2019 SysMod annual meeting
Integrating systems biology and bioinformatics
July, 2019 | Basel, CH
Dynamical modeling
Flux balance analysis
Logical modeling
Network modeling
Stochastic simulation
Developmental biology
Precision medicine

ETH Zürich

Jörg Stelling

Prof. Dr. Douglas Lauffenburger

Massachusetts Institute
of Technology

Douglas Lauffenburger

Edda Klipp

zu Berlin

Keynote speakers


Advances in genomics are creating new opportunities to understand biology that require both systems modeling and bioinformatics. The fourth annual SysMod meeting will be a forum for discussion about combined use of systems biology modeling and bioinformatics to understand biology and disease. The meeting will take place July 22, 2019 during the 2019 ISMB/ECCB conference external-link in Basel. The meeting will feature several keynote talks and contributed presentations.


Dynamical modeling
Flux balance analysis
Logical modeling
Network modeling
Stochastic simulation


Developmental biology
Precision medicine

Keynote speakers

Prof. Dr. Dr. h.c. Edda Klipp

Edda Klipp external-link
Humboldt-Universität zu Berlin, Germany

Prof. Dr. Douglas Lauffenburger

Douglas Lauffenburger external-link
Massachusetts Institute of Technology, USA

Prof. Dr. Jörg Stelling

Jörg Stelling external-link
ETH Zürich, Switzerland


8:30-9:30 ISMB keynote talk
9:30-10:00 Coffee break with exhibitors
10:15-12:40 Session I: Systems biology of human cells and diseases
Moderator: Julio Saez-Rodriguez, Systems Biomedicine, BioQuant, Heidelberg University, DE
10:15-10:20 Introduction to SysMod 2019
Andreas Dräger
Insititute for Biomedial Informatics (IBMI), University of Tübingen, DE
The community of special interest (COSI) in systems modeling (SysMod) organizes annual one-day gatherings. In 2019 the meeting comprises three sessions that cover a broad variety of topics, beginning with human cells and disease modeling, followed by the afternoon session on the systems biology of microorganisms and concludes with current trends in the field. Each of the three sessions features one of the keynote speakers, Douglas Lauffenburger, Edda Klipp, and Jörg Stelling. The event is hosted by Claudine Chaouiya, María Rodriguez Martinez, and Andreas Dräger on behalf of the ten COSI organizers. This brief talk introduces all speakers, organizers, and main topics of the 2019 meeting.
10:20-11:00 Cross-Species Translation of Biological Information via Statistical and Machine Learning Computational Frameworks
Douglas Lauffenburger
Massachusetts Institute of Technology, US
A vital challenge that the vast majority of biological research must address is how to translate observations from one physiological context to another—most commonly from experimental animals (e.g., rodents, primates) or technological constructs (e.g., organ-on-chip platforms) to human subjects. This is typically required for understanding human biology because of the strong constraints on measurements and perturbations in human in vivo contexts. Direct translation of observations from experimental animals to human subjects is generally unsatisfactory because of significant differences among organisms at all levels of molecular properties from genome to transcriptome to proteome and so forth. Accordingly, addressing inter-species translation requires an integrated experimental/computational approach for mapping comparable but not identical molecule-to-phenotype relationships. This presentation will describe methods we have developed for a variety of cross-species translation examples, centrally based on systems modeling frameworks.
11:00-11:20 Personalization of logical models using multi-omics data and its use in the study of clinical stratification and drug response
Jonas BÉAL, Arnau Montagud, Emmanuel Barillot, and Laurence Calzone
Institut Curie, FR
Mathematical models of cancer pathways are built by mining the literature for relevant experimental observations or extracting information from pathway databases. As a consequence, these models generally do not capture the heterogeneity of tumors and their therapeutic responses. We present here a novel framework, PROFILE, to tailor logical models to particular biological samples such as patient tumors, compare the model simulations to individual clinical data and investigate therapeutic strategies.Our approach makes use of MaBoSS, a tool based on Monte-Carlo kinetic algorithm to perform stochastic simulations on logical models resulting in model state probabilities. This semi-quantitative framework allows to integrate mutation data, copy number alterations (CNA), and transcriptomics/proteomics into logical models. These personalized models are validated by comparing simulation outputs with patients’ clinical data (subtypes, survival) and then used for cell line-specific investigations regarding the effects of drug perturbations, allowing both verification of the theoretical behavior of the model and comparison with experimental drug sensitivities.Our approach aims to combine the mechanistic insights of logical modeling with multi-omics data integration to provide patient or cell line-relevant models that can serve as tools for analyzing therapeutic responses.
11:20-11:40 Making Sense of Large Kinetic Models
Fabian Fröhlich, Luca Gerosa, and Peter Sorger
Harvard University, US
In recent years, simulation and training of large kinetic multi-pathway models with hundreds to thousands of species and parameters has become increasingly tractable. These large kinetics models are often constructed with the aim to deepen our mechanistic understanding of signaling pathways. Yet, the high complexity of large kinetic models impedes our ability to understand signal transduction in the model and thus limits possibilities for mechanistic insight and hypothesis generation.Here we propose the use of model structure to provide high- and low-level descriptions of signaling dynamics. We exploit the abstraction of rule-based models to provide protein-level summaries of signaling dynamics. To study emergent properties of the model, we apply a combination of causal compression and hierarchical modularization to provide pathway-level summaries of signal transduction.We apply these methods to an ordinary differential equation model of adaptive resistance in melanoma (EGFR and ERK pathways, >1k state variables, >10k reactions). We trained the model on absolute proteomic and phospho-proteomic as well as time-resolved immunofluorescence data, both in dose-response to small molecule inhibitors. We illustrate how low- and high-level descriptions can be used to probe signaling dynamics in the trained model and provide simple explanations for the observe nonlinear dose-response data.
11:40-12:00 Constraint-based modeling of human single cells to investigate metabolic heterogeneity in cancer subpopulations
Davide Maspero, Marzia Di Filippo, Riccardo Colombo, Dario Pescini, Alex Graudenzi, Hans V. Westerhoff, Lilia Alberghina, Marco Vanoni, Giancarlo Mauri, and Chiara Damiani
Biotechnology and Biosciences, University Milano-Bicocca, IT
Intratumour heterogeneity characterizing cancer populations represent a key factor in fostering the disease progression. In particular, metabolic intratumour heterogeneity increases the repertoire of possible cellular responses to drugs and boosts the adaptive nature of cellular behaviors, hindering the identification of effective treatments. Unfortunately, current metabolomics technologies depict the average cell population behavior, but disregard both internal interactions and differences. To explore such metabolic heterogeneity, characterization of metabolic programs at the single-cell level must be used. In this regard, single-cell metabolomics is still at its infancy thus is less advanced than single-cell sequencing. To bridge this gap, we present a computational framework to characterize metabolism at the single cell level and possible metabolic interactions among cells, by integrating bulk metabolomics and single-cell transcriptomics data. Than, we exploit constraint-based modeling to simulate a set of replicates of a human metabolic network corresponding to interacting distinct cells of a given population.
The integration of transcriptomics profiles of individual tumour cells isolated from lung adenocarcinoma and breast cancer patients allowed to compute single-cell fluxomes, to identify clusters of cells with different growth rates, and to point out the possible metabolic interactions among cells via exchange of metabolites by showing adherence to experimental evidences.
12:00-12:20 Modeling the propagation of the innate immune response to control influenza virus infection
Gregory Smith, Aleya Dhanji, Irene Ramos, Ciriyam Jayaprakash, and Stuart Sealfon
Icahn School of Medicine at Mount Sinai Hospital, US
Influenza remains a major threat to global health resulting in millions of severe infections and hundreds of thousands of deaths each year. The rapidly mutating nature and diverse strains of the influenza virus limit vaccine efficacy and highlight the necessity of novel thinking to produce more effective treatment options. Understanding the early innate immune response to infection is an essential component to this process. Despite considerable study into the dynamics of influenza infection, much is still unknown about the interplay between viral antagonism and the propagation of innate immune response across a cell population. Computational modeling provides an ability to measure this interplay with a real-time resolution that would be infeasible experimentally. We have devised a spatial, stochastic agent-based model of influenza virus infection of lung epithelium that tracks the spread of a viral infection and corresponding host cytokine response across a layer of epithelial cells. In order to fit our model, we apply in vitro infection time course data and single cell RNA sequencing data, including novel findings of paracrine signaling-induced IFNλ production. Our findings suggest this feed forward paracrine signaling loop can have a significant impact on the effectiveness of host immune response.
12:20-12:40 Modeling recovery of Crohn’s disease, by simulating microbial community dynamics under perturbations
Jorge Carrasco Muriel, Beatriz García-Jiménez, and Mark D. Wilkinson
Center for Plant Biotechnology and Genomics UPM – INIA, Universidad Politecnica de Madrid, Madrid, ES
There are few large longitudinal microbiome studies, and fewer that include planned, annotated perturbations between sampling-points. Thus, there are few opportunities to employ data-driven computational analyses of perturbed microbial communities over time.Our novel computational system simulates the dynamics of microbial communities under perturbations, using genome-scale metabolic models (GEM). Perturbations include modifications to a) the nutrients available in the medium, allowing modelling of prebiotics; or b) the microorganisms present in the community, to model probiotics or pathogen infection. These simulations generate the quantity and types of information used as input to the MDPbiome system, which builds predictive models suggesting the perturbation(s) required to engineer microbial communities to a desired state.We demonstrate that this novel combination, called MDPbiomeGEM, is able to model the influence of prebiotic fiber and probiotic in the case of a Crohn’s disease microbiome. The output’s recommended perturbation to recover from dysbiosis is to consume inulin, which promotes butyrate production to reach homeostasis, consistent with prior biomedical knowledge. Our system could also contribute to design (perturbed) microbial community dynamics experiments, potentially saving resources both in natural microbiome scenarios by optimizing sequencing sampling, or to optimize in-vitro culture formulations for generating performant synthetic microbial communities.
12:40-2:00 Lunch and community discussion
12:45-1:45 ISCB Town Hall Meeting
2:00-4:00 Session II: Systems biology of microorganisms
Moderator: Claudine Chaouiya, Polytech Marseille, Aix-Marseille University (AMU), PT
2:00-2:40 Systematic integration of models and data for yeast growth, division and stress response
Edda Klipp
Humboldt-Universität zu Berlin, DE
2:40-3:00 Stochastic system identification without an a priori chosen kinetic model — exploring feasible cell regulation with piecewise linear functions
Martin Hoffmann and Jörg Galle
Fraunhofer ITEM, Division of Personalized Tumor Therapy, Regensburg, DE
Background: Kinetic models are at the heart of system identification. A priori chosen rate functions may, however, be unfitting or too restrictive for complex or previously unanticipated regulation.
Methods: We applied general purpose piecewise linear functions for stochastic system identification in one dimension using published flow cytometry data on E.coli and report on identification results for equilibrium state and dynamic time series.
Results: In metabolic labelling experiments during yeast osmotic stress response, we find mRNA production and degradation to be strongly co-regulated. In addition, mRNA degradation appears overall uncorrelated with mRNA level. Comparison of different system identification approaches using semi-empirical synthetic data revealed the superiority of single-cell tracking for parameter identification. Generally, we find that even within restrictive error bounds for deviation from experimental data, the number of viable regulation types may be large. Indeed, distinct regulation can lead to similar expression behaviour over time.
Conclusion: Our results demonstrate that molecule production and degradation rates may often differ from classical constant, linear or Michaelis–Menten type kinetics.(1) NPJ Syst Biol Appl. 2018 Apr 11; 4:15. doi: 10.1038/s41540-018-0049-0, PMID 29675268
3:00-3:20 Pleiades Toolkit: Automatic rule-based modeling of bacterial gene regulation enables simulation, prediction, and perturbation of gene responses
Rodrigo Santibáñez, Daniel Garrido and Alberto Jm Martin
Network Biology Lab, Centro de Genómica y Bioinformática, Facultad de Ciencias, Universidad Mayor, CL
Regulation of gene expression is essential for cell homeostasis and adaptation. This regulation relies on transcription factors and other proteins that trigger specific genetic programs. However, the complexity of this regulatory network precludes efforts to model gene regulation at genome-scale. In this work, we developed the Pleiades toolkit that is currently composed by Atlas, Pleione, and Sterope. Atlas reconstruct a Rule-Based Model (RBMs) from biological networks. These Rules are similar to chemical equations and Atlas interpret nodes as model components and edges as a set of reactions, depending on the encoded nature of the networks. After model reconstruction, Pleione parameterizes RBMs employing one of four stochastic simulation software and distribute calculations with subprocesses or SLURM, taking advantage of high-performance computers and computational clusters. Finally, Sterope performs a global sensitivity analysis of selected parameters, calculating the interference or contribution of one Rule to itself and the remaining Rules. We validate the Pleiades employing the Escherichia coli regulatory and metabolic networks retrieved from Ecocyc and expression data from the literature. The developed Toolkit allows assessing of the impact of modifications like gene copy number, operon architecture, and other common genetic modifications to understand bacterial physiology, disease, and eventually, engineering of those systems.
