SystemsModels

From BioNetWiki

This page is being assembled as part of an exercise for the 2012 Cell and Systems Modeling Class at University of Pittsburgh and Carnegie Mellon University taught by James Faeder and Christopher James Langmead. The main idea here is to highlight papers from the last decade since the advent of Systems Biology in which a mechanistic mathematical or computational model of a biological system was used in conjunction with experimental approaches to make new discoveries.

Each mini-review consists of a short paragraph describing (1) the system being studied, (2) the type of model that was used, and (3) how the model and experiments were used to inform each other in some way.

Contents 1 NF-κB Signaling 2 In Vivo Human Metabolism 3 Influenza A 4 T Cell Antigen Discrimination 5 Synthetic Gene Networks 6 Environmental Signal Integration by a Modular AND Gate 7 Immune system simulation 8 Tumor Growth Dynamics Under DOX Treatment 9 Basal Signaling Heterogeneity 10 Modeling of the Nervous System 11 Roles of Ant Pheromones During Decision Making 12 Oscillations by the p53-Mdm2 feedback loop 13 Information in Noisy Networks 14 A method for zooming of nonlinear models of biochemical systems 15 Integrated Genomic and Proteomic Analyses of a Systematically Perturbed Metabolic Network 16 Computational Modeling of Cellular Signaling Processes 17 Functions of MdmX 18 Modeling drug perturbations 19 Response of C. elegans reproduction to chronic heat stress 20 Cholesterol 21 An Integrated Model of Epidermal Growth Factor Receptor Trafficking and Signal Transduction 22 Pou5f1-dependent Transcriptional Networks 23 Probabilistic model for Protein-Protein Interaction Prediction 24 Predicting HIV-1 drug resistance 25 Rapid Evolution in Predator Prey Dynamics 26 Environmental and genetic pertubations reveal different networks of metabolic regulation 27 Representation of morphed odors in the rat olfactory bulb 28 RANK-RANKL-OPG Feedback in Bone Remodeling

NF-κB Signaling

This paper explores the role of negative feedback regulators on activation of the transcription factor NF-κB, which plays a critical role in the activation of immune responses.

Reference 

A. Hoffmann, A. Levchenko, M. L. Scott, and D. Baltimore (2002) The IκB-NF-κB Signaling Module: Temporal Control and Selective Gene Activation, Science, 298, 1241-1245. link.

The Model 

A differential equation model of NF-κB signaling was developed and analyzed that includes both biochemical and transciptional events. The model treats separately three different isoforms of IκB, which negatively regulate by binding to it and holding it in an inactive state. Parameters of the model for the different isoforms were determined by separately fitting data from knockouts in which one or two of the three isoforms were deleted.

The Interaction 

The model reproduces the damped oscillations in NF-κB activity that are observed in wild type and shows them arising as a sum of oscillatory behavior induced by the alpha isoform and sustained negative feedback from the beta and epsilon isoforms. The model was then used to explore how the system would respond to transient versus sustained signals induced by ligand. It was found that the strong negative feedback through the alpha isoform was required to rapidly turn off signals induced by transient signals. This was one of the first papers that used a mathematical model to explore the signal processing characteristics of a signal transduction network.

Reviewer 

Jim Faeder

In Vivo Human Metabolism

In this paper an energy balance was created to capture the effects of food intake on the three main components of metabolism; fat, glycogen, and protein.

Reference 

K. D. Hall (2006) Computational model of in vivo human energy metabolism during semistarvation and refeeding, AJP -Endocrinology and Metabolism, 291, E23-E37. link.

The Model 

Fat, glycogen, and protein stores in the body were modeled using differential balances and physiologically based transport parameters. An overall model was developed to track the effects of starvation on overall body mass and fat mass. A minimal amount of fitted parameters were used to integrate the model to a previous in vivo study followed by a verification using an entirely different study. Biologically determined metabolic rates were used with measured metabolizable energy intake as inputs for the model.

The Interaction 

This model accurately captures the fluctiations in overall body mass and fat mass induced by starvation and refeeding. The purpose of the model was to gain insight into the changes in human energy composition caused by variations in dietary energy sources. The model also captured the body's reaction to starvation and the overshoot in fat storage induced when refeeding, which was a major goal of this model. Insight into the fat-storage overshoot should allow for a more dynamic approach to weight loss and other model-driven dietary plans.

