Research Projects

This section introduces the current projects of the Molecular Systems Physiology group.


Reconstruction of biochemical reaction networks

Our manually assembled, highly curated reconstructions are build in a bottom-up approach using genome annotations, literature, biochemical data, and omics data. We have developed infrastructure to enable high level of quality control and assurance (QC/QA) for the reconstruction process, the network validation and computational analysis.

Our multiscale biochemical reconstruction repertoire includes, beyond others:

Metabolism: genome-scale reconstructions for

  • Global human metabolism, which is the most comprehensive, manual curated reconstruction currently available, based on (Duarte et al.), and constructed in a community-driven effort ( Thiele et al., NBT, 2013). We also assigned reaction directionality to the human metabolic reconstruction based on thermodynamic data (Haraldsdottir et al.). Moreover, we extended the fatty acid oxidation pathway significantly to enable mapping of standardly measured metabolites (e.g., acylcarnitines) onto the human metabolic reconstruction (Sahoo et al.)
  • Human small intestinal epithelial cells, which we used to study their role in diet and enzymopathies (Sahoo and Thiele).
  • Human cardiomyocyte mitochondrion, which we used to study the impact of diabetes and diets on mitochondrial metabolic capabilities (Thiele et al.)
  • Global mouse metabolism, which was generated by mapping mouse homologous genes onto the human metabolic reconstruction (Sigurdsson et al.) and which has been recently extended and updated (Heinken et al.)
  • the cyanobacterium Synechocystis sp.PCC6803, which we employed to study the robustness of its metabolic network and its photosynthetic apparatus (Nogalges et al.)
  • the hyperthermophilic, hydrogen producing bacterium Thermotoga maritima, which includes protein structure information for all enzymes (Zhang et al., Nogales et al.)
  • the human pathogen Salmonella typhimurium LT2, which was assembled in a community-driven effort (Thiele et al.)
  • the human pathogen Helicobacter pylori(Thiele et al.)
  • the human beneficial gut microbe Bacteroides thetaiotaomicron(Heinken et al.)
  • the human beneficial gut microbe Faecalibacterium prausnitzii(Heinken et al.)


  • a comprehensive, large-scale, stoichiometric network for Toll-like receptor signaling pathwaysin mammalian cells (Li et al., Aurich and Thiele),

Macromolecular synthesis:

  • a genome-scale, gene-and sequence specific, highly detailed stoichiometric network for the macromolecular synthesis machinery in E. coli(Thiele et al, Thiele et al). This network is multiscale as it combines cellular processes occurring on multiple time scales.

Truly multiscale:

  • an integrated host-microbe metabolic model of mouse and the beneficial human gut inhabitant Bacteroides thetaiotaomicron, which we employ to study mutual and competitive metabolic interactions and their impact onto human health. (Heinken et al).

Further reading

Creation and analysis of predictive multiscale models of human metabolism

We aim to continuously refine and expand the human metabolic reconstruction by incorporating missing information and reactions and making them available to the research community via the website, The refinement and expansion is a direct product of identifying missing metabolic routes in the human metabolic reconstruction of metabolites discovered in metabolomics studies, in published work and by collaborators. We are developing computational methods to analyze various omics data with the human metabolic models, with particular emphasis on metabolomic data. These methods result in personalized computational models, consistent with the measured omic data from individuals. These models allow us to predict the metabolic status of an individual and to compare it with other individuals using state-of-the-art machine learning approaches. With personalized metabolic models at hand, one can now envision proposing dietary supplementations or drug administrations that would move a disease-perturbed metabolic model into a more healthy state. We develop methods that permit us to propose such personalized nutrition-based therapeutic approaches. We are working closely with clinicians to ensure that this modeling approach could be translated into clinical practice and that it is of biomedical relevance. A current challenge in modeling with the human metabolic network is its global nature, as it accounts for metabolic transformations occurring in at least one human cell, but not necessarily in all. Consequently, the network exhibits a high gene and reaction redundancy (i.e., multiple alternate metabolic routes are present that would not be present as such in a single cell), leading to a reduced predictive sensitivity of gene knockouts. To overcome this limitation, we are working on organ- and cell-type specific metabolic reconstructions.  

Further reading

Creation of metabolic models for human gut microbes, modeling of microbial communities and their interactions with the human host

The creation of high-quality metabolic reconstructions relies heavily on the availability of high quality genome annotations as well as biochemical and phenotypic data. As many gut microbes are either only poorly studied or not studied at all, the use of automated reconstruction tools becomes increasingly important. Our efforts have therefore focused on accessing how well those automated methods perform by developing metabolic reconstructions for medium well studied gut microbes using this tool and performing subsequent manual curation. Based on this experience, we are now expanding genome annotations to generate semi-automated metabolic reconstructions for numerous important gut microbes. We are developing computational approaches to integratively investigate the metabolic crosstalk between these microbes and their role within the gut microbial community. We are therefore using the constraint-based modeling approach. We develop approaches to integrate meta-genomic and meta-transcriptomic data, obtained from publications and collaborators, with microbial community models to enable further insight into these complex data and the role of the human gut microbiota in human health and disease. The integration of genomic and transcriptomic data has been already published for single organisms but has so far not been applied for microbial communities consisting of multiple species. With such microbial or community metabolic reconstructions at hand, we can now start to address the question of their metabolic interactions with the human host. We have recently developed a scalable constraint-based modeling approach that permits study of the metabolic interactions between a host and one or more microbes. We are now expanding on this work to include more microbes and a more detailed representation of the host metabolism, thereby enabling investigation of how the microbiota contributes to homeostasis and dysbiosis. The gut microbiota also plays an essential role in studying the effect of different diets on human metabolism. Which microbes can breakdown which dietary nutrients? Which microbes produce which important metabolites (e.g., short chain fatty acids) for the host? Can we use such insight to assist in designing (personalized) functional food or propose probiotics? We use our models to address those and related questions.  

Further reading

Development of efficient computational methods for large- and multi-scale modeling

As the complexity of the metabolic models described above increases, more efficient modeling approaches are required. Traditionally, the COBRA field solved many important problems, including generation of tissue-specific metabolic models and in silico formulation of minimal growth medium, using mixed-integer linear programming, which scales poorly with increasing model size and complexity. We, and other, have already developed linear approximation algorithms, which have better performance on large scale. We will continue to develop those algorithms and make them freely available to the research community.

Our repertoire includes:

  • rBioNet - an open source tool for multiscale, biochemical network reconstruction (Thorleifsson and Thiele)
  • von Bertalanffy 1.0 - an open source tool for assigning unbiased reaction directionality to network reactions based on thermodynamic principles (Fleming and Thiele)
  • COBRA toolbox v2 for analysis metabolic networks (Schellenberger et al), and
  • fastFVA - an open source tool for assessing network redundancy in a computationally efficient & tractable manner both large-scale and multiscale biochemical reaction networks (Gudmunsson and Thiele)
  • Robust flux balance analysis of multiscale biochemical reaction networks (Sun et al)
  • Prediction of missing reactions in multi-compartment biochemical reaction networks: fastGapFill (Thiele et al)

Further reading