3:20-3:40 Condition-specific series of metabolic sub-networks and its application for gene set enrichment analysis
Van Du T. Tran, and Marco Pagni
Vital-IT group, SIB Swiss Institute of Bioinformatics, CH
Genome-scale metabolic networks and transcriptomic data represent complementary sources of knowledge about an organism’s metabolism, yet their integration to achieve biological insight remains challenging. We investigate here condition-specific series of metabolic sub-networks constructed by successively removing genes from a comprehensive network. The optimal order of gene removal is deduced from transcriptomic data and the produced sub-networks are evaluated via a fitness function, which estimates their degree of alteration. We then consider how a gene set, i.e. a group of genes contributing to a common biological function, is depleted in different series of sub-networks to reveal the difference between experimental conditions. The method, named metaboGSE, was validated on public data for Yarrowia lipolytica. It was shown to produce GO terms of higher specificity compared to popular gene set enrichment methods like GSEA or topGO. Furthermore, metaboGSE permits identifying genes that are not necessarily differentially expressed, but nevertheless responsible for functional differences between conditions. We are currently investigating this aspect as part of a study about the early modifications leading to metaflammation in white adipose tissue of mice under high-fat diet. The metaboGSE R package is available at https://CRAN.R-project.org/package=metaboGSE.
3:40-4:00 µbialSim: Simulating complex microbial communities at their natural diversity
Florian Centler and Denny Popp
Helmholtz Centre for Environmental Research – UFZ, DE
Microbial communities are ubiquitous in nature and impact human well-being in many ways. They close global elemental cycles, are harnessed in biotechnological applications such as biogas production, and play an important role in human health. To uncover the complex web of metabolic interactions in these systems, we introduce µbialSim (pronounced ‘microbialSim’), a novel numerical simulator that implements the dynamic Flux-Balance-Analysis approach. By employing a novel numerical integration scheme, our simulator can consider communities at their natural diversity, going beyond current simulator codes which are restricted to few species only. As an example, we apply µbialSim to the entirety of a model collection of 773 species of the human gut microbiome. We demonstrate how the predicted pattern of compound exchange and its dynamics can be analyzed as the community feeds on a western-diet substrate pulse. While quantitative predictions have to be interpreted in the light of the simulator’s current limitations – being restricted to metabolic interactions only – we envision µbialSim as a starting point for an extensive in silico characterization of community dynamics at an unprecedented level of detail and helping in elucidating general principles in microbial ecology, and as a tool for experimental design and the design of communities.
4:00-4:40 Coffee break with exhibitors
4:40-6:10 Session III: Current trends in systems biology
Moderator: María Rodriguez Martinez, IBM Zurich Research Laboratory, CH
4:40-5:00 Learning dynamical information from static protein and sequencing data
Philip Pearce, Francis Woodhouse, Aden Forrow, Halim Kusumaatmaja, and Jorn Dunkel
Massachusetts Institute of Technology, US
Many complex processes, from protein folding and virus evolution to brain activity and neuronal network dynamics, can be described as stochastic exploration of a high-dimensional energy landscape, but little is understood about the reliable inference of dynamics from static data in such settings. Here, we introduce a flexible and robust numerical framework to infer Markovian transition networks directly from time-independent data sampled from stationary equilibrium distributions. Our approach combines Gaussian mixture approximations and self-consistent dimensionality reduction with minimal-energy path estimation and multi-dimensional transition-state theory. We demonstrate the practical potential of the inference scheme by reconstructing the network dynamics for several protein folding transitions, gene regulatory network motifs and HIV evolution pathways. The predicted network topologies and relative transition time scales agree well with direct estimates from time-dependent data in each case. The underlying numerical protocol thus allows the recovery of relevant dynamical information from instantaneous ensemble measurements, effectively alleviating the need for time-dependent data in many situations. Owing to its generic structure, the framework introduced here will be applicable to modern experimental technologies including cryo-electron-microscopy and high-throughput single-cell RNA sequencing data.
5:00-5:05 Towards Homogeneous Modeling and Simulation of Whole-Cells
Paulo Eduardo Pinto Burke and Luciano Costa
University of São Paulo, BR
Computational models of biological systems are growing in complexity, approaching the whole-cell scale. Both modeling and simulation of such systems are far from trivial, yet, significant advances in this direction have already been performed. Current whole-cell models yield heterogeneous representations of cellular processes, each one being approached using established methods. Their integration is achieved by exchanging information between them from time to time. Although this approach proved to be useful, its organism-specificity makes it hard to scale and adapt to other organisms. Here, we present a homogeneous approach to model and simulate whole-cells where all cellular process are represented through their underlying biochemical reactions. Such a representation results in a map of all possible biochemical interactions between molecular entities of a cell in the form of a single biochemical network, naturally integrating cellular processes. We discuss the implications of such an approach on automated model generation, user-friendliness, parameter estimation, scalable simulation methods, and computational costs. We also present an example of the entire pipeline extending from model construction up to simulation and analysis using toy models. In addition, we present a biochemical network model of a whole real organism.
5:05-5:10 The regulation of aquaporin 2 vesicle transport by localized cyclic AMP pools
Christoph Leberecht and Dirk Labudde
Hochschule Mittweida, DE
The average basal cAMP concentration in eukaryotic cells is 1 micro mole per liter. The reported cAMP concentration to half-maximally activate protein kinase A (PKA) in vitro is about 200 nano mole per liter. This relationship suggests that PKA should be constantly active. However, in vivo studies determined the sensitivity of PKA to be significantly lower. A promising hypothesis for the apparent low sensitivity is, that cAMP abundance is highly regulated by concentration gradients. As a model system we choose collecting duct principal cells that require PKA signaling for the transport of vesicles storing the water channel aquaporin 2.We modeled the interplay of localized cAMP, PKA, and phosphodiesterase and their effect on vesicle transport in a spatial model. To model the movement and behavior of vesicles as well as reaction kinetics and diffusion a hybrid simulation technique was devised. We have found that cAMP concentration forms localized sinks around vesicles that act as a threshold to prevent unjustified transport initiation. Further, cAMP concentration is further decreased along the path of traveling vesicles. The paths might temporarily prevent other vesicles from following the initial vesicles and therefore regulate the throughput of vesicles to the membrane.
5:10-5:15 Optimal information acquisition of the molecular systems in living organisms require a non-minimal level of noise
Eugenio Azpeitia and Andreas Wagner
University of Zürich, CH
Organisms are constantly acquiring information from the environmental. Information improves the decisions made by organisms, directly affecting their survival and reproductive success. Signaling pathways are the basic mechanisms used by cells to obtain information. They rely on reversible reactions for the binding of signals to receptors, the activation of molecules via allosteric regulation, and the binding of transcription factors to the DNA. However, reversible reactions are noisy, because of random fluctuations in the concentration and activity of molecules. Noise causes uncertainty about the information conveyed by transforming an input into a distribution of possible outputs. For this reason, it is commonly stated that noise reduces the capacity to acquire information. Interestingly, our results show that, under realistic biological conditions, reversible chemical reactions unavoidably produce non-minimal levels of noise for information acquisition. We study how this phenomenon affects the capacity of signaling pathways to acquire and transmit information. We show that the non-minimal levels of noise are transmitted from reversible reactions to the production of mRNA and protein. However, the strength of the binding of a reversible reaction modulates information acquisition and noise levels. Finally, we test our results using the nuclear receptor signaling pathway as an example.
5:15-5:20 An in silico mechanistic representation of an in vitro neutropenia assay to explore dose and schedules
Cristina Santini, Carla Guarinos, Alicia Benitez, Estela Torano, Mark McConnell, Matthew Trotter, James Carmichael, Soraya Carrancio, and Alex Ratushny
Celgene, ES
Objectives: Lenalidomide, an immunomodulatory agent, is approved for the treatment of multiple myeloma, del 5q myelodysplastic syndrome and mantle cell lymphoma. Lenalidomide causes a reversible block in neutrophil maturation. To investigate the dose and schedule that allows for neutrophil recovery, we developed an in silico model based on an in vitro assay. This model is applied to explore dosing regimens.
Methods: A compartmental model was developed to represent the in vitro maturation assay [1]. Donor related parameters were fitted to DMSO treatment data and compound related parameters were fitted to the effect upon treatment with a concentration range of lenalidomide.
Results: The proposed model quantitatively represents the in vitro neutropenia maturation system and the block in neutrophil maturation caused by lenalidomide. In silico predictions for neutrophil recovery after off-drug period were validated experimentally (predicted vs experimental data R2 = 0.985).
Conclusions: An in silico model that represents an in vitro neutropenia assay was developed. Good parameter fit and validated predictions support the applicability of the model to explore dose and schedule of lenalidomide in silico and propose regimens that could minimize a key clinical toxicity of this compound.
References:[1] Chiu et al., Br J Haematol 2019 Feb 14
5:20-6:00 Systems Analysis of Cell-to-Cell Variability
Jörg Stelling
ETH Zürich, CH
A key step for understanding heterogeneity in cell populations is to disentangle sources of cell-to-cell and intra-cellular variability. Single-cell time-lapse data provides potential means for this, but single-cell analysis with dynamic models is a challenging open problem. Most of the existing inference methods address only single-gene expression or neglect correlations between processes that underlie heterogeneous cell behaviors. The focus of the talk will be a simple, flexible, and scalable method for estimating cell-specific and population-average parameters to characterize sources and effects of cell-to-cell variability. The framework relies on non-linear mixed effects models of cellular networks. Its accuracy and performance compared to state-of-the-art methods from pharmacokinetics is demonstrated with a published model and data set. An application to endocytosis in yeast demonstrates that one can develop dynamic models of realistic size for the analysis of single-cell data. Combined with sensitivity analysis for identifying which biological sub-processes quantitatively and dynamically contribute to cell-to-cell variability, this application shows that shifting the focus from single reactions or parameters to nuanced and time-dependent contributions of sub-processes helps biological interpretation.
6:00-6:10 Closing remarks and poster awards SysMod 2019
Claudine Chaouiya
Polytech Marseille, Aix-Marseille University (AMU), PT
This concluding talk aims to briefly discuss the diversity of topics presented at the “Computational Modeling of Biological Systems“ (SysMod) COSI track. This diversity illustrates the importance of the field and the broad range of applications in systems biology and disease. Then, forthcoming meetings of interest will be announced, and the three poster awards will be delivered as a closing event.
6:10-7:00 Poster presentations
Integrated Flux Analysis of Susceptible and Resistant Escherichia coli under Antibiotic Stress
Sean Mack, Eric Hill, Young-Mo Kim, Lye-Meng Markillie, Teressa Palazzo, Robert Young, Karl Weitz, Ganesh Sriram, and Daniel Dwyer
The surge in antimicrobial resistance requires urgent development of innovative approaches to address the numerous bacterial pathogen threats outlined by the CDC and WHO. Notably, a growing body of evidence suggests that the presumed fitness disadvantages of resistant pathogens conferred by expression of resistance genes is not fully accurate. Arising from these data is the increasingly attractive hypothesis that modification of metabolism is a key component of antibiotic resistance. Further exploration of the relationship between metabolism, antibiotic stress, and resistance is clearly needed.
To address these gaps in our fundamental understanding, we have compared the metabolic behaviors of wildtype and resistant strains Escherichia coli through a combined transcriptomic and fluxomic analysis. Differential expression analysis identified significant shifts in activity in a multitude of pathways between the WT and resistant strains as well as the resistant strains with and without antibiotic stress. Additionally, the resistant strains produced significantly more CO2 than the wildtype strain. Our preliminary findings suggest that the resistant strains reductively constrain their metabolism upon genomic and/or antibacterial stress. To elucidate the specific metabolic alterations, we are generating comprehensive, genome-scale flux predictions through the integration of transcriptomics data with metabolic flux analysis simulations.