Reviewer 

Ari Pritchard-Bell

Influenza A

This paper demonstrates a mathematical model of the replication of influenza A virus in animal cells. Understanding these dynamics is essential for creating more potent influenza vaccines.

Reference 

L. Mohler, D. Flockerzi, H. Sann, and U. Reichl (2004) Mathematical Model of Influenza A Virus Production in Large-Scale Microcarrier Culture, Biotechnology and Bioengineering, 90, 46-58. link.

The Model 

The authors first infected a suspension of Madin-Darby canine kidney (MDCK) cells with influenza A, and then they tracked the progression of the influenza infection rate through the healthy cells. A time-delay differential equation model of virus infection, similar to the standard 3-equation SIR model, was created and analyzed in MATLAB for dynamics of the solution set. The parameters used in this model were primarily derived from experimental data from the authors’ own MDCK experiments.

The Interaction 

The model represents well the true dynamics of the influenza A virus. The model’s solutions showed the healthy cells died off after 3 to 4 days, which corresponded to the time frame garnered from the MDCK experiments. Experiments showed the maximum hemagglutinin (HA) titer levels about 50 hours after the initial infected cells were exposed to the healthy culture, and the model was able to reflect this maximum virus level at 50 hours. This paper provided valuable information as to the values of parameters of these influenza models, as well as good insight into the basic timeline for influenza infection in animal cells.

Reviewer 

Ericka Mochan

T Cell Antigen Discrimination

This paper constructed a new mathematical model of proximal T cell receptor(TCR)-dependent signaling, based on the observed competition between digital positive and negative feedback loops on ERK (extracellular signal-regulated kinase) activation.

Reference 

Gregoire Altan-Bonnet, Ronald N. Germain (2005) Modeling T Cell Antigen Discrimination Based on Feedback Control of Digital ERK Responses, PLoS Biol. 2005 Nov;3(11):e356. link.

The Model 

To trigger T cell antigen recognition, TCR and pMHC (peptide–major histocompatibility complex, on the surface of antigen-presenting cells) form a initial complex. The rest of the model is a feedback control system of two positive loops and two negative loops. Downstream event MAPK activation further amplify the input signal (up-regulates the reactions from the TCR-pMHC complex) while another event, Shp-1 cascade down-regulates them.

The Interaction 

Three predictions were made and tested. 1) Nonlinear lengthening of the response time at low ligand densities (predicted feature from the model and also observed in simulations). From experiments, the authors obtained qualitative agreement with their model, but there is a systematic quantitative discrepancy that could not be resolved by parameter adjustment. 2) The hierachy of antagonism, or the fact that better binders will be better inhibitors until a bifurcation point is reached. Previous experimental results are consistent with this prediction. 3) Precise positioning of the kinetic threshold between agonist and non-agonist ligants is set by the balance between positive and negative feedback loops. The balance should be highly sensitive to small changes in concentration of key components in the system. The authors confirmed this by experiments w.r.t. Shp1.

Reviewer 

Lingxue Zhang

Synthetic Gene Networks

This paper presents and demonstrates a new approach to engineering synthetic gene networks. Their modeling method allows for construction of functionally detailed libraries and modules, greater predictability, and decreased development time. They prove their system by creating a gene network to control sedimentation timing in S. cerevisiae.

Reference 

T. Ellis, X. Wang, J. J. Collins (2009) Diversity-based, model-guided construction of synthetic gene networks with predicted functions, Nature Biotechnology, 27, 465-471. link.

The Model 

The authors experimentally determined maximum and minimum output values for an engineered library of regulatory promoters. Using these data, they were able to create a mathematical model of feed-forward loop networks containing the promoters and in turn predict the network output using in silico models (devoid of interconnected network interference). These models were tested in vivo and were well correlated to the computational predictions.

The Interaction 

The verified in silico model of the promoter library in addition to another profiled library allowed construction of a more complex gene regulatory network: a toggle switch on a timer. A mutual inhibitory network was used to create this timed toggle switch which, when activated, triggered flocculation (sedimentation) in yeast. An intermediate experiment was required to assess the finer quantitative parameters that would affect the reset time of the switch. Using the input-output data from the libraries and the temporal data the authors were able to create a model that correctly predicted phenotype activation. These methods can be rapidly deployed to industrial synthetic biology and present a new, more efficient paradigm for creating functionally exploitable and predictable gene networks.

