Create Presentation
Download Presentation

Download Presentation
## Double Description Method (1953)

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -

**V16 The Double Description method:Theoretical framework**behind EFM and EP /Integration Algorithms in „Combinatorics and Computer Science Vol. 1120“ edited by Deza, Euler, Manoussakis, Springer, 1996:91 Bioinformatics III**Double Description Method (1953)**• The Double Description method is the basis for simple & efficient algorithms • for the task of enumerating extreme rays. • For example, it serves as a framework for popular methods to compute • elementary flux modes and extreme pathways. • Analogy with Computer Graphics problem: • How can one efficiently describe the space • in a dark room that is lighted by a torch • shining through the open door? Bioinformatics III**Review: Duality of Matrices**This is the duality Left: all points above the dividing line (the shaded area) fulfill the condition x 0. Middle: the points in the grey area fulfill the conditions x1 0 and x2 0. But how could we describe the points in the grey area on the right side in a correspondingly simple manner? Obviously, we could define a new coordinate system (r1, r2) as a new set of generating vectors. But we could also try to transform this area back into the grey area of the middle panel and use the old axes x1 and x2. In 2D, this transformation can be obviously best performed by multiplying all vectors inside the grey area by a two-dimensional rotation matrix. Bioinformatics III**The Double Description Method**A pair (A,R) of real matrices Aand R is said to be a double description pair or simply a DD pair if the relationship A x 0 if and only if x = R for some 0 holds. The column size of Ahas to be equal to the row size of R, say d. For such a pair, the set P(A) represented by Aas is simultaneously represented by R as A subset P of d is called polyhedral cone if P = P(A) for some matrix A, and Ais called a representation matrix of the polyhedral cone P(A). Then, we say Ris a generating matrix for P. Each column vector of a generating matrix Rlies in the cone P and every vector in P is a nonnegative combination of some columns of R. Bioinformatics III**The Double Description Method**Theorem 1 (Minkowski‘s Theorem for Polyhedral Cones) For any m n real matrix A, there exists some d m real matrix Rsuch that (A,R) is a DD pair, or in other words, the cone P(A) is generated by R. The theorem states that every polyhedral cone admits a generating matrix. The nontriviality comes from the fact that the row size of R is finite. If we allow an infinite size, there is a trivial generating matrix consisting of all vectors in the cone. Also the converse is true: Theorem 2 (Weyl‘s Theorem for Polyhedral Cones) For any d n real matrix R, there exists some m d real matrix A such that (A,R) is a DD pair, or in other words, the set generated by Ris the cone P(A). Herrmann Minkowski 1864-1909 Herrmann Weyl 1885-1955 Bioinformatics III**The Double Description Method**Task: how does one construct a matrix Rfrom a given matrix A, and the converse? These two problems are computationally equivalent. Farkas‘ Lemma shows that (A,R) is a DD pair if and only if (RT,AT) is a DD pair. A more appropriate formulation of the problem is to require the minimality of R: find a matrix R such that no proper submatrix is generating P(A). A minimal set of generators is unique up to positive scaling when we assume the regularity condition that the cone is pointed, i.e. the origin is an extreme point of P(A). Geometrically, the columns of a minimal generating matrix are in 1-to-1 correspondence with the extreme rays of P. Thus the problem is also known as the extreme ray enumeration problem. No efficient (polynomial) algorithm is known for the general problem. Bioinformatics III**Double Description Method: primitive form**Suppose that the m d matrix Ais given and let (This is equivalent to the situation at the beginning of constructing EPs or EFMs where S is given.) The DD method is an incremental algorithm to construct a d m matrix R such that (A,R) is a DD pair. Let us assume for simplicity that the cone P(A) is pointed. Let Kbe a subset of the row indices {1,2,...,m}of A and let AKdenote the submatrix of A consisting of rows indexed by K. Suppose we already found a generating matrix Rfor AK, or equivalently, (AK,R) is a DD pair. If A = AK , we are done. Otherwise we select any row index i not in K and try to construct a DD pair (AK+i, R‘) using the information of the DD pair (AK,R). Once this basic procedure is described, we have an algorithm to construct a generating matrix Rfor P(A). Bioinformatics III**Geometric version of iteration step**The procedure can be understood geometrically by looking at the cut-section C of the cone P(AK) with some appropriate hyperplane h in d which intersects with every extreme ray of P(AK) at a single