Inverse engineering metabolomics data to infer regulations of biological network
Jiahang Li, Wolfram Weckwerth, and Xiaoliang Sun

Background: One central goal of systems biology is to infer biochemical regulation from large-scale Omics data. We previously developed an experimentally validated inverse Jacobian approach that solves the Lyapunov equation JC + CJ^T = -2D to obtain the biochemical Jacobian directly from metabolomics data [1-3]. However, these algorithms rely on prior biological knowledge, e.g., genome annotation, to reconstruct metabolic network and obtain the stoichiometric matrix. In addition, they are difficult to capture feedback loops such as allosteric regulations.
Methods: Here we present a novel inverse Jacobian algorithm. We samples m times of the fluctuation matrix Di (i=1, 2 …m). Then using the corresponding covariance matrix Ci, we are able to solve the over-determined Jacobian matrix by machine learning-based optimization methods. This approach is illustrated by in silico stochastic simulation with different-sized metabolic networks from the BioModel database. The advantages of the new approach are that 1) it does not rely on prior biological knowledge; 2) it can be applied on networks with feedback loops; 3) it requires fewer replicates.
1. Steuer R et al. (2003) Bioinformatics 19(8):1019-26
2. Sun X & Weckwerth W (2012) Metabolomics 8(1):81-93
3. Sun X et al. (2015) Frontiers in Bioengineering and Biotechnology 3:188

Metabolic models to screen gut microbial metabolites augmenting mitochondrial function
Prashant Bajpai, Shakti Sagar, and Anurag Agrawal

Gut microbes have been shown to play an important role in human health and disease, affecting host physiology and homeostasis by secretion of small molecules.A few such circulating blood metabolites derived from gut microbes have been shown to influence mitochondrial function.Despite sharing the common evolutionary origin, there are no such studies so far which could identify metabolites affecting mitochondrial health on a large scale. Here,we have utilized constraint based model of mitochondria to screen such gut microbial metabolites.Out of 437 metabolites taking part in mitochondrial pathways,325 were common between metabolites produced by gut microbes and mitochondrial metabolites. The effect of these metabolites on mitochondrial function was tested using the metabolic model of mitochondria.We simulated hypoxic condition,a proxy to mitochondrial dysfunction by restricting the oxygen uptake from 19.8μM-5.0μM which resulted in decrease in ATP production from 102.7μM-30μM.In this condition, we simulated uptake of various gut microbial metabolites to identify which metabolites restore ATP levels to normal.Of the 127 metabolites tested so far,21 metabolites showed positive results and are being validated in-vitro. This is the first study of its kind that uses metabolic model to screen large number of metabolites that are poised to be a part of future mitochondrial targeted therapy

