be formulated within the framework of embedded graphs and polygon gluings.
Methods
Embedded graphs and glued surfaces
We recall the following definition from the topological graph theory:
Definition 8

the edges do not intersect (except at the vertices);

the complement of G _{ Σ } in Σ represents a collection of regions (called faces), and each face is a polygon.^{5}
An embedded graph with m connected components is defined as the union \(\{G^{(1)}_{\Sigma _{1}},G^{(2)}_{\Sigma _{2}},\dots, G^{(m)}_{\Sigma _{m}}\}\) of m connected embedded graphs \(G^{(i)}_{\Sigma _{i}}\) (each on its own surface).
We remark that the complement of the connected embedded graph G _{ Σ } in Σ can be viewed as the result of cutting Σ along the edges of G _{ Σ }. Conversely, G _{ Σ } can be obtained by gluing the sides of its faces, which are polygons. Let us denote this collection of polygons by \(\mathcal {P}\). Since each edge of G _{ Σ } has two sides on Σ, the total number of sides in \(\mathcal {P}\) is twice the number of edges in G _{ Σ }, and the edges of G _{ Σ } define a (perfect) matching on the sides in \(\mathcal {P}\). Since the surface Σ is orientable, we can orient sides of each face clockwise. Then the matched sides of \(\mathcal {P}\) are glued in G _{ Σ } headtotail.
A polygon gluing according to a nonperfect matching is called partial. It results in an embedded graph G _{ Σ } on a surface Σ with boundary. Connected components of the boundary are called holes. In this case, some edges of G _{ Σ } represent glued pairs of sides, while the others represent nonglued sides and form holes.
where h(Σ) is the number of holes in Σ and g(Σ) is the topological genus (number of handles) of Σ. Unless G _{ Σ } is the result of a partial gluing, we have h(Σ)=0.
RGGHP and embedded graphs
We start with establishing a correspondence between contracted breakpoint graphs and embedded graphs.
Recall that for an allduplicated genome A, the Aedges in \(\hat {\mathfrak {G}}(A)\) form a collection of Acycles. Let us fix some orientation o of these Acycles. For each Acycle with k edges, we assign a kgon whose sides correspond to the cycle vertices (such that adjacent sides correspond to adjacent vertices). Then the sides of each polygon inherit labels from the corresponding cycle vertices, and the polygon itself inherits the orientation from the cycle. We denote the collection of these labeled oriented polygons by \(\mathcal {P}_{o}(A)\).
Lemma 9
Let A be an allduplicated genome, R be an ordinary genome, and o be some orientation of the Acycles. Then the vertices of G _{ o }(A,R) are in onetoone correspondence with the ARcycles in the htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\) with respect to the orientation o.
Proof
Recall that the vertices of \(\mathcal {P}_{o}(A)\) correspond to the Aedges in \(\hat {\mathfrak {G}}(A)\). Any vertex of G is an image of some vertices of \(\mathcal {P}_{o}(A)\) under gluing. Let us prove that two vertices of \(\mathcal {P}_{o}(A)\) are glued iff the corresponding Aedges belong to the same ARcycle in the htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\) (Fig. 4 c, d). Consider an arbitrary directed Aedge (U _{1},U _{2}) in \(\hat {\mathfrak {G}}(A)\). Let this edge belong to some subpath (W _{1},V _{1}), {V _{1},U _{1}}, (U _{1},U _{2}), {U _{2},V _{2}}, (V _{2},W _{2}) in ARcycle in the htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\). Note that (W _{1},V _{1}), (U _{1},U _{2}), (V _{2},W _{2}) are Aedges and {V _{1},U _{1}}, {U _{2},V _{2}} are (undirected) Redges in \(\hat {\mathfrak {G}}(A,2R)\). Then in G _{ o }(A,R) the side V _{1} is glued with U _{1} and the side V _{2} is glued with U _{2} (in headtotail fashion), and so the vertex corresponding to (U _{1},U _{2}), which is the head of the side U _{1} and the tail of the side U _{2}, is glued with the vertices corresponding to (W _{1},V _{1}) (the tail of V _{1}), and (V _{2},W _{2}) (the head of V _{2}). Conversely, since every gluing of matched sides implies gluing of vertices that correspond to Aedges from the same ARcycle, vertices that correspond to Aedges from distinct ARcycles can not be glued. By transitivity we obtain the statement of the lemma. □
Lemma 10
Let \(\mathcal {P}\) be a set of k polygons with an even number of sides (evengons) and 2l polygons with an odd number of sides (oddgons). Then the graph obtained by gluing the sides of \(\mathcal {P}\) contains at most n+k vertices, and this upper bound is achieved by the embedded graphs on k+l spheres.