point. Such a cutsection is illustrated in the Figure. Here, C is the cube abcdefgh. Bioinformatics III**Geometric version of iteration step**The newly introduced inequality Aix 0 partitions the space dinto three parts: Hi+ = {x d : Aix> 0 } Hi0 = {x d : Aix = 0 } Hi- = {x d : Aix < 0 } The intersection of Hi0 with P and the new extreme points i and j in the cut-section C are shown in bold in the Figure. Bioinformatics III**Geometric version of iteration step**Let J be the set of column indices of the current generating matrix R. The rays rj (j J ) are then partitioned into three parts accordingly: J+ = {j J : rj Hi+ } J0 = {j J : rj Hi0} J- = {j J : rj Hi-} We will call the rays indexed by J+, J0, J-the positive, zero, negative rays with respect to i, respectively. To construct a matrix R‘ from R, we generate new | J+| | J-| rays lying on the ith hyperplane Hi0 - by taking an appropriate positive combination of each positive ray rj and each negative ray rj‘ and - by discarding all negative rays. Bioinformatics III**Geometric version of iteration step**The following lemma ensures that we have a DD pair (AK+i ,R‘), and provides the key procedure for the most primitive version of the DD method. Lemma 3 Let (AK,R) be a DD pair and let i be a row index of A not in K. Then the pair (AK+i ,R‘) is a DD pair, where R‘ is the d |J‘ | matrix with column vectors rj (j J‘) defined by J‘ = J+ J0 (J+ J-), and rjj‘= (Airj)rj‘ – (Airj‘)rjfor each (j,j‘) J+ J- Proof omitted. Bioinformatics III**Finding seed DD pair**It is quite simple to find a DD pair (AK,R) when |K| = 1. This can serve as the initial DD pair. Another simple (and perhaps the most efficient) way to obtain an initial DD form of P is by selecting a maximal submatrix AK of A consisting of linearly independent rows of A. The vectors rj‘s of matrix R are then obtained by solving the system of equations AK R = I where I is the identity matrix of size |K|. As we have assumed rank(A) = d, i.e. R = AK-1 , the pair (AK,R) is clearly a DD pair, since AKx 0 x = AK-1, 0. Bioinformatics III**Primitive algorithm for DoubleDescriptionMethod**This algorithm is very primitive. The straightforward implementation will be quite useless because the size of J increases extremely fast. This is because many vectors rjj‘ generated by the algorithm defined in Lemma 3 are unnessary. We need to avoid generating redundant vectors! To avoid generating redundant vectors, we will use the zero set or active set Z(x) which is the set of inequality indices satisfied by x in P(A) with equality. Noting Ai• the ith row of A, Z(x) = {i : A i• x = 0} Bioinformatics III**Towards the standard implementation**Two distinct extreme rays r and r‘ of P are adjacent if the minimal face of P containing both rays contains no other extreme rays. Proposition 7. Let r and r‘ be distinct rays of P. Then the following statements are equivalent (a) r and r‘ are adjacent extreme rays, (b) rand r‘ are extreme rays and the rank of the matrix AZ(r) Z(r‘) is d – 2, (c) if r‘‘is a ray with Z(r‘‘) Z(r) Z(r‘) then either r‘‘ ≃ ror r‘‘ ≃r ‘. Bioinformatics III**Towards the standard implementation**Lemma 8. Let (AK,R) be a DD pair such that rank(AK) = d and let i be a row index of A not in K. Then the pair (AK+i , R‘) is a DD pair, where R‘ is the d |J‘| matrix with column vectors rj (j J‘) defined by J‘ = J+ J0 Adj Adj = {(j,j‘) J+ J- : rjand rj‘ are adjacent in P(AK)} and r = (Ai rj ) rj‘ – (Airj ) rjfor each (j,j‘) Adj. Furthermore, if R is a minimal generating matrix for P(AK) then R‘ is a minimal generating matrix for P(AK+i). Bioinformatics III**Algorithm for standard form of double description method**This is now a straightforward variation of the DD method which produces a minimal generating set for P: DDMethodStandard(A) such that R is minimal Lemma 8 To implement DDMethodStandard, we must check for each pair of extreme rays rand r‘of P(AK) with Air> 0 and Ai r‘< 0 whether they are adjacent in P(AK). This completes our quick look at the Double Description method. Bioinformatics III Bioinformatics III**V16 – part II – applications of FBA and EFM**Review: (1) The concept of metabolic networks required revising the traditional picture of separate biochemical pathways into a densely-woven metabolic network (2) Connectivity of substrates in this network follows a power-law (Yeong&Barabasi). (3) Constraint-based modeling (FBA) enables to analyze the capabilities of cellular metabolism including e.g. - its capacity to predict deletion phenotypes - the ability to calculate the relative flux values of metabolic reactions, and - the capability to identify properties of alternate optimal growth states in a wide range of simulated environmental conditions Open questions - what parts of metabolism are involved in adaptation to environmental conditions? - is there a central essential metabolic core? - what role does transcriptional regulation play? Bioinformatics