Bayesian Metabolic Flux Analysis reveals intracellular flux couplings
Markus Heinonen, Maria Osmala, Henrik Mannerstrom, Janne Wallenius, Samuel Kaski, Juho Rousu, and Harri Lähdesmäki

Motivation: Metabolic flux balance analysis is a standard tool in analysing metabolic reaction rates compatible with measurements, steady-state and the metabolic reaction network stoichiometry. Flux analysis methods commonly place unrealistic assumptions on fluxes due to the convenience of formulating the problem as a linear programming model, and most methods ignore the notable uncertainty in flux estimates.
Results: We introduce a novel paradigm of Bayesian metabolic flux analysis that models the reactions of the whole genome-scale cellular system in probabilistic terms, and can infer the full flux vector distribution of genome-scale metabolic systems based on exchange and intracellular (e.g. 13C) flux measurements, steady-state assumptions, and objective function assumptions. The Bayesian model couples all fluxes jointly together in a simple truncated multivariate posterior distribution, which reveals informative flux couplings. Our model is a plug-in replacement to conventional metabolic balance methods, such as flux balance analysis (FBA). Our experiments indicate that we can characterise the genome-scale flux covariances, reveal flux couplings, and determine more intracellular unobserved fluxes in C. acetobutylicum from 13C data than flux variability analysis.
Availability: The COBRA compatible software is available at github.com/markusheinonen/bamfa
Contact: markus.o.heinonen@aalto.fi

Whole Cell Simulation of Bacteria from Genomic Sequence
Kazunari Kaizu, Kozo Nishida, and Koichi Takahashi

Whole cell modeling is one of the grand scientific challenges in the 21st century. However, despite the accumulation of a huge amount of experimental data, the way of modeling is still a “black-box”, where a model manually emerges without formalization of its procedure. Here, we present a novel framework for precise genome-scale simulation of bacteria, i.e. Escherichia coli, based on the genome sequence. To achieve the prediction of phenotype from genotype, this framework accepts a genome sequence instead of a mathematical model as an input and automatically annotates the given sequence utilizing information stored in multiple bioinformatic databases. Based on the annotation, it generates a dynamic, stochastic, and single-cell model consisting of multiple pathways, such as gene expression, signaling, and replication, involving more than 10 thousand reactions and 1 million agents. Additionally, the whole cell simulation enables us to predict omics profiles directly from a genome sequence without knowledge about the mathematical details of the model. The automated modeling can facilitate the construction and maintenance of large complex models against massive knowledge updated constantly and allow us to design a genome from scratch toward an era of synthetic cells.

A Stochastic Model for Topographically Influenced Cell Migration
Adam Mitchinson

Topography is one of many features of the microenvironment known to affect cell migration. This interaction is critical to many physiological and pathophysiological processes (e.g. wound repair and cancer metastasis) and is exploited for application in biomedicine (e.g. medical implants and tissue engineering scaffolds). Despite a large body of experiment-based literature establishing cell response to different topographic configurations, little is understood about how topographies influence migration behaviour. This work aims to use mathematical modelling within a systems biology framework to better understand the dynamics of topographically influenced migration. Based on an Ornstein-Uhlenbeck process, the model describes velocity-time evolution of an individual point cell undergoing resisted Brownian motion, extended to incorporate directional bias caused by physical surface gradients. Numerical simulations produce individual cell paths with average properties comparable to that measured from experimental migration on grooved topographies. Preliminary comparison between model and experimental data suggest grooved topographies stimulate changes to both intrinsic (kinesis) and directional (taxis) migration for different groove widths, implying considerable motile sensitivity to physical dimension (consistent with literature findings). The intention is to use the model to predict migration upon topographies for potential use as an implant surface or tissue scaffold.

Repurposing drugs and identifying interventional targets for cardiovascular disease using qualitative logical modeling
Amel Bekkar, Julien Dorier, Isaac Crespo, Cristina Casals-Casas, Anne Estreicher, Anne Niknejad, Alan Bridge, and Ioannis Xenarios

Cardiovascular-diseases are multifactorial and complex pathologies that cannot be described by reductionist approaches. In order to tackle this, we developed a logical modeling framework composed of three components. The first step is an expert-based curation related to the body of literature we call Prior Knowledge Network (PKN). The PKN is assembled from the existing knowledge and experimental evidence. It includes the relevant components for Cardiovascular-diseases as well as the regulations between them. As compared to databases that register facts and summarize them, we have encoded the logical rules of regulations, enabling the use of the PKN for modeling and simulation. The second step simulates the cellular decision process and identifies the phenotypes attained by the regulatory network. As the PKN is large a manual optimization would be time consuming. Therefore we use Optimusqual, a method that uses a genetic algorithm to find in the PKN the regulatory sub-graph that fit to a training-set. In the final step we simulate several in silico perturbations. That allows to evaluate the pertinence of our model and to make predictions and generate testable hypotheses about driver nodes able to switch the network from a disease to the healthy state and ultimately find interventional targets.

An Extended Kappa Simulator for Agent and Rule Based Models
Ignacio Fuenzalida, Alejandro Bernardin, Pablo Monares, Alvaro Bustos, and Tomas Perez-Acle

The study of protein interactions is mainly accomplished with network-analysis. Although it’s a powerful tool, it fails in showing what the dynamic properties of these interactions are. Simulation tools help with this task but they are barely known in the area. Kappa is a modeling language based on rules and agents. Agents are particle-like entities which can bind and be modified according to rules.
We expand Kappa models with PISKaS, a software based on KaSim v3.0 which is capable to extend the modeling bounds efficiently.
Our implementation’s new features include spatiality, resource optimization, modeling flexibility, and statistical analysis. We can simulate volumes where agents travel among them using a parallel algorithm. Agents now have numeric internal-values, allowing rules to operate with them and even vary their reaction-rate. Several trajectories can run at once, which allows to optimize simulations based on previous results and extract some basic statistics.
Nowadays, PISKaS can perform just as KaSim, with no significant differences in time or accuracy. Furthermore, internal-values allow agents to represent more complex entities than proteins, such as cells or humans, making our software suitable to perform social simulations.

Partial economic support from FA9550−18−1−0438 AFOSR. AFB-170004 Fundacion Ciencia y Vida, ICM-Economia P09-022 CINV.

Logical models and cell-cell communication networks to investigate pattern formation
Pedro Varela, Pedro T. Monteiro, and Claudine Chaouiya

Models of multi-cellular systems need to account not only for cellular molecular networks but also for cell-cell communication that altogether orchestrate the dynamics of the whole. We present EpiLog, a software tool implementing a logical modelling framework to handle pattern formation on epithelia [1]. Briefly, this framework defines a cellular automaton in which each cell carries a logical regulatory model whose input nodes embody cell receptors. Integration functions specify how these receptors are activated depending on signals from neighbouring cells (how many, at what distance). EpiLog defines a fixed grid of hexagonal cells, with parametrisable size and boundary conditions.

To explore the validity of this fixed topological configuration, we consider different cell-cell communication networks and assess the resulting patterns of a simple lateral inhibition model. This study suggests that reasonable deviations from a hexagonal grid do not change much the characteristics of the resulting patterns. Furthermore, our study indicates that measures such as the number of shared neighbours between pairs of contacting cells and network regularity are relevant to qualify such deviations.

[1] Varela PL et al. EpiLog: A software for the logical modelling of epithelial dynamics [version 2; peer review: 3 approved]. F1000Research 2019, 7:1145

Computational Studies on Ureter Smooth Muscle: Modeling Ion Channels and their Role in Generating Electrical Activity
Chitaranjan Mahapatra and Rohit Manchanda

Abnormal peristaltic contraction of the ureter smooth muscle (USM) causes the pathophysiological condition to the urinary system. The USM contractions are discretely initiated by the USM cell action potentials (APs). Therefore, the USM AP is a dynamic parameter to investigate the abnormal USM contractions. In the interest of figuring out the internal membrane ionic currents responsible for USM AP origination, this paper aimed at developing a computational model of the USM cell AP. From various published electrophysiological recordings, we listed all major USM cell ion channels associated with generating cellular electrical activities. Then, we constructed all ion channel models after extracting biophysical details from the documented voltage clamp experiments in guinea pig USM tissue. The individual ion channel model properties were validated by comparing the simulation results with the experimental voltage-clamp results. Then, all ion channels were integrated to generate the USM AP after introducing a current stimulus to a single cell model. This model reproduced USM AP successfully that replicates the experimental AP. This model also allows analyzing the ion channel implications at a different phase of the AP. In the future, this primary model can be further extended to explore new intracellular insights for abnormal USM contraction.