Reviewer 

Collin McCormack

Environmental Signal Integration by a Modular AND Gate

This paper discusses the design of a synthetic AND gate in E. Coli. The AND gate uses two promoters as inputs and another promoter as output. Specifically, one input promoter is used to control the transcription of T7 RNA polymerase, which has been engineered to contain two amber stop codons. These stop codons, by themselves, will cause premature termination of translation. However, when the amber suppressor, controlled by the other input promoter, is transcribed, the amber codons will be translated as serine. Thus, the T7 polymerase is expressed only when both promoters are activated. The T7 polymerase will then activate the output promoter.

Reference 

J. C. Anderson, C. a Voigt, and A. P. Arkin, “Environmental Signal Integration by a Modular AND Gate.,” Molecular Systems Biology, vol. 3, no. 133, p. 133, Jan. 2007 link

The Model

A transfer-function model is derived based on the underlying biochemical reaction and a translation model.

The Interaction

The output promoter has been connected to GFP, whose intensity can be measured under different input conditions. The data collected is then used to fit and determine the parameters of the transfer-function model. The AND gate is modular in the sense that the input promoters can be changed while the output promoter can be connected to different genes. This allows the AND gate to connect to different cellular signals. To demonstrate modularity, experiments are performed to connect different input and output signals to the AND gate and the resulting circuit is shown to be working.

Reviewer

Wing Chiu Tam

Immune system simulation

In this paper a dynamic system model called lattice gas automata is employed to simulate the immune system of HER-2/neu mice, and this model provides researchers a new tool to evaluate different vaccination scenarios that can be used to prevent tumor occurrence.

Reference 

F. Pappalardo, P.-L. Lollini, F. Castiglione, and S. Motta(2005) Modeling and simulation of cancer immunoprevention vaccine, Bioinformatics, 21(12), 2891-2897.link

The Model

The model is developed based on lattice gas automata method which is often used to simulate fluid flows. In this model, the whole immune system is represented as numerous lattice cells, each lattice cell integrates several different cellular or molecular entities information such as cell age, cell state, antigen, antibody and etc. The evolution of this system is carried out by simulating several specific events such as diffusion, interaction or mutation. Parameters of this model are basically determined according to physical law and physiological facts.

The Interaction

By setting initial parameters such as the state of each cell, vaccine concentration or the administration time, the author performed 200 in silico experiments. The result is highly agreed with those of in vivo experiments with same parameters. In addition, the simulation reveals that after several cycles of vaccination, there is a prevalent antibody response and a lack of cytotoxic T cell activity, which is validated experimentally. The simulation also suggests that there is early activation of the cytotoxic T cell response after first cycles of vaccination, which needs further experiments to validate.

Reviewer

Yue Yu

Tumor Growth Dynamics Under DOX Treatment

The authors explore using ODEs to create a hierarchical model for several pathways known to be dysfunctional when several types of cancer are present while DOX is administered to a patient. Furthermore, the authors employ a system of ODEs to link interrelated pathways.

Reference 

D. Ribeiro and J. M. Pinto (2009) An integrated network-based mechanistic model for tumor growth dynamics under drug administration, Computers in Biology and Medicine, 39, 368-384. link.

The Model 

The model estimates the effect of a drug on a tumor by having a differential equations based method with parameters for accumulation of the drug in the cancerous region, as well as dissipation in the blood and effects downstream in the signaling pathway. The amount of DOX entering the tissue then may be absorbed into the cancerous portion of the tissue, may be held in a healthy cell within the cancerous region, may re-enter the bloodstream, or it may eliminated from the system.

The Interaction 

The initial parameters for the speed and dosage of the drug were varied and the system was solved numerically, with other parameters determined from previous work in the PD and PK subsystems. The result reflected higher of DOX in cancerous tissue and lower levels of dissipation of DOX in the blood for long-time low-dose administration of the drug, in agreement with the medical literature. However, the hierarchical network-style approach employed by the authors is admittedly a first step in cancer dynamics, which would require a better understanding of the high-dimensional cell-circuit space before predictive results can be easily interpreted.

Reviewer 

Ted Roman

Basal Signaling Heterogeneity

This paper investigates whether patterns of basal signaling heterogeneity contain relevant biological information predictive of subsequent population response to perturbation. Understanding these patterns is very important for drug design and specifically, understanding drug sensitivity.

Reference 

Singh DK, Ku CJ, Wichaidit C, Steininger RJ, Wu LF, Altschuler SJ. Patterns of basal signaling heterogeneity can distinguish cellular populations with different drug sensitivities. Molecular Systems Biology. 2010; 6:369. link.