Proof
So, G has a maximal number of vertices (equal n+k) iff it has k+l connected components (each on a sphere). □
We remark that Lemmas 9 and 10 provide a topological interpretation of the GHP and essentially give a new proof of Theorem 3, which is much simpler than previous ones [25, 26].
Lemma 11
Let A be an allduplicated genome, R be an ordinary genome, and o be some orientation of the Acycles. Then a DCJ on the genome R corresponds in the embedded graph G _{ o }(A,R) to cutting two edges and gluing the resulting four sides in a new order (we call such operation a DCJsurgery).
Proof
Let R ^{′} be the result of a DCJ on R. Then the Rmatching and R ^{′}matching on the sides of \(\mathcal {P}_{o}(A)\) differ only in two pairs of matched sides. The corresponding DCJsurgery on G _{ o }(A,R) cuts the two pairs of sides matched in R and glues the resulted four sides according to R ^{′}. □
Lemmas 9, 10, and 11 inspire us to pose the following problem:
Problem
Given an embedded graph G, find a shortest sequence of DCJsurgeries that results in an embedded graph G ^{′} on a maximum number of spheres.
Theorem 12
 (1)
The RGGHP for an allduplicated genome A and an ordinary genome B is equivalent to the GSP for G _{ o }(A,B), where o is some orientation of Acycles.
 (2)
The IGMP for ordinary genomes P, Q, and an outgroup genome T is equivalent to the GSP for G _{ o }(P Q,T), where o is any orientation of PQcycles.
Proof
(1) Let R be a solution to the RGGHP for an allduplicated genome A and an ordinary genome B. Let \(\mathcal {S}\) be a shortest DCJ scenario \(\mathcal {S}\) between B and R. By Lemma 5, there exists an orientation o of Acycles such that the htdecomposition of \(\hat {\mathfrak {G}}(A,2R)\) is maximal. By Lemmas 9 and 10, G _{ o }(A,R) is an embedded graph on a maximum number of spheres. By Lemma 11, the DCJ scenario \(\mathcal {S}\) corresponds to a shortest sequence of DCJsurgeries transforming G _{ o }(A,B) into G _{ o }(A,R). Thus, the RGGHP for the genomes A and B is equivalent to the GSP for the embedded graph G _{ o }(A,B).
(2) Since all PQcycles in \(\mathfrak {G}(PQ,R)\) are even, the htdecomposition of \(\mathfrak {G}(PQ,R)\) has a maximum number of PR and QRcycles for any orientation o of PQcycles. Thus, the IGMP for genomes P, Q, T is equivalent to the GSP for G _{ o }(P Q,T) with any orientation o of PQcycles. □
Results
Cardinality of the GHP solution space
Let us enumerate all the solutions to the GHP for a given allduplicated genome A. For each solution R, there exists some orientation o such that G _{ o }(A,R) is an embedded graph on the maximum number of spheres. This inspires us to define a maximal gluing as a polygon gluing that results in an embedded graph on the maximum number of spheres. By Lemma 10, each connected component of this graph has either one evengon face or two oddgon faces.