Modelling bistable tumour population dynamics to design effective treatment strategies
Andrei Akhmetzhanov, Jong Wook Kim, Ryan Sullivan, Robert Beckman, Pablo Tamayo, and Chen-Hsiang Yeang

Drug resistance is driven by mutations and dynamic plasticity deregulating pathway activities and regulatory programs of a heterogeneous tumour. We propose a model to simulate population dynamics of heterogeneous tumour cells with reversible drug resistance. Drug sensitivity of a tumour cell is determined by its internal states demarcated by coordinated activities of multiple interconnected pathways. Transitions between cellular states depend on the effects of drugs and regulatory relations between the pathways. We build a simple model to capture drug resistance characteristics of BRAF-mutant melanoma, where cellular states are determined by two mutually inhibitory – main and alternative – pathways. Cells with an activated main pathway are proliferative yet sensitive to the BRAF inhibitor, and cells with an activated alternative pathway are quiescent but resistant to the drug. We describe a dynamical process of tumourgrowth, and compare efficacy of three treatment strategies from simulated data: static treatments with constant dosages, periodic treatments with regular intermittent active phases and drug holidays, and treatments derived from optimal control theory (OCT). Periodic treatments outperform static treatments with a considerable margin, while treatments based on OCT outperform the best periodic treatment. Our results provide insights regarding optimal cancer treatment modalities for heterogeneous tumours.

Taxonomic Gap Filling of Metabolic Networks Yields Increased Accuracy
Wai Kit Ong and Peter Karp

One of the most significant bottlenecks in metabolic-model development is the time required to add reactions to the network that were initially omitted due to incompleteness in the genome annotation. Algorithms have been developed to address this problem, called network gap filling, by choosing a minimal number of reactions from a universal reaction database such as MetaCyc, such that adding those reactions to the metabolic network enables the metabolic model to produce all biomass metabolites defined for the model. Our past studies have shown that reaction gap filling has a significant error rate. Here we present an enhanced taxonomic gap-filling algorithm with significantly improved performance. The algorithm assigns a lower cost to MetaCyc reactions that are found in other organisms within the same phylum as the organism being gap filled, on the assumption that different taxonomic groups are biased toward using different metabolic reactions. For example, when gap filling the Escherichia coli metabolic network, the algorithm assigns a lower cost to reactions found in other BioCyc databases for organisms in the same phylum (Proteobacteria) as E. coli. For gap filling an E. coli metabolic network containing randomly introduced gaps, gap filling accuracy increases from 87.8% to 97.3%.

Prediction of pharmacokinetic profiles using chemical structure information: Fraction unbound in plasma and renal excretion
Reiko Watanabe, Tsuyoshi Esaki, Hitoshi Kawashima, Yayoi Natsume-Kitatani, Chioko Nagao, Rikiya Ohashi, Hiroshi Komura, and Kenji Mizuguchi

In the early stages of drug development, prediction of pharmacokinetic profiles of new chemical entities is essential to minimize the risks of potential withdrawals. The pharmacokinetic profile of a drug depends on various properties including solubility, intestinal absorption, plasma protein binding, metabolism, biliary excretion, distribution and renal excretion. Recently, computer-aided drug design using in silico models to predict ADMET parameters (absorption, distribution, metabolism, excretion, and toxicity) has attracted much attention. In this study, we introduce several prediction models for pharmacokinetic parameters, mainly the fraction unbound in plasma (fu,p), renal excretion (fe) and renal clearance (CLr) of drugs in human. fu,p is an important determinant of drug efficacy and the excretion of unchanged compounds by the kidney is a major route in drug elimination and these parameters play an important role in pharmacokinetics. In this study, prediction models for fu,p were created initially, additional models for fe and CLr in human were generated using machine learning methods such as Random Forest and Support Vector Machine with predicted values of fu,p as a descriptor. Our prediction system, consisting of fu,p, renal excretion and other models, is freely available to the public and it can be used in screening processes of drug discovery.

Computational modeling of the immune response to Clostridium difficile infection
Meghna Verma, Josep Bassaganya-Riera, Nuria Tubau-Juni, Stanca M. Ciupe, and Raquel Hontecillas

Background: Clostridium difficile is a gram-positive spore-forming bacterium [1, 2]. Upon germination in the colon, it produces toxins that cause death of epithelial cells, severe colitis and potentially death [3].
Motivation: We built a mathematical model of immune responses to C.difficile generalizable to other gram-positive toxin-producing select agents potentially used as bioweapons, to determine the immunological mechanisms responsible for pathogenesis.
Method: The ODE model has 58 equations involving 34 species at the interface of host-pathogen interactions and describes the dynamics of: i) bacterium and toxins (tcdA/tcdB, binary); ii) innate responses: type 3 innate lymphoid cells (ILC3s), macrophages, neutrophils, eosinophils, iii) cytokines: IL22, IL25, IL1β (IL1beta), and iv) T and B cells. The model was calibrated using data from a mouse C.difficile infection (CDI) model. Sensitivity analysis identified critical factors influencing outcome of infection.
Results: A novel modeling prediction was that macrophage-derived IL1β enhanced the production of ILC3s which in turn controlled IL22 levels. Increased levels of colonic IL22 was beneficial by having a negative impact on epithelial cell damage during CDI.
Conclusion: The results highlighted IL1β production as key factor in reducing epithelial cell damage during CDI. These findings will be validated experimentally using mouse models and human primary cells.

Development of DruMAP, Drug Metabolism and pharmacokinetics Analysis Platform
Hitoshi Kawashima, Reiko Watanabe, Tsuyoshi Esaki, Chioko Nagao, Yayoi Natsume-Kitatani, Rikiya Ohashi, Hiroshi Komura, and Kenji Mizuguchi

We began an initiative “Development of a Drug Discovery Informatics System” supported by the Japan Agency for Medical Research and Development. The main aim of this initiative is to develop accurate prediction systems for DMPK (drug metabolism and pharmacokinetics), primarily targeting academic scientists.
We collected pharmacokinetic and physicochemical parameters from ChEMBL. However, since ChEMBL compiles data obtained in different experimental conditions, we selected the data measured in compatible conditions and reformatted the results as appropriate for our prediction system. In addition to the public data, we have acquired experimental data under unified protocols. The in vitro data include solubility, distribution coefficient, metabolic stability, fraction unbound in plasma, fraction unbound in brain homogenate, and blood-to-plasma concentration ratio. The in vivo data include the drug concentrations in plasma and several tissues after oral or intravenous administration of the drug and pharmacokinetic parameters calculated therefrom.
We stored these data to DruMAP database. We are currently developing prediction models for several pharmacokinetic parameters using these data. DruMAP also provides the ability to predict those parameters for user input compounds using our prediction models. DruMAP can be used for early DMPK studies and for candidate compound selection to accelerate novel drug development.

Metabolic Network Reconstruction of Treponema pallidum spp. pallidum
Silvia Morini, Isabella Casini, Thomas M. Hamm, Kay Nieselt, and Andreas Dräger

Since the discovery of Treponema pallidum ssp. pallidum (Tp) as the etiologic agent of syphilis in 1905, still no vaccine is available, and the world is still burdened by syphilis, which in its early stages enhances the transmission of HIV. Continuous in vitro culture of Tp has still not been achieved, imposing a substantial roadblock to its experimental inspection, and even the sequencing of its genome did not yield an obvious solution to the cultivation problem.

We present iSM161, a first manually curated draft reconstruction of the metabolic network in Tp towards a genome-scale metabolic model (GEM), comprising 161 genes (1039 predicted open reading frames), 239 reactions, and 277 metabolites. The model is still under development and steadily updated. For the reconstruction, COBRApy has been used, where subsystem information is added and parsed as SBML groups extension using libSBML.

Using this reconstruction, we anticipate to gather new insights into the pathogen’s physiology and pathology, and in how this spirochete, which has earned the designation of “stealth pathogen,” succeeds in making a living and eluding human’s immune defenses as well as cultivation attempts. It is planned to make the model available to the community in SBML format.

Mechanistic modelling reveals a mixed phosphorylation mechanism of Hog1 MAPK switching between low and high processivity
Maximilian Mosbacher, Matthias Peter, and Manfred Claassen

Receptor mediated signals are often propagated via a sequence of activating double phosphorylation events. The phosphorylation mechanism, which is commonly thought to be either distributive or processive, as well as potential positive and negative feedback loops, can strongly impact the response behavior. We try to pinpoint the mechanism and feedbacks needed to generate the ultrasensitive Hog1 response, a MAPK pathway in Saccharomyces cerevisiae.
We generated an ODE model that summarizes current knowledge of the pathway. To provide information we collected various publicly available data sets and our own measurements of Hog1 activation in different conditions. These were used to fit and parameterize different model variants, encoding different mechanisms of Hog1 phosphorylation and putative feedback loops.
The best fitting model incorporates a mixed phosphorylation mechanism that switches between a mainly distributive nature before and a more processive nature after activation. This change is induced by a Hog1 mediated positive feedback loop. In our simulations, this arrangement displays robustness to model perturbations, mediated by the more distributive mechanism, while the induced processivity is needed for quicker and total activation.