III**Central metabolism of E.coli characterized by EFMs**Catabolic part: substrate uptake reactions, glycolysis, pentose phosphate pathway, TCA cycle, excretion of by-products (acetate, formate, lactate, ethanol) Anabolic part: conversions of precursors into building blocks like amino acids, to macromolecules, and to biomass. Stelling et al. Nature 420, 190 (2002) Jörg Stelling ETH Zürich Bioinformatics III**Metabolic network topology phenotype**Question: Can the total number of EFMs for given conditions be used as quantitative measure of metabolic flexibility? i : deletion mutant of gene i µ : ability to grow N : number of EFMs enabling wild-type or deletion mutants in E. coli to grow Shown are results for 90 deletions of different individual genes relative to the situation for wild-type. Stelling et al. Nature 420, 190 (2002) Answer: Yes, for more than 90% of single gene deletions, the number of EFMs for the mutant strain was correctly associated with the growth phenotype. Bioinformatics III**EFM-based robustness analysis**The # of EFMs qualitatively indicates whether a mutant is viable or not, but does not describe quantitatively how well a mutant grows. Define maximal biomass yield Ymax as the optimum of: eiis the single reaction rate (growth and substrate uptake) in EFM i selected for utilization of substrate Sk. Thus, Ymax selects the EFM where most substrate medium is converted into biomass ( 0 Ymax 1). Stelling et al. Nature 420, 190 (2002) Bioinformatics III**EFM-based robustness analysis**X-axis: fraction of elementary modes operational in the mutants. Y-axis: Open squares: relative network diameter D(Δi) / D (is essentially constant) Open circles: maximal growth yield of the mutant Ymax(Δi) (open circles) Stelling et al. Nature 420, 190 (2002) → Central metabolism of E.coli behaves in a highly robust manner. Even mutants with significantly reduced metabolic flexibility ( > 15% or so) show a growth yield similar to wild type. Bioinformatics III**Distribution of fluxes in E.coli**Aim: understand principles that govern the use of individual reactions under different growth conditions. Nature 427, 839 (2004) Stoichiometric matrix for E.coli strain MG1655 containing 537 metabolites and 739 reactions was taken from Palsson et al. Apply FBA to characterize solution space (all possible flux states under a given condition). vjis the flux of reaction j and Sijis the stoichiometric coefficient of reaction j. Bioinformatics III**Optimal states**Denote the mass carried by reaction j producing (consuming) metabolite i by Observation: Fluxes vary widely: e.g. the dimensionless flux of the succinyl coenzyme A synthetase reaction is 0.185, whereas the flux of the aspartate oxidase reaction is 10.000 times smaller, 2.2 10-5. Use FBA to compute flux states that optimize cell growth on various substrates. Focus on active (non-zero flux) reactions of E.coli. Compare growth on glutamate- or succinate-rich substrate media. Bioinformatics III**Overall flux organization of E.coli metabolic network**a, Flux distribution for optimized biomass production on succinate (black) and glutamate (red) substrates. Solid line : power-law fit d, Experimentally determined fluxes for reactions of the central metabolism of E. coli. Clear power-law behaviour. Best fit with P(v) v- with = 1. Both computed and experimental flux distribution show wide spectrum of fluxes. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**Response to different environmental conditions**Is the flux distribution independent of environmental conditions? Black: Flux distribution for optimized biomass on pure succinate substrate. Red / green / blue : Flux distributions when an additional 10%, 50%, or 80% of randomly chosen subsets of the 96 input channels (substrates) are added to succinate. The flux distribution was averaged over 5,000 independent random choices of uptake metabolites. Yes, the flux distribution is independent of the external conditions. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**Use scaling behavior to determine local connectivity**• The observed flux distribution is compatible with two different potential local flux structures: • a homogenous local organization would imply that all reactions producing (consuming) a given metabolite have comparable fluxes • (b) a more delocalized „high-flux backbone (HFB)“ is expected if the local flux organisation is heterogenous such that each metabolite has a dominant source (consuming) reaction. All fluxes vij are the same, say v. One flux dominates -> replace sum by this flux vmax. Bioinformatics III Almaar et al., Nature 427, 839 (2004)**Characterizing the local inhomogeneity of the flux net**FBA-computed kY(k) as a function of k, averaged over all metabolites shows linear dependence k×Y(k) k0.73 with slope 0.73. This is true for incoming and outgoing reactions. an intermediate behavior is found between the two extreme cases discussed before. the large-scale inhomogeneity observed in the overall flux distribution is also valid at the level of the individual metabolites. The more reactions consume (produce) a given metabolite, the more likely a single reaction carries most of the flux, see inset (FAD). Inset shows non-zero mass flows producing (consuming) FAD on a glutamate-rich substrate. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**Clean up metabolic network**Use simple algorithm that removes for each metabolite systematically all reactions but the one providing the largest incoming (outgoing) flux distribution. This algorithm uncovers the „high-flux-backbone“ of the metabolism. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**High-flux backbone of E.coli metabolic network**glutamate rich medium succinate rich medium Directed links: Metabolites A and B are connected with an arc from A to B if the reaction with maximal flux consuming A is the reaction with maximal flux producing B. Shown are all metabolites that have at least one neighbour after completing this procedure. Background colours : known biochemical pathways. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**FBA-optimized high-flux backbone on glutamate-rich medium**Blue coloredMetabolites (vertices) have at least one neighbour in common in glutamate- and succinate-rich substrates. Red colored nodes have no common neighbors („rewiring“) Reactions (lines) are coloured blue if they are identical in glutamate- and succinate-rich substrates, green if a different reaction connects the same neighbour pair, and red if this is a new neighbour pair („rewiring“). Black dotted lines indicate where the disconnected pathways, e.g., folate biosynthesis (4), would connect to the cluster through a link that is not part of the HFB. Thus, the red nodes and links highlight the predicted changes in the HFB when shifting E. coli from glutamate- to succinate-rich media. Dashed lines indicate links to the biomass growth reaction. Almaar et al., Nature 427, 839 (2004) Bioinformatics III**FBA-optimized high-flux backbone on glutamate-rich medium**(1) Pentose Phospate (2) Purine Biosynthesis (3) Aromatic Amino Acids (4) Folate Biosynthesis (5) Serine Biosynthesis (6) Cysteine Biosynthesis (7) Riboflavin Biosynthesis (8) Vitamin B6 Biosynthesis (9) Coenzyme A Biosynthesis (10) TCA Cycle (11) Respiration (12) Glutamate Biosynthesis (13) NAD Biosynthesis (14) Threonine, Lysine and Methionine Biosynthesis (15) Branched Chain Amino Acid Biosynthesis (16) Spermidine Biosynthesis (17) Salvage Pathways (18) Murein Biosynthesis (19) Cell Envelope Biosynthesis (20) Histidine Biosynthesis (21) Pyrimidine Biosynthesis (22) Membrane Lipid Biosynthesis (23) Arginine Biosynthesis (24) Pyruvate Metabolism (25) Glycolysis Almaar et al., Nature 427, 839 (2004) Bioinformatics III**Interpretation**Only a few pathways appear disconnected. This indicates that although these pathways are part of the HFB, their end product is only the second-most important source for another HFB metabolite. Groups of individual HFB reactions largely overlap with traditional biochemical partitioning of cellular metabolism Almaar et al., Nature 427, 839 (2004) Bioinformatics III**How sensitive is the HFB to changes in the environment?**Fluxes of individual reactions on glutamate-rich and succinate-rich medium. Black squares: reactions belonging to the HFB, bluedots : remaining reactions Green squares : reactions in which the direction of the flux is reversed. Reactions with negligible flux changes follow the diagonal (solid line). Some reactions are turned off in only one of the conditions (shown close to the coordinate axes). Only reactions in the high-flux territory undergo noticeable differences! Type I: reactions turned on in one conditions and off in the other. Type II: reactions remain active but show an orders-in-magnitude shift in flux under the two different growth conditions. Bioinformatics III Almaar et al., Nature 427, 839 (2004)**Flux distributions for individual reactions**Shown is the flux distribution for 4 selected E. coli reactions on a 50% random medium. Reactions with small fluxes have unimodal/gaussian distributions (a and c). Shifts in growth-conditions only lead to small changes of their flux values. Off-diagonal reactions have multimodal distributions (b and d), showing several discrete flux values under diverse conditions. Triosphosphate isomerase; Carbon dioxide transport NAD kinase guanosine kinase Almaar et al., Nature 427, 839 (2004) Bioinformatics III**Summary**Metabolic network use is highly uneven (power-law distribution) at the global level and at the level of the individual metabolites. Whereas most metabolic reactions have low fluxes, the overall activity of the metabolism is dominated by several reactions with very high fluxes. E. coli responds to changes in growth conditions by reorganizing the