We remark that there exists a method [27] that for any collection of polygons enumerate their gluings into an embedded graph on a surface of a given genus. Since the case of spheres is much easier than the general case, we can derive explicit formulas here.
Lemma 13
The number of ways to obtain a sphere by gluing the sides of a 2kgon equals the kth Catalan number \(C_{k} = \frac {1}{k+1}\binom {2k}{k}\).
Lemma 14
Proof
Lemmas 13 and 14 lead to the following formula for the number of solutions to the GHP.
Theorem 15
Since the IGMP represents a particular case of the RGGHP, where all cycles are even and the maximal gluings correspond to the intermediate genomes, Theorem 15 implies the following corollary (first observed in [13]):
Corollary 16
For given ordinary genomes P and Q, the number of intermediate genomes equals \(\prod _{i=1}^{k} C_{n_{i}}\), where 2n _{1},…,2n _{ k } are the lengths of the PQcycles in \(\mathfrak {G}(P,Q)\).
Solving the RGGHP in a particular case
Lemma 17
Let G be an embedded graph on a torus with one face. If G contains a simple cycle of length 2l, then G can be transformed into an embedded graph on a sphere with l DCJsurgeries.
Proof
Consider a simple cycle of length 2l in G. If l>1, we apply a DCJsurgery to two adjacent edges of this cycle such that the graph remains on a torus, thus decreasing the cycle length by 2 (Fig. 5 a, b). After l−1 such DCJsurgeries, we obtain a graph on a torus with a cycle of length 2 (i.e., with l=1).
If l=1, we apply a DCJsurgery that cuts the edges of this cycle, resulting in a sphere with two holes of length 2, and then glues each of these holes, resulting in a sphere. So, we have transformed G into an embedded graph on a sphere with l DCJsurgeries. □
Lemma 18
Let G be an embedded graph on a torus with one face. If G contains two simple odd cycles that have the total length 2l and share exactly one vertex, then G can be transformed into an embedded graph on a sphere with l DCJsurgeries.
Proof
Similarly to Lemma 17, we can apply l−1 DCJsurgeries on G and obtain two loops (cycles of length 1) that share the vertex. We then apply a DCJsurgery that cuts these loops, resulting in a sphere with a hole of length 4, and then glues this hole, resulting in a sphere. So, we have transformed G into an embedded graph on a sphere with l DCJsurgeries. □
Lemma 19
 1.
Let g be the genus of the surface of G and G ^{′} be obtained from G by gluing a pair of sides from different holes. Then the surface of G ^{′} has genus g ^{′}=g+1.
 2.
If G has one face and can be glued into an embedded graph on a sphere, then G is an embedded graph on a sphere with holes of even length. Furthermore, all simple cycles in G are holes.
Proof
(1) Let G have v vertices, e edges, f faces and h holes. Let C _{1} and C _{2} be the holes that contain the pair of sides we are gluing. If at least one of the holes C _{1}, C _{2} has length greater than 1, then G ^{′} has v ^{′}=v−2 vertices, e ^{′}=e−1 edges, f ^{′}=f faces, and h ^{′}=h−1 holes. If both C _{1} and C _{2} have length 1, then G ^{′} has v ^{′}=v−1 vertices, e ^{′}=e−1 edges, f ^{′}=f faces, and h ^{′}=h−2 holes. By the Euler formula (1), we have g ^{′}=g+1 in both cases.
(2) Since G has one face, it results from a partial gluing of a polygon. Obviously, any partial gluing resulting in a sphere with holes of even length can be extended to a gluing resulting in a sphere. Let us prove that any other gluing can not be extended in such a way. Let g the genus of the surface of G. Consider a gluing of G into an embedded graph on a sphere. If g>0, such gluing does not exist, since the genus cannot be decreased by such gluing. Hence, g=0 and thus G is on a sphere with holes. If there are holes of odd lengths, then some side from one of these holes has to be glued with a side from some other hole, which would increase the genus. So, all holes must be of even length.
