Enumerating constrained elementary flux vectors of metabolic networks

The concept of elementary vector is generalised to the case where the steady-state space of the metabolic network is not a flux cone but is a general polyhedron due to further inhomogeneous constraints on the flows through some of the reactions. On one hand, this allows to selectively enumerate elementary modes which satisfy certain optimality criteria and this can yield a large computational gain compared with full enumeration. On the other hand, in contrast to the single optimum found by executing a linear program, this enables a comprehensive description of the set of alternate optima often encountered in flux balance analysis. The concepts are illustrated on a metabolic network model of human cardiac mitochondria.

Yield of serial sputum specimen examinations in the diagnosis of pulmonary tuberculosis: a systematic review

Current international tuberculosis (TB) guidelines recommend the microscopic examination of three sputum specimens for acid-fast bacilli in the evaluation of persons suspected of having pulmonary TB. We conducted a systematic review of studies that quantified the diagnostic yield of each of three sputum specimens. By searching multiple databases and sources, we identified a total of 37 eligible studies. The incremental yield in smear-positive results (in studies using all smear-positive cases as the denominator) and the increase in sensitivity (in studies that used all culture-positive cases as the denominator) of the third specimen were the main outcomes of interest. Although heterogeneity in study methods and results presented challenges for data synthesis, subgroup analyses suggest that the average incremental yield and/or the increase in sensitivity of examining a third specimen ranged between 2% and 5%. Reducing the recommended number of specimens examined from three to two (particularly to two specimens collected on the same day) could benefit TB control programs, and potentially increase case detection for several reasons. A number of operational research issues need to be addressed. Studies examining the most effective and efficient means to utilize current technologies for microscopic examination of sputum would be most useful if they followed an internationally coordinated and standardized approach, both to strengthen the country-specific evidence base and to permit comparison among studies.

Functional stoichiometric analysis of metabolic networks

MOTIVATION

An important tool in Systems Biology is the stoichiometric modeling of metabolic networks, where the stationary states of the network are described by a high-dimensional polyhedral cone, the so-called flux cone. Exhaustive descriptions of the metabolism can be obtained by computing the elementary vectors of this cone but, owing to a combinatorial explosion of the number of elementary vectors, this approach becomes computationally intractable for genome scale networks.

RESULT

Hence, we propose to instead focus on the conversion cone, a projection of the flux cone, which describes the interaction of the metabolism with its external chemical environment. We present a direct method for calculating the elementary vectors of this cone and, by studying the metabolism of Saccharomyces cerevisiae, we demonstrate that such an analysis is computationally feasible even for genome scale networks.

An improved algorithm for stoichiometric network analysis: theory and applications

MOTIVATION

Genome scale analysis of the metabolic network of a microorganism is a major challenge in bioinformatics. The combinatorial explosion, which occurs during the construction of elementary fluxes (non-redundant pathways) requires sophisticated and efficient algorithms to tackle the problem.

RESULTS

Mathematically, the calculation of elementary fluxes amounts to characterizing the space of solutions to a mixed system of linear equalities, given by the stoichiometry matrix, and linear inequalities, arising from the irreversibility of some or all of the reactions in the network. Previous approaches to this problem have iteratively solved for the equalities while satisfying the inequalities throughout the process. In an extension of previous work, here we consider the complementary approach and derive an algorithm which satisfies the inequalities one by one while staying in the space of solution of the equality constraints. Benchmarks on different subnetworks of the central carbon metabolism of Escherichia coli show that this new approach yields a significant reduction in the execution time of the calculation. This reduction arises since the odds that an intermediate elementary flux already fulfills an additional inequality are larger than when having to satisfy an additional equality constraint.

Learning curves for mutual information maximization

An unsupervised learning procedure based on maximizing the mutual information between the outputs of two networks receiving different but statistically dependent inputs is analyzed [S. Becker and G. Hinton, Nature (London) 355, 161 (1992)]. For a generic data model, I show that in the large sample limit the structure in the data is recognized by mutual information maximization. For a more restricted model, where the networks are similar to perceptrons, I calculate the learning curves for zero-temperature Gibbs learning. These show that convergence can be rather slow, and a way of regularizing the procedure is considered.

Universal learning curves of support vector machines

Using methods of statistical physics, we investigate the role of model complexity in learning with support vector machines (SVMs), which are an important alternative to neural networks. We show the advantages of using SVMs with kernels of infinite complexity on noisy target rules, which, in contrast to common theoretical beliefs, are found to achieve optimal generalization error although the training error does not converge to the generalization error. Moreover, we find a universal asymptotics of the learning curves which depend only on the target rule but not on the SVM kernel.