An integrative workflow to visualize Elementary Flux Modes in genome-scale metabolic models
Chaitra Sarathy, Martina Kutmon, Michael Lenz, Marian Breuer, Michiel Adriaens, Chris T. Evelo, and Ilja C. W. Arts

Elementary Flux Modes (EFMs) are an indispensable tool for constraint-based modelling and metabolic network analysis. However, systematic and automated visualization of EFMs, capable of integrating various data types is still a challenge. In this study, a semi-automated, customizable, MATLAB-based workflow was developed for graphically visualizing EFMs as a network of reactions, metabolites and genes. The workflow integrates COBRA and RAVEN toolboxes with the open-source tool Cytoscape and offers a platform for comprehensive EFM analysis, starting with EFM generation followed by visualization and data mapping. Once processed, network manipulations in Cytoscape were semi-automated using R along with application of the widely accepted SBGN layout, thus minimizing both time and user effort. The biological applicability of the workflow is demonstrated using EFMs generated from two genome-scale models, (1) a medium-sized E. coli model (iAF1260) and (2) a large-scale human model (Recon 2.2). Additionally, two different types of data, gene expression and reaction fluxes, were mapped onto the visualized EFMs, thereby illustrating that such integrated visualization can enable better understanding of the metabolic described by the EFMs.

Agent-Based Intelligent System Modelling of Immune-Tumour Interactions for In Silico Bioengineering
Clara Pavillet, Francesca Buffa, and Tudor Fulga

Agent-based modelling belongs to a class of discrete mathematical approaches used to model complex phenomena as dynamical systems of interacting autonomous entities (agents). They have gained increased attention in the field of cancer research, and a recently developed modelling framework, microC (https://microc.org), successfully illustrates how they can be used to model cellular behaviour within the broader three-dimensional physical environment in which they act. Cellular agents are equipped with network models, mathematical structures used to model pairwise relations, representing the interaction between genes and molecules which influence cellular behaviour. The combination of network modelling and agent-based modelling provides a unique opportunity to model genotype-phenotype relationships in-silico.

My current work expands on the microC framework to model the role of the immune system as a critical regulator of tumour development and progression. It makes use of Java’s object orientated paradigm to create sets of immune agents and components. This comprehensive model enables the in-silico study of immune-tumour interactions in three-dimensional space across different cancer types and mutation profiles. It provides a framework to predict patient-specific response to cancer therapy and to design new engineering-based approaches harnessing the patient’s immune system.

SBML to Knowledge Graph in Neo4j
Thorsten Tiede and Oliver Kohlbacher

The Systems Biology Markup Language (SBML) is often used to represent biological network models in a standardized and interoperable file format. Searching the models and accessing relational information stored in them is an aspect that is not easily covered with a file-based approach.
The aim of this project is to use the power of graph databases to make these biological models queryable, traverse pathways in them, and integrate multiple models into one unified graph representation.
In an Extract-Map-Connect (EMC) process we extract the model information, map the enclosed entities to graph nodes using object-graph mapping (OGM), and add relationship information between the elements according to the SBML specification.
Multiple models with matching identifier systems can be integrated in the same database by connecting them through common entities.
In this process, additional information from online services, like the public KEGG API or the Systems Biology Ontology, can be fetched and added to the emerging Knowledge Graph.
We store this Graph in a Neo4j database which offers sub-second query response times for retrieving model-spanning subnetworks and biologically relevant network-contexts. Graph algorithms like shortest path, target identification via breadth-first search as well as mapping of omics data are accessible through a RESTful API.

Modelling Xenopus egg extract aster microtubule autonucleation with RDMEcpp
Lukas A. Widmer and Jörg Stelling

Modelling microtubule dynamics in cells together with regulatory networks requires integrating the spatiotemporal evolution of regulators with stochastically growing and shrinking microtubules. In principle, such stochastic spatiotemporal models of cells can be simulated using the reaction-diffusion master equation (RDME)-type framework. However, no simulation software exists for RDME models with embedded filaments that, themselves, evolve according to a stochastic model. We therefore developed RDMEcpp, a high-performance, extensible, and cross-platform solution in C++ for simulating 1D-3D RDME-type models that require subvolume coordinate lookup at runtime, e.g., to determine concentrations along dynamic microtubules. RDMEcpp exhibits similar or better performance compared to the state-of-the art URDME on the MinD oscillation model from Escherichia coli. For 2D Xenopus laevis egg extract spindle autonucleation, we incorporate experimental nucleation angle measurements between microtubules as well as explicit microtubules; microtubules evolve according to a stochastic microtubule tip model, which determines the time until catastrophe occurs. Compared to previous deterministic partial differential equation models, the RDME model’s more mechanistically accurate microtubule nucleation and tip evolution lead to a more realistic predicted aster microtubule density . We anticipate future extensions of RDME models to include regulators of microtubule dynamics for detailed investigation of control mechanisms in vivo.

Workflow for TF knockout simulations on transcriptome primed whole genome metabolic model reconstructions for 10 antibiotic resistant E. coli strains
Daria Gaidar, Amanuel Ghirmay Araya, and Volkhard Helms

Based on published gene expression datasets from directed evolution experiments on E. coli exposed to 10 antibiotics and iML1515 genome wide metabolic reconstruction model, we developed a library of treatment specific genome scale models. Metabolic network reconstruction was performed with the COBRA (iMAT) and CORDA methods. Interaction probabilities for 1753 TF-gene pairs with strong experimental evidence were computed using the E. coli M3D dataset (264 samples). We used PROM to construct integrated regulatory-metabolic networks and run TF knockout simulations. For each virtual TF knockout, the growth defect of the model, Δgrowth, due to knockout was computed. If Δgrowth is larger for the resistant model than the wildtype model, this TF may be considered as potential target for future antibiotic intervention.
A total of 73 TFs associated with resistance to at least one antibiotic were identified. Their target genes engage predominantly in catalytic, binding and transporter activity and include known resistance associated genes like acrD, cyoC, folA, marA, tsx, ompF, and oppA. The TFs GadE, SoxR, and FliZ were predicted to be associated with resistance for most models. CORDA models were more responsive to TF knockout simulations.

Construction of Discrete Model of Human Pluripotency in Predicting Lineage-Specific Outcomes and Targeted Knockdowns of Essential Genes
Priyanka Narad, Lakshay Anand, Romasha Gupta, and Dr. Abhishek Sengupta

A network consisting of 45 core genes was developed for the genes/proteins responsible for loss/gain of function in human pluripotent stem cells. The nodes were included on the basis of literature curation. The initial network topology was further refined by constructing an inferred Boolean model from time series
RNA-seq expression data. The final Boolean network was obtained by integration of the initial topology and the inferred topology into a refined model termed as the integrated model. Expression levels were observed to be bi-modular for most of the genes involved in the mechanism of human pluripotency. Thus, single and combinatorial perturbations/knockdowns were executed using an insilico approach. The model perturbations were validated with literature studies. A number of outcomes are predicted using the knockdowns of the core pluripotency circuit and we are able to establish the minimum requirement for maintenance of pluripotency in human. The network model is able to predict lineage-specific outcomes and targeted knockdowns of essential genes involved in human pluripotency which are challenging to perform due to ethical constraints surrounding human embryonic stem cells. Source code to run the insilico simulations is provided in an open-source repository (GitHub) https://github.com/pnarad/hPluriNet-Boolean-Modelling.

Metabolic modeling of retinoblastoma in Indian population
Swagatika Sahoo, Ranjith Kumar Ravi Kumar, Brandon Nicolay, Omkar Mohite, Karthikeyan Sivaraman, Vikas Khetan, Pukhraj Rishi, Suganeswari Ganesan, Krishnakumar Subramanyan, Karthik Raman, Wayne Miles, and Sailaja Elchuri

Retinoblastoma (RB) is a childhood eye cancer, caused by loss of RB1 gene. It affects one or both the eyes, with a current global incidence of in 15,000 to 20,000 births. Chemotherapy, local therapy, and enucleation are the main ways in which RB is managed. However, these treatment strategies exhibit severe side effects, warranting a systems-level analysis of RB to predict novel diagnostics and safer therapeutics. Herein, we used mathematical modeling approach (i.e., constraint-based reconstruction and analysis) to identify and explain RB-specific survival strategies. While there was over-utilization of amino acids by RB for energy, there was under-utilization of cholesterol synthesis for preservation of redox potential. Variable synthesis of long-chain/very-long chain fatty acids was found to be classifying RB-subtypes. Further, our model-specific secretion profile was also found in RB1-depleted human retinal cells in vitro and suggests that novel biomarkers involved in lipid metabolism may be important. Finally, RB-specific synthetic lethals have been predicted as lipid and nucleoside transport proteins that can aid in novel drug target development.

Assessing the impact of biological aerosols on rainfall: effects of land cover diversity and landscape properties
Rachel Kohn, Davide Martinetti, Brent Christner, and Cindy Morris

Biological aerosols may influence atmospheric processes that lead to rainfall; however, their constant presence makes it difficult to identify when and where they might play an important role. To investigate this, we utilized data on rainfall feedback, a phenomenon where relatively heavy rainfall has a measurable effect on subsequent rainfall. Previous studies suggest that rainfall feedback persistent over several weeks is due to environmental phenomena that involve microbial growth. Rain-induced increases in surface populations of microorganisms leads to an increase in cloud-active bioaerosols which have the capability to influence precipitation formation. In this study, we utilized a geographically weighted regression model to explore the underlying mechanisms of rainfall feedback in the context of the continental USA. We found that known predictors of microbial aerosol diversity and abundance and precipitation pattern, such as mean temperature, mean precipitation, and landcover composition, are potential determinants of rainfall feedback. Relationships between rainfall feedback and landcover type are spatially non-stationary: both the strength and the direction of the relationships differ over space. Our study also identified that certain landcover types are more favorable for positive rainfall feedback, and thus likely favorable to the production and emission of bioaerosols in non-limiting quantities.