It remains to show that all the simple cycles in G are holes. Let L be the total length of the holes, and v and e be the number of vertices and edges of G, respectively. Consider the embedded graph G ^{′} resulting from contraction of the edges belonging to holes in G. Then G ^{′} is an embedded graph on a sphere, which has v+h−L vertices, e−L edges, and one face. From the Euler formula (1), we conclude that G ^{′} is a tree, thus all its edges are bridges. So, all edges of G except the edges belonging to the holes are bridges. □
Theorem 20
Let \(\mathcal {S}\) be a shortest sequence of DCJsurgeries transforming an embedded graph G with a single face on a torus into some embedded graph \(\tilde {G}\) on a sphere. Then there exists a cycle of length \(2\mathcal {S}\) in G.
Proof
Denote the face of G (and \(\tilde {G}\)) by F; clearly, F represents an evengon. Let M and \(\tilde {M}\) be the (perfect) matchings on the sides of F that define gluings resulting in G and \(\tilde {G}\), respectively. Let G ^{′} be the result of a partial gluing of F defined by the (nonperfect) matching \(M\cap \tilde {M}\). Then G ^{′} can be glued into each of G and \(\tilde {G}\). Since \(\tilde {G}\) is on a sphere, by Lemma 19 G ^{′} is an embedded graph on a sphere with holes of even length. Let 2m be the total length of these holes. Note that every nonglued edge in G ^{′} represents a side of an edge in G that should be cut by some DCJsurgery from \(\mathcal {S}\). Since each DCJsurgery in \(\mathcal {S}\) can create at most 4 nonglued sides, we have \(4 \mathcal {S} \geq 2m\).
Let b be a bridge (i.e., an edge whose removal disconnects the graph) in G such that its sides s _{1}, s _{2} are not glued in G ^{′}. We will show that gluing of these sides into b in G ^{′} transforms this graph into another embedded graph \(G^{\prime }_{b}\) still on a sphere with holes of even lengths. Since b is a bridge, s _{1} and s _{2} cannot belong to distinct holes in G ^{′}. Let C be a hole in G ^{′} that contains both sides s _{1} and s _{2}. In \(G^{\prime }_{b}\), C is transformed into two holes C _{1} and C _{2} (possibly empty) connected by the edge b. It is clear that the lengths of C _{1} and C _{2} have the same parity. It remains to show that both lengths are even. Assume that they are odd. Since b is a bridge, no side of C _{1} is glued with a side of C _{2} in G. Hence, at least one side from C _{1} is glued with a side from a hole different from C _{1} and C _{2}. Similarly, at least one side from C _{2} is glued with a side from a hole different from C _{1} and C _{2}. By Lemma 19, gluing of two sides from different holes creates a handle, implying that G should contain at least two handles, a contradiction to G being an embedded graph on a torus (i.e., G has exactly one handle). Thus, both holes C _{1} and C _{2} in \(G^{\prime }_{b}\) have even length, while the other holes in \(G^{\prime }_{b}\) are inherited from G ^{′}. This proves that \(G^{\prime }_{b}\) is an embedded graph on a sphere with holes of even lengths.
Let H ^{′} be an embedded graph obtained from G ^{′} by gluing all nonglued sides of bridges in G. Then H ^{′} is on a sphere with holes of even lengths. Note that any edge in G, whose sides are nonglued in H ^{′}, is not a bridge and thus belongs to some simple cycle in G.