Efficiently learning multilayer perceptrons

A learning algorithm for multilayer perceptrons is presented which is based on finding the principal components of a correlation matrix computed from the example inputs and their target outputs. For large networks our procedure needs far fewer examples to achieve good generalization than traditional on-line algorithms.

Online learning with ensembles

Supervised online learning with an ensemble of students randomized by the choice of initial conditions is analyzed. For the case of the perceptron learning rule, asymptotically the same improvement in the generalization error of the ensemble compared to the performance of a single student is found as in Gibbs learning. For more optimized learning rules, however, using an ensemble yields no improvement. This is explained by showing that for any learning rule f a transform f exists, such that a single student using f has the same generalization behavior as an ensemble of f students.

A Large Committee Machine Learning Noisy Rules

Statistical mechanics is used to study generalization in a tree committee machine with K hidden units and continuous weights trained on examples generated by a teacher of the same structure but corrupted by noise. The corruption is due to additive gaussian noise applied in the input layer or the hidden layer of the teacher. In the large K limit the generalization error εg as function of α, the number of patterns per adjustable parameter, shows a qualitatively similar behavior for the two cases: It does not approach its optimal value and is nonmonotonic if training is done at zero temperature. This remains true even when replica symmetry breaking is taken into account. Training at a fixed positive temperature leads, within the replica symmetric theory, to an α-k decay of εg toward its optimal value. The value of k is calculated and found to depend on the model of noise. By scaling the temperature with α, the value of k can be increased to an optimal value kopt. However, at one step of replica symmetry breaking at a fixed positive temperature εg decays as α−kopt. So, although εg will approach its optimal value with increasing sample size for any fixed K, the convergence is only uniform in K when training at a positive temperature.

[Therapeutic approaches in diseases caused by mycobacteria]

Following the historical introduction (the review by Schlossberger, 1928) the evolution of individual chemotherapy phases is briefly discussed. Then after following points are discussed. 1) The usual controlled clinical trials for evaluation of new drugs must be replaced (perhaps by estimates of the "early bactericidal activity" and of the 2 months' sputum conversion rates). 2) Interesting and of importance are research activities connected with the reasons for properties of TB bacilli in the persistence state, and of drugs capable to combat such bacilli (metronidazole?). 3) Since the human immunity against TB normally operates well (from 100 infected approximately 10 will get a clinically manifest disease), it is necessary to investigate possible combinations of chemo- and immunotherapy, either of combinations of antituberculous and immunomodulating drugs (eg. TNF alpha-antagonists), or of chemotherapy combined with vaccination (eg. M. vaccae). 4) There are still unsolved problems in formulations of drugs (particularly R in fixed combinations) and of drug resorption (in HIV-pos. patients). 5) Other points of research interest could be the adsorption of TB bacilli on the surface of mammalian cells (1st step in pathogenesis), drugs other than antituberculotics (eg. amoxycillin + clavulanic acid), or even drugs used for completely other diseases (eg. antidepressive). 6) All these attempts should serve primarily the purpose of further reduction of the necessary therapy duration (M+ pulmonary TB) to 1 - 2 months in total.

[Is there a reliable new diagnostic procedure in mycobacterial infections including tuberculosis?]

Microscopy remains, for various reasons, even today the most important diagnostic tool for pulmonary TB. Because of the need to monitor drug resistance rates (pulmonary TB) and with other forms of TB or other mycobacterioses radiometric culture techniques, particularly if combined with conventional egg media cultures and with identification of grown bacilli using genetic probes are required on a broad scale; the speed of complete investigation is of utmost importance. Thus, even susceptibility testing should be performed with the radiometric technique. Serology, so far, is unfortunately still incapable of clear-cut answers. This applies also--for the time being--to polymerase chain reaction and other techniques in the field of molecular biology: the sensitivity, specificity and predictive values of such techniques are still not sufficiently high to warrant unequivocal results. This applies also to novel drug resistance tests e.g. luciferase techniques or genetic probes for resistance genes.

[The Mycobacteria Working Group: Investigation of resistance of tuberculosis bacteria in Germany 1991 and 1992]

A tuberculosis caused by resistant or even multidrug-resistant TB bacilli seems to be of increasing importance in some parts of the world. An exact investigation of rates of initial and acquired drug resistance of tubercle bacilli for the whole state of Germany is--considering the corresponding technical, logistic and economic necessities--rather demanding. The "Arbeitskreis Mykobakterien" performed therefore, as a first step, an estimation of global resistance rates of tubercle bacilli in Germany based on questionnaires sent to TB laboratories throughout the country: only the first strain isolated in a patient in 1991 and 1992 should have been considered. The results are presented in Table 1.