Identifying knock-outs when the desired target chemical is toxic for the microorganism
Irene Ziska, Mariana Ferrarini, Nuno Mira, Susana Vinga, and Marie-France Sagot

Metabolic engineering is a common technique used to improve the production of target chemicals in microorganisms. There are already various published methods that identify potential metabolic engineering strategies such as gene and reaction knock-outs. However, most of the available approaches do not explicitly take into account that in some cases, the target chemical can be toxic for the microorganism itself, which might render the production unstable.
We are currently developing a method that aims to identify knock-outs which increase the production of the target and which, at the same time, ensure that the microorganism keeps a high resistance against the toxic target. In a first step, our approach uses multi-objective linear optimization to find valid trade-offs between growth, target production and toxicity resistance against the target. Afterwards, potential knock-outs are enumerated and then ranked to choose the best candidates for a desired trade-off. The toxicity resistance is measured by the activity of a set of critical reactions that have to be known or identified experimentally as a prerequisite.
To test our method, we are applying it to identify knock-outs for the production of ethanol in yeast.

Modelling potentially virulence-associated metabolic pathways in Pseudomonas aeruginosa PA14 including experimental verification
Alina Renz, Michael Sonnabend, Kristina Klein, Erwin Bohn, Monika Schütz, and Andreas Dräger

Pseudomonas aeruginosa (Pa) is a Gram-negative opportunistic pathogen. Its potential virulence, its intrinsic and acquired resistance, and its ability to cause primary health-care associated infections, motivate the research of pathogenicity of Pa.

Based on the hypothesis that the absence of genes can potentially increase the pathogenicity of Pa, we aim to identify such genes and to validate their modulating role for the pathogen’s virulence. Whole-genome sequencing data of the Pa strain PA14 and a patient isolate associated with high mortality were used to identify single nucleotide polymorphisms (SNPs). The potential consequences of the genetic differences were analysed using a genome-scale metabolic network model of PA14 and information about virulence factors. Assuming SNP variants could cause the loss of function of a gene product, the SNP-affected genes were subsequently knocked out in the model, and its effect on virulence-associated reactions was monitored using flux balance analyses. Promising candidate genes were validated in the Galleria mellonella infection model using mutant Pa strains. First experiments suggest a modulating role of the urocanate hydratase gene in Pa by increasing the virulence when being absent.

The insight into factors enhancing and modulating the virulence could be used to detect new targets for therapeutic approaches.

Understanding Regulatory Mechanisms Underlying Stem Cells Helps to Identify Cancer Biomarkers
Maryam Nazarieh

Detection of biomarker genes plays a crucial role in disease detection and treatment. These computational approaches enhance the insights derived from experiments and reduce the efforts of biologists and experimentalists. This is essentially achieved through prioritizing a set of genes with certain attributes.
In this work, I show that understanding the regulatory mechanisms underlying stem cells helps to identify cancer biomarkers.
We got inspired by the regulatory mechanisms of the pluripotency network in mouse embryonic stem cells to formulate the problem where a set of master regulatory genes in regulatory networks is identified with two combinatorial optimization problems namely as minimum dominating set and minimum connected dominating set in weakly and strongly connected components. We applied the developed methods to regulatory cancer networks to identify disease-associated genes and anti-cancer drug targets in breast cancer and hepatocellular carcinoma.
As not all the nodes in the solutions are critical, I developed a new prioritization method named TopControl to ranks a set of candidate genes which are related to a certain disease. Moreover, this work shows that the topological features in regulatory networks surrounding differentially expressed genes are highly consistent in terms of using the output of several analysis tools.

Representation of ME-Models in SBML
Marc Alexander Voigt, Colton J. Lloyd, Laurence Yang, Zachary A. King, Oliver Kohlbacher, Kay Nieselt, and Andreas Dräger

Metabolism and Expression models (ME-models) are a constraint-based modeling approach, which explicitly accounts for the cost of macromolecular biosynthesis. Explicitly representing these processes allows ME-models to investigate genotype-phenotype relationships with quantitative incorporation of ‘-omics’ data. Common standards are a prerequisite for interoperability of systems biology tools. Novel approaches often require additional or changed data structures that existing standards can often not directly represent. ME-models are powerful tools, but a lack of standards for encoding and creation, and the increased complexity of these models prevented widespread use.

SBMLme extends current model encoding standards and enables the representation of the ME-Model variant COBRAme in SBML. A prototype of the extension has been created in Java together with a standalone, bi-directional converter, between this extension and COBRAme’s model storage format. The converter showed that SBMLme could fully and correctly encode a COBRAme model.

The use of SBMLme enables sharing ME-models more efficiently and a wider variety of tools to access ME-models, promoting their propagation. SBMLme is a proof-of-concept towards an official SBML package for ME-models.

SBMLme is freely available at https://github.com/draeger-lab/SBMLme (under MIT license).

Using Protein Localization Studies to Improve Genome-scale Metabolic Models
Helen Tung, Rong Ma, Sebastian Lee, and Uri David Akavia

Cancer cells reprogram metabolism to support rapid proliferation and survival. Energy metabolism is particularly important for growth. A genome scale metabolic model (GSMM) is a mathematical formalization of metabolism which allows simulation and hypotheses testing of metabolic strategies using Constraint-Based Modeling.
GSMMs are based on the accumulated knowledge of metabolic reactions and enzymes in the published literature where reactions are represented mathematically as a stoichiometric matrix. One of the widely used generic human metabolic models is Recon 2.2, containing 7785 reactions, 5324 metabolites and 1675 genes.
The accuracy of predictions using GSMMs is dependent on accurate gene associations to reactions. Current GSMMs do not consider intracellular protein localization in annotating GPR relationships which may lead to false-positives. In this study, protein localization data obtained from 10 different studies were used to verify the intracellular location of each protein. Mismatches between protein localization and reaction location were corrected.
The modified Recon 2.2 includes 8135 reactions, 5801 metabolites and 1756 genes, an addition of 5%, 10% and 5%, respectively. The improved model can predict energy productions correctly and perform more known metabolic tasks. Improving prediction accuracy of Recon 2.2 will facilitate identification
of biomarkers and drug targets using contextualized GSMMs.

Inferring subcellular compartmentalized flux in cancer cells: A new approach integrating isotope tracing with thermodynamic analysis
Alon Stern, Tomer Shlomi, Boris Sarvin, Won Dong Lee, and Elina Aizenshtein

The inability to inspect metabolic activities within distinct subcellular compartments has been a major barrier to our understanding of eukaryotic cell metabolism. Numerous isozymes catalyze the same metabolic transformation in different compartment, having different flux, potentially in opposite directions. A direct approach for quantifying intracellular metabolic flux is isotope tracing coupled with computational Metabolic Flux Analysis (MFA). However, utilizing this approach with metabolic measurements performed on a whole-cell level typically limits its applicability to inferring only the total flux through all subcellular organelles. Here, we developed a computational method for inferring cytosolic and mitochondrial specific metabolic fluxes based on whole-cell level measurements of metabolite isotopic labeling and concentrations. This is made possible by integrated modeling of compartment-specific isotope tracing as well as reaction and membrane transporter thermodynamics – where inferred Gibbs free energy of reactions in each compartment is associated with rates of isotope exchange (forward-to-backward flux ratio). We applied our method to several proliferating cancer cell lines, deriving a first comprehensive view of the interplay between mitochondrial versus cytosolic fluxes in central metabolism under physiological conditions. We expect this approach to be a highly useful tool for probing cytosolic and mitochondria metabolic dysfunction in cancer and other human diseases.

Modelling gene regulatory networks in oncogene-induced senescence
José Américo Nabuco Leva Ferreira de Freitas, Pierre-François Roux, Ricardo Iván Martínez-Zamudio, Lucas Robinson, Gregory Doré, Nir Rozenblum, Benno Schwikowski, and Oliver Bischof

Cellular senescence (CS) is a cell fate that arrests cell proliferation in response to numerous stresses most notably oncogenes such as RAS. The phenotypic transformations that occur in CS include cell cycle arrest, inflammatory responses and a complex metabolic shift. An emerging paradigm stipulates that senescent cells are major contributors to health and age-related illnesses, particularly cancer. As such, research on therapeutic strategies targeting senescence to treat cancer and improve healthspan has gained enormous momentum in recent years. The phenotypic and transcriptomic changes that occur in CS can be interpreted as transitions in a high-dimensional state space, where each dimension corresponds to a molecular species. These transitions are specified by the architecture of its underlying gene-regulatory network (GRN), which represents the possible molecular interactions encoded in the genome. In order to describe and predict the mechanisms governing CS, we applied the Sparse Identification of Nonlinear Dynamics algorithm (SINDy) in a high performance computing environment to datasets containing time-course gene expression data on cells undergoing senescence. Our proposed predictive modeling approach will provide a deeper understanding of cellular senescence and has the potential to unravel unknown vulnerabilities of senescent cells that may be exploited to promote healthspan.