Consider a gluing of H ^{′} into G. A handle in G can be created by gluing either two sides from distinct holes, say C _{1} and C _{2}, or from one hole, say C, in H ^{′}. In the former case, sides from C _{1} and C _{2} cannot be glued with sides from any other holes (otherwise, there would be at least two handles in G by Lemma 19). The sides from C _{ i } (i=1,2) cannot be glued with any other side from C _{ i }, since this would result in a bridge missing in H ^{′}. Thus, the sides from C _{1} and C _{2} are glued into edges that form a simple cycle in G of length 2l (equal the length of each C _{ i }). Since C _{1}+C _{2}≤2m, we have 4l≤2m. In the latter case, we claim that the edges resulted from gluing of the sides of C form two simple cycles in G, which share a vertex. Indeed, let 2p be the length of C, and H ^{′} have V+2p vertices, E+2p edges, and h holes. After gluing the sides of C (as in G), we obtain a graph on a torus with V+v vertices, E+p edges, and h−1 holes, where v vertices and p edges are obtained from vertices and edges in C and form a (possibly nonsimple) cycle \(\tilde {C}\) in G. By the Euler formula (1), we have v=p−1, and so \(\tilde {C}\) is formed by two simple cycles sharing a vertex. Clearly, either one of these simple cycles has an even length, or \(\tilde {C}\) itself has an even length. Let the even cycle have the length 2l, then 4l≤2p≤2m.
Since \(\mathcal {S}\) transforms G into \(\tilde {G}\), the above analysis implies that some cycle of length 2l should be cut by DCJsurgeries from \(\mathcal {S}\). Hence, \(4l \leq 2m \leq 4\mathcal {S}\). By Lemmas 17 and 18, we have \(\mathcal {S} \leq l\). Thus, \(\mathcal {S} = l\), and there exists a cycle of length \(2\mathcal {S}=2l\) in G. □
 1.
Construct \(\hat {\mathfrak {G}}(A,B)\) and fix an arbitrary^{6} orientation o on its Acycle.
 2.
From \(\hat {\mathfrak {G}}(A,B)\) and o, construct the embedded graph G _{ o }(A,B).
 3.
Using the breadthfirst search (BFS) starting at each vertex in G _{ o }(A,B), find a shortest even cycle C in G _{ o }(A,B).
 4.
Construct a sequence of C/2 DCJsurgeries that cut the edges of C and transform G _{ o }(A,B) into an embedded graph on a sphere.
 5.
Apply the corresponding DCJs to the genome B and return the resulting genome as a solution to the RGGHP.
We remark that our algorithm runs in polynomial time. Indeed, the most timeconsuming step is the BFS starting at each vertex of G _{ o }(A,B). Since in G _{ o }(A,B) the number of edges equals n=B=A/2 and the number of vertices equals n−1, this step runs in O(n ^{2}) time.
Discussion
In the present study we establish a somewhat unexpected link between the restricted variants of genome median and halving problems and embedded graphs. We provide a new simple proof for existence of the GHP solutions as well as completely describe the structure of the GHP solution space and determine its cardinality. We also show how the topological framework can be applied for solving the restricted guided genome halving problem (and the intermediate genome median problem) in a particular case. In further development we plan to address the topological problem of an embedded graph surgery (GSP) on an arbitrary orientable surface (i.e., a sphere with handles), which may provide better heuristic solutions for the RGGHP and IGMP.
We remark that similar topological interpretations exist for other comparative genomics problems and can provide intuition for their solution. For example, analysis of nonorientable surfaces (such as Klein bottle) seems to be relevant to the double distance problem asking for a maximal cycle decomposition of the contracted breakpoint graph of a given allduplicated genome and an ordinary genome. Also, embedded graphs on surfaces with boundaries (holes) can be related to models including genome rearrangements along with gene insertions and deletions [29, 30].
Endnotes
^{1} Some studies base their analysis on synteny blocks rather than genes. We will use the term “gene” to refer to an actual gene or a synteny block.
^{2} Here we view genome P as being transformed and Pedges as changing.
^{3} A WGD event can simultaneously duplicate each circular chromosome in genome Q either into a single circular chromosome or into two identical circular chromosomes, which have the same contracted genome graph [25]. We assume that a doubled genome 2R may contain duplicated chromosomes of both types.