The Model 

The authors generated a collection of 49 clonal populations from the highly metastatic non-small cell lung cancer cell line H460. Then they observed patterns such as growth rate, local cell density, and cell morphology from diverse signaling pathways associated with cancer. The used computational tools to decompose the heterogeneous cellular distributions into a small set of homogeneous cellular phenotypes . They also applied a applied a classification method for predicting drug sensitivity to determine functional differences in heterogeneity of subpopulations. Model parameters were mostly learned from experimental data.

The Interaction 

The models created did successfully find cell signaling phenotypes within the population. The models were then used to cluster subpopulations based on patterns of heterogeneity, identify a significant relationship between heterogeneity and function, and finally test a predictive model of drug sensitivities across different cancer populations. This paper brings up many interesting future questions such as the evolution of phenotypic origin and why is their predictive model able to classify drug responsiveness based on cellular pathway information.

Reviewer 

Santosh Bhavani

Modeling of the Nervous System

This paper explores an integrated multi-scale simulation in the brain and the complex interactions taking place at the distinctive steps from cellular to systems level.

Reference 

Bouteiller, Jean-Marie C.; Allam, Sushmita L.; Hu, Eric Y.; Greget, Renaud; Ambert, Nicolas; Keller, Anne F.; Pernot, Fabien; Bischoff, Serge; Baudry, Michel; Berger, Theodore W (2011), Modeling of the nervous system: From molecular dynamics and synaptic modulation to neuron spiking activity."IEEE".2011.link.

The Model 

The model was divided based on the spatio-temporal hierarchical complexity where the first, molecular level was modeled using variable step numerical methods. The following, synaptic level takes into account the higher level of inter-connectivity, which was modeled using the EONS/RHENOMS platforms represented by thousands of differential equations. The subsequent step to neurons and extension to network level needed the calculations to be performed in parallel so they could communicate with each other as the simulation evolves. Various modeling tools were chosen to be integrated and each from its area of proficiency.

The Interaction 

The modeling approach incorporates complex non-linear dynamics from the sub synaptic bio molecular level up to the neuron level. The model was directly applied to study the effect of drugs on the nervous system, as it provided an integration of the effects at the molecular level into a multi-scale “pathophysiological modeling framework”. This model could prove useful in explaining the changes across networks in different multi-scale models if it proves flexible enough to adapt to different systems.

Reviewer 

Nitesh Turaga

Roles of Ant Pheromones During Decision Making

This paper models attractive and repellent trail pheromones used by Pharaoh's ants and how they affect foraging dynamics. I recognize that this model does not fall into the expected category of cell biology, but it is an interesting agent-based modeling paper that looks at the behavioral system and dynamics of an ant colony using the same type of methods and analysis that could be found in an agent-based model applied to the subject of cell biology.

Reference 

Robinson E, Ratnieks F, Holcombe M. An agent-based model to investigate the roles of attractive and repellent pheromones in ant decision making during foraging. "Journal of Theoretical Biology." 2008; 225:250-258. link

The Model 

The aim of the model is to investigate how foraging success is affected by the use of attractive and repellent pheromone. In the model, ants choose between two branches at a trail bifurfaction: one leading to food and one not leading to food. Success is quantified by the percentage of ants that return with food and the amount of time taken for the trip. In this model, ants are specified as the 'agent' and follow one of three states: nest ant, unfed foraging and, fed foraging ant. The agents can transition between these states depending on environment input and individual memory. The model environment consists of (1) nest (2) stem of trail (3) branch A of trail bifurcation (that leads to food) and (4) branch B of bifurcation (that has no food). The agent (ant) can sense pheromone and food in the system and react to the system by laying more pheromone. Agents do not directly affect other agents, only indirectly by affecting the pheromone environment. All agents/ants have and ID number and leave the nest in an ID order, at a rate determined by the traffic flow. As the ants/agents move into the stem of the trail, they sense the pheromones in the environment. The ants/agents in this model have a set of behavioral rules that they all follow. Each simulation experiment was ran for 200 time-steps which corresponds to 20 minutes.

The Interaction 

Simulations demonstrate that the use of attractive pheromones increase foraging success and reduces journey rate, as expected. The use of repellent pheromone also affects journey time and foraging success - its absence has the strongest affect in those cases when a random switch in branch attractiveness occurs (a phenomenon that emerged from the agent based model). The switches occur as a result of the regulation of positive pheromone by positive feedback. The attractive pheromone's positive feedback mechanism by which the Pharaoh's ant mediates its responses to food has potential disadvantages when used alone without repellent pheromone. In particular, it can cause random errors where the bulk of a foraging colony picks an unproductive branch when the location of food changes. The repellent pheromone allows the colony to escape ongoing suboptimal states caused by these positive feedback loops, providing the system with robustness to random fluctuations.