[Diagnosis of mycobacterioses in the laboratory of the future]

Microscopy is, for various reasons, even today the most important diagnostic measure for pulmonary Tb. With other forms of Tb and with other mycobacterioses there is a considerable increase in importance of the radiometric culture techniques, particularly if these are combined with serologic (ELISA and its variants) or molecular-genetic assays for the presence of mycobacteria. The latter, especially the so called genetic probes combined with the polymerase chain reaction, represent probably the assay of the future for mycobacteria; however, a stern warning must be sounded against the use of such techniques without previous experience in the field of the "classic" mycobacteriology. Serology is, inspite of recent enormous advances, still not capable of producing unequivocal results.

Isoniazid protects mice against endotoxin lethality without influencing tumor necrosis factor synthesis and release

Treatment of NMRI mice with isoniazid (INH; 25 mg/kg) intraperitoneally induced significant protection when it was injected before or after a lethal intravenous challenge with endotoxin. The INH preparation used was not contaminated with endotoxin. Tumor necrosis factor (TNF) was not elevated in sera from NMRI mice 2 h after the injection of INH. INH did not influence TNF synthesis or release determined in human monocytes in vitro. Therefore, it is concluded that the protective effect of INH against lethal endotoxin is not due to a suppressive effect of INH on TNF production.

Present position of microscopy and of culture in diagnostic mycobacteriology

The relative merits of microscopy and culture are discussed; it is attempted to review the possible reasons why microscopy is considered as unreliable in TB diagnosis, including the probably false application of Bayes' theorem. In the author's opinion, microscopy still holds a firm position in the rapid and relatively reliable detection of pulmonary TB. As for culture, it is highly questionable if "better media" could accelerate the growth rate of tubercle bacilli. Therefore, it should be attempted to introduce novel techniques which will allow to decrease considerably the mycobacterial mass needed for detection. Laboratories equipped with such devices could offer their services even on an intercontinental basis, in order to have a favourable cost-benefit ratio.

On the possibility of cross-resistance between rifampicin and clofazimine in mycobacteria--a laboratory study

One-way cross-resistance to rifampicin in clofazimine-resistant substrains of Mycobacterium smegmatis ATCC 607 was seen in 4 instances, whereas at most a borderline situation (probably within the experimental error) occurred in this respect with Mycobacterium parafortuitum 16009 and with strains of Mycobacterium "lufu" and Mycobacterium fortuitum. Clofazimine-resistant sub-strains were not isolated from the parental strains of Mycobacterium tuberculosis. Methodological influences (the gradient plate technique was used in this study) are discussed.

Antimycobacterial activity of isoniazid + prothionamide + dapsone against a number of randomly selected 'wild' strains of Mycobacterium tuberculosis

46 strains of Mycobacterium tuberculosis were studied as to their susceptibility to isoniazid (INH) or to INH + prothionamide (PTH) in the presence or absence of dapsone (DDS) by microdilution transfer plate technique. Synergy with DDS as defined by Berenbaum in 1978 was seen in 78% with INH and in 91% with INH + PTH. It concerned the minimal inhibitory concentration, whereas the minimal bactericidal concentration (MBC) was unaffected in the case of INH and it was increased in only 52% with INH + PTH. The latter MBC proportion mentioned would be decreased to roughly 16% accepting an only twofold decrease in the MBC as possibly due to dilution errors. The possible effect of dimethyl sulphoxide on the reported results is discussed.

The realistic possibilities of chemotherapy for pulmonary tuberculosis in South America

The slow integration of antituberculosis programmes into general health service activities is the most important problem encountered. Short course regimes with high cost drugs require the supervision which, for the time being, is difficult to realize on a large scale except, perhaps, in big cities. Though pilot studies of short course regimes are welcomed, for the immediate use in both the capitals and the rural areas mainly self-administered low-cost systems may be considered for the chemotherapy of tuberculosis, with basic division of patients into groups of previously untreated and previously treated persons (which would affect the drug combination used), and smear (and culture?) positive or negative persons (which might affect the duration of therapy administered). For the latter, it is of considerable importance to continue improving the network of bacteriological laboratories.

On the mechanism of the antimycobacterial activity of isoniazid + prothionamide + dapsone (Isoprodian)

Increased antimycobacterial activity of Isoprodian (isoniazid + prothionamide + dapsone) may be due to (i) decreased mutation rate for INH resistance provoked in mycobacteria by DDS (in 2 of 3 strains tested); (ii) leakage of K+, Na+ or Ca++ caused by INH and/or by PTH (in all 3 strains tested), and (iii) indicating damages which may increase the low level penetration of DDS into the cell in sub-MIC concentrations as shown by phage proliferation changes (tested in 1 strain with 1 phage).