^{4} Under a surface we understand a 2dimensional compact orientable manifold without boundary (e.g., a sphere or a torus). We distinguish surfaces up to homeomorphisms.
^{5} Under a polygon (ngon) we understand a topological disc, whose boundary is formed by a collection of n sides.
^{6} There exist two orientations of the Acycle in \(\hat {\mathfrak {G}}(A,B)\), both corresponding to the same htdecomposition.
Declarations
Acknowledgements
The project is supported by the National Science Foundation under the grant No. IIS1462107.
Declarations
Publication charges for this article have been funded by the National Science Foundation under Grant No. IIS1462107.
This article has been published as part of BMC Bioinformatics Vol 17 Suppl 14, 2016: Proceedings of the 14th Annual Research in Computational Molecular Biology (RECOMB) Comparative Genomics Satellite Workshop: bioinformatics. The full contents of the supplement are available online at https://bmcbioinformatics.biomedcentral.com/articles/supplements/volume17supplement14.
Availability of data and material
Not applicable.
Authors’ contributions
The research project was performed by NA and PA under the direction of MAA. All authors participated in writing this article, PA also prepared illustrations. All authors read and approved the final article.
Competing interests
The authors declare that they have no competing interests.
Consent for publication
Not applicable.
Ethics approval and consent to participate
Not applicable.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Authors’ Affiliations
References
 Gagnon Y, Blanchette M, ElMabrouk N. A flexible ancestral genome reconstruction method based on gapped adjacencies. BMC bioinforma. 2012; 13(Suppl 19):4.Google Scholar
 Hu F, Zhou J, Zhou L, Tang J. Probabilistic reconstruction of ancestral gene orders with insertions and deletions. IEEE/ACM Trans Comput Biol Bioinforma. 2014; 11(4):667–72.View ArticleGoogle Scholar
 Zheng C, Sankoff D. On the PATHGROUPS approach to rapid small phylogeny. BMC bioinforma. 2011; 12(Suppl 1):4.View ArticleGoogle Scholar
 Avdeyev P, Jiang S, Aganezov S, Hu F, Alekseyev MA. Reconstruction of ancestral genomes in presence of gene gain and loss. J Comput Biol. 2016; 23(3):150–64.View ArticlePubMedGoogle Scholar
 Kellis M, Birren BW, Lander ES. Proof and evolutionary analysis of ancient genome duplication in the yeast Saccharomyces cerevisiae. Nature. 2004; 428(6983):617–24.View ArticlePubMedGoogle Scholar
 Postlethwait JH, Yan YL, Gates MA, Horne S, Amores A, Brownlie A, Donovan A, Egan ES, Force A, Gong Z, et al. Vertebrate genome evolution and the zebrafish gene map. Nat Genet. 1998; 18(4):345–9.View ArticlePubMedGoogle Scholar
 Guyot R, Keller B. Ancestral genome duplication in rice. Genome. 2004; 47(3):610–4.View ArticlePubMedGoogle Scholar
 Dehal P, Boore JL. Two rounds of whole genome duplication in the ancestral vertebrate. PLoS Biol. 2005; 3(10):314.View ArticleGoogle Scholar
 Alekseyev MA, Pevzner PA. Colored de Bruijn graphs and the genome halving problem. IEEE/ACM Trans Comput Biol Bioinforma (TCBB). 2007; 4(1):98–107.View ArticleGoogle Scholar
 Mixtacki J. Genome Halving under DCJ Revisited In: Hu X, Wang J, editors. Computing and Combinatorics: 14th Annual International Conference, COCOON 2008. Berlin: Springer: 2008. p. 276–86, doi:10.1007/9783540697336_28.Google Scholar