Reviewer 

Phillip Lee

Oscillations by the p53-Mdm2 feedback loop

This paper elucidates a simple mathematical model suggesting the oscillations in p53 and Mdm2 proteins as result of a stress signal which has significant role in the development of cancer and several other human malignancies.

Reference

Lev Bar-Or R, et al. Generation of oscillations by the p53-Mdm2 feedback loop: a theoretical and experimental study. Proc Natl Acad Sci U S A. 2000;97:11250 link

The Model

A p53-Mdm2 interaction model was developed and it was proved that certain features of the p53-Mdm2 interaction can generate oscillations in the levels of both proteins, in response to a sufficiently high stress signal. In fact a delay in p53-dependent induction of Mdm2 leads to an oscillatory behavior in the system. By tuning the length of this delay the authors were able to determine the period of oscillation. Also it was shown that for the delay to occur the strength of p53-Mdm2 interactions has to lie in the intermediate range.

The Interaction

The model accurately reproduced the kinetics characteristics observed in the experimental results. Oscillation can hence be viewed as an arrangement that allows repetitive repair efforts where a first pulse of p53 is delivered and the system then waits to see whether the damage has been fixed or not. If not then a second pulse is generated and so forth until the damage is restored. If the damage is extensive then the peaks are found to be very high which triggers apoptosis. This work shows the intelligent mechanism cell uses to determine its fate in response to various stress signals.

Reviewer

Sombeet Sahu

Information in Noisy Networks

This paper looks at the interference of noise within signaling pathways and discusses methods employed by the cell to mitigate the impact of noise on a signal.

Reference 

R. Cheong, A. Rhee, C. J. Wang, I. Nemenman, and A. Levchenko (2011) Information Transduction Capacity of Noisy Biochemical Signaling Networks, Science, 334, 354-358. link.

The Model 

The bush and tree models were both used to analyze signal transduction networks. In the bush model, the root node is directly connected to multiple sub nodes, forming a bush-like structure; the tree model passes all information through a single "trunk" to some intermediate node before splitting and distributing the signal. The tree model highlights a possible information bottleneck at the trunk which is absent in the bush model.

The Interaction 

Both models were used to estimate the amount of information (a measure of TNF concentration) encoded in two separate, but redundant, signaling pathways: the NK-κB and ATF-2 pathways. Each pathway individually encodes less than 1 bit of information, but the combination of the pathways yields a higher capacity. The bush and tree models predicted a theoretical capacity for the network, but only the prediction from the tree model was close to experimental results. As a result, the authors were able to predict something about these pathways, specifically that there exists some kind of upstream bottleneck. Their predictions match well with experimental data that suggests TNF receptor activation is a bottleneck in these networks.

Reviewer 

David Farrow

A method for zooming of nonlinear models of biochemical systems

This paper examines a mathematical approach to simplifying complex biochemical models while retaining their predictive ability and interpretability.

Reference 

Mikael Sunnåker, Gunnar Cedersund, and Mats Jirstrand (2011) A method for zooming of nonlinear models of biochemical systems, BMC Systems Biology 2011, 5:140 doi:10.1186/1752-0509-5-140n. link.

The Model 

Use of a lumping methodology to a secondary model which can be used for analysis and "retrieval of original states and parameters" without need for a new simulation. The reduced model can be considered as a different degree of zooming relative to the original model.

The Interaction 

Conservation relations in the original model were identified, state variables of the reduced model defined, then fraction parameters linking the reduced model to the original were derived. After rate of change of the state variables of the reduced model were derived, the variables were back translated to the fraction parameters allowing a comparison between the predictions of the original and reduced models.

The method was demonstrated on two models and the robustness of the reduced model tested: 1. an Enzyme Kinetics Model, 2. Glucose transport in budding yeast. The reduced model for the latter did not perform as well as hoped and equations had to be adjusted, but both models were successfully reduced.

Reviewer 

Jayse Embry

Integrated Genomic and Proteomic Analyses of a Systematically Perturbed Metabolic Network

This paper uses DNA microarrays and tandem mass-spectrometry to measure the effects of perturbations of cellular pathways.