rates of selected fluxes predominantly within this high-flux backbone. Apart from minor changes, the use of the other pathways remains unaltered. Bioinformatics III**The same authors as before used FBA to examine utilization**and relative flux rate of each metabolite in various simulated environmental conditions for E.coli, H. pylori and S. cerevisae: For each system they considered 30.000 randomly chosen combinations where each uptake reaction is assigned a random value between 0 and 20 mmol/g/h. adaptation to different conditions occurs by 2 mechanisms: (a) flux plasticity: changes in the fluxes of already active reactions. E.g. changing from glucose- to succinate-rich conditions alters the flux of 264 E.coli reactions by more than 20% (b) less often, adaptation includes structural plasticity, turning on previously zero-flux reactions or switching off active pathways. Bioinformatics III**Emergence of the Metabolic Core**The two adaptation mechanisms enable a group of reactions that are not subject to structural plasticity to be active under all environmental conditions. Are these core reactions randomly distributed? If typically a fraction q of the metabolic reactions were active under a specific growth condition, we would expect for n distinct conditions an overlap of at least qn reactions. This converges quickly to 0. Bioinformatics III**Emergence of the Metabolic Core**(a–c) Average relative size of the number of reactions that are always active as a function of the number of sampled conditions (black line). As the number of conditions increases, the curve converges to a constant marked by the dashed line, identifying the metabolic core of an organism. Red line : number of reactions that are always active if activity is randomly distributed in the metabolic network. The fact that it converges to zero indicates that the real core represents a collective network effect, forcing a group of reactions to be active in all conditions. Bioinformatics III**Emergence of the Metabolic Core**The number of metabolic reactions (d) and the number of metabolic core reactions (e) in the 3 studied organisms. As the complexity of the organism increases (the prokaryote H. pylori has fewest reactions, the prokaryote E. coli has more, and the eukaryote S. cerevisiae has most), the number of core reactions decreases. Complex organisms have more flexible metabolic networks. Fewer reactions are always on. Bioinformatics III**Metabolic Core of E.coli: The constantly active reactions**form a tightly connected cluster! Shown are all reactions that are found to be active in each of the 30,000 investigated external conditions. Blue: Metabolites that contribute directly to biomass formation, Red(green): core reactions (links) catalyzed by essential (or nonessential) enzymes. Black-colored links: enzymes with unknown deletion phenotype. Blue dashed lines: multiple appearances of a metabolite (to simplify the plot), links with arrows: unidirectional reactions. Bioinformatics III**Metabolic Core of E.coli: The constantly active reactions**form a tightly connected cluster! 20 out of the 51 metabolites necessary for biomass synthesis are not present in the core. This indicates that they are produced (or consumed) in a growth-condition-specific manner. Blue and brown shading: folate and peptidoglycan biosynthesis pathways White numbered arrows denote current antibiotic targets inhibited by: (1) sulfonamides, (2) trimethoprim, (3) cycloserine, and (4) fosfomycin. Bioinformatics III**Metabolic Core Reactions**The metabolic cores contain 2 types of reactions: (a) reactions that are essential for biomass production under all environment conditions (81 of 90 in E.coli) (b) reactions that assure optimal metabolic performance. Bioinformatics III**Characterizing the Metabolic Cores**(A) Number of overlapping metabolic reactions in the metabolic core of H. pylori, E. coli, and S. cerevisiae. The metabolic cores of simple organisms (H. pylori and E.coli) overlap to a large extent. The largest organism (S.cerevisae) has a much larger reaction network that allows more flexbility the relative size of the metabolic core is much lower. (B) The fraction of metabolic reactions catalyzed by essential enzymes in the cores (black) and outside the core in E. coli and S. cerevisiae. Reactions of the metabolic core are mostly essential ones. (C) One could assume that the core represents a subset of high-flux reactions. This is apparently not the case. The distributions of average metabolic fluxes for the core and the noncore reactions in E. coli are very similar. Bioinformatics III**Summary**• Adaptation to environmental conditionsoccurs via structuralplasticityand/orfluxplasticity. • Here: a surprisinglystablemetaboliccoreofreactions was identifiedthataretightlyconnectedtoeachother. • - thereactionsbelongingtothiscorerepresentpotential targetsforantimicrobialintervention. Bioinformatics III