Mathematical Modeling of Macronutrient Signaling in Saccharomyces cerevisiae
Amogh P. Jalihal, Pavel Kraikivski, T. M. Murali, and John Tyson

Cells continuously assess the nutrient sufficiency of their environment in order to adapt to environmental fluctuations. In the budding yeast S. cerevisiae nutrient sensing and signaling is carried out by nutrient-specific signaling pathways. For example, the cAMP/Ras2/PKA pathway senses glucose sufficiency, and the TORC1 pathway conveys amino acid sufficiency. While these pathways have been extensively characterized for a nutrient specific response, signal integration in order to regulate global cellular state remains poorly understood at the level of the entire signaling network. Here, we present a dynamical model of the nutrient signaling system in yeast and parameterize it with data curated from the literature. The model readouts include the major transcription factors (TFs) responsible for growth control and nutrient adaptation responses. We use this model to explore the space of nutrient responsive cellular states defined by the TF levels. Next, we investigate the effect of single gene deletions on the space of cellular states. Our study of systematic gene deletions reveals the creation of novel nutrient responsive cellular states, as well as the loss of wild type states in various mutants. This approach provides a methodology for systematically simulating models to suggest testable experimental hypotheses.

A mechanistically detailed model of the cell cycle in Saccharomyces cerevisiae
Ulrike Münzner, Edda Klipp, and Marcus Krantz

Computational models of biological systems aid to understand how cellular properties and functions emerge from individual reactions. Model formalization requires accurate and scalable mapping of the available data. However, both scalability issues and data coverage challenge the development of large-scale models of signaling networks.
Here, we present a mechanistically detailed model of the molecular network that controls the cell cycle in Saccharomyces cerevisiae. We use the reaction-contingency (rxncon) language to establish a knowledge base, enabling translation into a bipartite Boolean model (bBM). We use the bBM to evaluate the knowledge base and to predict genotype-to-phenotype relationships. Our model reproduces wildtype behavior on the level of macroscopic observables, and correctly predicts 62 out of 85 tested phenotypes.
In the future, a similar effort may pave the way towards human whole-cell models.

Logical approach to identify Boolean networks modeling cell differentiation
Stéphanie Chevalier, Christine Froidevaux, Andrei Zinovyev, and Loïc Paulevé

Modeling effect of experimental perturbations on the functioning of biological networks governing cell differentiation in health or disease can improve our capacity to achieve desired system behaviours (for example, reprogram the differentiation process). To this aim Boolean networks (BNs) enable modeling large biological networks offering the level of abstraction that can match the current knowledge and the measurement accuracy, but identifying BNs whose dynamics is compatible with the data is a heavy combinatorial problem. Some of the existing methods focus on the logical model inference from constraints resulting from the requirement of reachability between sequential temporal observations. However, this type of constraint is insufficient to represent differentiating trajectories, characterized by irreversible bifurcations into distinct final stable states. Here our contribution is to take these features into account, formalizing them in Answer Set Programming as trap spaces and non reachability constraints. We have tested the new constraint types on a toy example of neuron precursor differentiation model. The method allows identifying few hundreds BNs compatible with the data from hundreds of millions of possible BNs. Our approach can be readily applied to binarized bulk cell population molecular data, and to single-cell data after proper pre-processing steps.

Systems biology in hematopoietic cell stem and progenitor populations: Integrating multiple *omics datasets to understand differentiation
Jens Lichtenberg, Guanjue Xiang, Elisabeth Heuston, Belinda Girardine, Cheryl Keller, Ross Hardison, and David Bodine

Systems biology integrates genomic profiles of specific cell types to generate functionally-testable hypotheses of lineage-specificity. Here we compare RNA expression, DNA methylation, chromatin accessibility, DNA binding proteins and histone modification profiles in seven different hematopoietic populations using a Bayesian non-parametric hierarchical latent-class mixed-effect model known as IDEAS to characterize epigenetic changes associated with hematopoietic differentiation.
Previous hematopoietic epigenome segmentation studies have focused on histone modifications, chromatin accessibility and DNA binding protein profiles. DNA methylation has been shown to vary markedly in hematopoietic populations. Inclusion of DNA methylation in these segmentation studies increased the original 36-state model of regulatory interactions to 41 states. These new DNA methylation-related states were associated with repressive marks, active RNA transcription, and a novel state regulated by DNA methylation alone. Imputing epigenetic models on inputs systematically perturbed for hematopoietic populations resulted in epigenetic models of varying degrees of overlap, which were quantified and set in context with underlying biological processes. We furthermore leveraged these imputation-related differences to infer potential lineage-specific impacts on regulation.
Our data show that methylation has a strong impact on functional genomic modeling and can be used to discern cell type specific epigenetic regulatory behavior by leveraging imputation for missing cell type data.

Laura Baumgartner, Miguel Ángel González Ballester, and Jérôme Noailly

Intervertebral disc (IVD) failure is closely related to low back pain, i.e. one of the biggest health-care burdens worldwide. Despite the extensive effort in IVD research, underlying mechanisms of IVD failure are unknown, which might be related to a lack of information about the behavior of IVD cells exposed to a multifactorial environment. This work aims at estimating cellular behavior within their corresponding micro-environment by developing a methodology that couples experimental cell-culture results with a systems biology approach. Therefore, Agent-based modelling (ABM) is introduced as a new technique in IVD research. Nutritional factors as well as inflammation are considered to estimate cell viability and mRNA expression of nucleus pulposus cells. Results are in good agreement with data from the literature, since qualitative cell viability prediction and mRNA expression fit well experimental data. The latter might furthermore provide unique explanation possibilities for surprising results from in-vitro cell culture experiments. Both results underline the utility of this new methodology to create cell regulation networks. Furthermore, provided results might be extended to multiscale models, including cell signaling or gene regulation networks to provide a deeper understanding of cellular interaction with the microenvironment.

A Hybrid Framework based on Network Analysis and Regression Optimisation Models for QSAR Applications
Jonathan Silva, Lazaros Papageorgiou, and Sophia Tsoka

Quantitative Structure-Activity Relationship (QSAR) methods employ features of chemical compounds to model molecular properties such as activity against a target. QSAR models are important in drug discovery, for example in lead optimisation and virtual screening of molecular compounds. Recently, the need for models that are not only predictive but also interpretable has been highlighted.
We report a methodology for QSAR modelling by combining elements of complex network analysis and piecewise linear regression. The algorithm, modSAR, employs a two-step procedure where first, compounds associated with a common target are represented as a network in terms of their structural similarity, revealing modules of similar chemical properties through network community detection. Second, a novel mixed integer programming optimisation model is used to subdivide every module into subsets (regions), each region modelled by an independent linear equation. The piecewise linear optimisation step involves determining an optimal molecular feature to separate data into regions as well as linear equations to predict the outcome variable in each region, and includes a regularisation term to prevent overfitting and implicitly selecting most informative features.
Comparative analysis shows that modSAR offers similar predictive accuracy to popular algorithms, such as RF and SVM, while also being interpretable and mathematically descriptive.

In silico study of protein-inhibitor interactions using PTP1B as molecular target
Andrea Coronell Tovar and Martin González Andrade

Protein tyrosine phosphatase 1B (PTP1B), an anti-diabetic molecular target, is a cytosolic phosphatase which plays an important role in the negative regulation of the insulin signaling pathway, so it is a key element in maintaining glucose homeostasis and in the molecular mechanism of insulin resistance, a relevant condition in the pathogenesis of diabetes and some other metabolic disorders.
The identification of selective PTP1B inhibitors has increased considerably in recent years. Nevertheless, owing to the subcellular location and structural properties of the enzyme, the design of pharmaceutically acceptable inhibitors remains a challenge. With the aim to find an inhibitor targeting PTP1B, an extended search with the allosteric and active site of the protein through structure-based virtual screening (with AutoDock Vina and iDock) and molecular docking was carried out with crystal structures of PTP1B and 20410 chemical compounds from Zinc database. They were selected according to their origin (biogenic or not), FDA approval, human use, and commercial availability. Molecular dynamics simulation shows complex stability formed by PTP1B and irinotecan, which was the substance that presented best bind affinity to the enzyme (-37.0 Kcal/mol). Findings suggest that irinotecan as a potentially good PTP1B inhibitor. Future enzyme assays can be done to prove it.

Key dates

April 11, 2019: Abstract submission deadline
May 9, 2019: Abstract acceptance notification
May 15, 2019: Late poster submissions deadline

May 23, 2019: Late poster acceptance notifications
June 20, 2019: Early registration deadline
Sunday-Thursday July 21-25, 2019: ISMB/ECCB conference
July 22, 2019: SysMod meeting

Abstract submission

We will accept abstracts for contributed oral presentations and posters. Abstracts should briefly (approximately 200 words) summarize the background/motivation, methods, results, and conclusions of your study.

For consideration for an oral presentation or poster, please submit your abstract via ISMB online system external-link by April 11, 2019.

For consideration for a late poster presentation, please submit your abstract via ISMB online system external-link by May 15, 2019.

Registration and fees

Please register for the SysMod meeting through the ISMB conference registration external-link. Detailed information will be available starting March, 2019.


We intend to provide a small number of travel scholarships to students and postdocs.


Accommodations will be available through the ISMB/ECCB conference. Please see the ISMB/ECCB website external-link for more information.

More information

For more information, please contact the SysMod coordinators 🔗.