 Warren R, Sankoff D. Genome halving with double cut and join. J Bioinforma Comput Biol. 2009; 7(02):357–71.View ArticleGoogle Scholar
 Haghighi M, Sankoff D. Medians seek the corners, and other conjectures. BMC Bioinforma. 2012; 13(19):1.Google Scholar
 Feijão P. Reconstruction of ancestral gene orders using intermediate genomes. BMC Bioinforma. 2015; 16(Suppl 14):3.View ArticleGoogle Scholar
 Swenson KM, Moret BM. Inversionbased genomic signatures. BMC Bioinforma. 2009; 10(1):1.View ArticleGoogle Scholar
 Zheng C, Zhu Q, Sankoff D. Genome halving with an outgroup. Evol Bioinforma. 2006; 2:295–302.Google Scholar
 Zvonkin A. Matrix integrals and map enumeration: an accessible introduction. Math Comput Model. 1997; 26(8):281–304.View ArticleGoogle Scholar
 Haagerup U, Thorbjørnsen S. Random matrices with complex gaussian entries. Expo Math. 2003; 21(4):293–337.View ArticleGoogle Scholar
 Harer J, Zagier D. The Euler characteristic of the moduli space of curves. Invent Math. 1986; 85(3):457–85.View ArticleGoogle Scholar
 Erickson J, HarPeled S. Optimally cutting a surface into a disk. Discrete Comput Geom. 2004; 31(1):37–59.View ArticleGoogle Scholar
 Colin de Verdière É. Shortening of curves and decomposition of surfaces (Raccourcissement de courbes et décomposition de surfaces). PhD thesis, Université Paris 7. 2003. http://www.di.ens.fr/~colin/textes/03thesee1.pdf.
 Penner R, Waterman MS. Spaces of RNA secondary structures. Adv Math. 1993; 101(1):31–49.View ArticleGoogle Scholar
 Andersen JE, Penner RC, Reidys CM, Waterman MS. Topological classification and enumeration of RNA structures by genus. J Math Biol. 2013; 67(5):1261–1278.View ArticlePubMedGoogle Scholar
 Yancopoulos S, Attie O, Friedberg R. Efficient sorting of genomic permutations by translocation, inversion and block interchange. Bioinformatics. 2005; 21(16):3340–346. doi:10.1093/bioinformatics/bti535.View ArticlePubMedGoogle Scholar
 Alekseyev MA, Pevzner PA. Multibreak rearrangements and chromosomal evolution. Theor Comput Sci. 2008; 395(2):193–202. doi:10.1016/j.tcs.2008.01.013.View ArticleGoogle Scholar
 Alekseyev MA, Pevzner PA. Whole genome duplications, multibreak rearrangements, and genome halving problem. In: Proceedings of the Eighteenth Annual ACMSIAM Symposium on Discrete Algorithms (SODA). Philadelphia: Society for Industrial and Applied Mathematics: 2007. p. 665–79.Google Scholar
 ElMabrouk N, Sankoff D. The reconstruction of doubled genomes. SIAM J Comput. 2003; 32(3):754–92.View ArticleGoogle Scholar
 Alexeev NV, Andersen JE, Penner RC, Zograf PG. Enumeration of chord diagrams on many intervals and their nonorientable analogs. Adv Math. 2016; 289:1056–1081.View ArticleGoogle Scholar
 Goulden IP, Slofstra W. Annular embeddings of permutations for arbitrary genus. J Comb Theory Ser A. 2010; 117(3):272–88. doi:10.1016/j.jcta.2009.11.009.View ArticleGoogle Scholar
 Braga MDV, Willing E, Stoye J. Double Cut and Join with Insertions and Deletions. J Comput Biol. 2011; 18(9):1167–1184. doi:10.1089/cmb.2011.0118.View ArticlePubMedGoogle Scholar
 Compeau P. DCJIndel sorting revisited. Algoritm Mol Biol. 2013; 8(1):6. doi:10.1186/1748718886.View ArticleGoogle Scholar