Reference 

Trey Ideker, Vesteinn Thorsson, Jeffrey A. Ranish, Rowan Christmas, Jeremy Buhler, Jimmy K. Eng, Roger Bumgarner, David R. Goodlett, Ruedi Aebersold, Leroy Hood (2002) Integrated Genomic and Proteomic Analyses of a Systematically Perturbed Metabolic Network Science 292, 929-934 (2001);

DOI: 10.1126/science.292.5518.929 link.

The Model 

To demonstrate the usefulness of a systems approach in general, they look at the case of galactose utilization in yeast. They start with the known interaction edges and vertices, subject cells to different conditions to perturb each pathway component, and use the gene expression data to infer new pathways, which are then tested for in the lab. The process is then iterated, with new hypothetical pathways postulated, perturbed , etc.

The Interaction 

The galactose utilization network in Saccharomyces cerevisiae was studied as (perhaps?) the first large-scale comparison of mRNA and protein responses used to investigate cellular signaling pathways. This paper was part of Lee Hood's ongoing attempts (at the time) to take a systems approach to explain the responses of signaling networks using some known partial knowledge of regulatory and physical interactions in tandem with large scale assays taking advantage of fully sequenced genomes. The paper also looked at some refinements to the GAL model suggested by the integrative studies -- for instance, the analysis of their gene expression data suggested that Gal4p directly regulates genes in several of the galactose utilization processes through novel protein- DNA interactions.

Reviewer 

Rory Donovan

Computational Modeling of Cellular Signaling Processes

This paper outlines a technique for modeling cellular signaling, and uses MAPK activation in yeast to compare simulated and experimental data.

Reference 

Bastian R Angermann, Frederick Klauschen, Alex D Garcia, Thorsten Prustel, Fengkai Zhang, Ronald N Germain & Martin Meier-Schellersheim (2012) Computational modeling of cellular signaling processes embedded into dynamic spatial contexts, Nature Methods, Advance online publication, Jan 29, 2012 link.

The Model 

This model uses information for local geometry and molecular components of individual cellular elements to build a reaction network as opposed to mapping a global reaction network onto individual elements. Partial differential equations are used to model diffusion through spatial regions of the simulated cell. This technique is implemented in the Simmune modeling tool.

The Interaction 

The model was applied to the MAPK activation pathway in yeast and simulated concentrations of Fus3 were compared to a experimentally reported intensity profile and confocal microscopy images of yeast cells expressing fluorescent Fus3. When the classical MAPK signaling model was constructed in Simmune the results were not similar to experimental observations. The authors then constructed a more recently proposed MAPK signaling model and found that the simulated Fus3 concentrations were very similar to those observed in the fluorescence experiments. This indicates that use of the computational model can provide insight into the underlying biological system.

Reviewer 

AJ Sedgewick

Functions of MdmX

This paper explores the function of MdmX, a protein that regulates the tumor suppressor protein p53. The regulation of p53 by MdmX in response to DNA damage was investigated by generating and simulating the mathematical model of p53-Mdm2-MdmX network.

Reference 

Kim S, Aladjem MI, McFadden GB, Kohn KW (2010) Predicted Functions of MdmX in Fine-Tuning the Response of p53 to DNA Damage. PLoS ComputBiol 6(2) link.

The Model 

A mathematical model based on molecular interaction map (MIM) of simplified p53-Mdm2-MdmX network was generated to understand the function of MdmX. The MIM of p53-Mdm2-MdmX network was constructed and extracted from previously published MIMs. The model displays all interactions in association or dissociation, which can be translated into reaction form. Ordinary differential equations (ODEs) were written based on reactions obtained from the MIM of p53-Mdm2-MdmX network, and the model was simulated by solving ODEs. Constants for initial simulation were also obtained from previously published models.

The Interaction 

It is known that MdmX and Mdm2 regulate the tumor suppressor protein p53 in some way, but it is not clear how exactly they interact with p53. Simulated mathematical model suggested that MdmX may controls the dependency of p53 on Mdm2, and it positively or negatively regulates p53, which led to a switch-like behavior in early response to DNA damage. Study also suggested that MdmX inhibits oscillations of p53 level in late response to DNA damage.

Reviewer 

Sunguk Choi

Modeling drug perturbations

The paper builds a mathematic model to simulate the network in a cell to predict its response to certain events, more specifically, the reaction under combined drug treatment.

Reference

Sven Nelander, Weiqing Wang, et al., (2008) Models from experiments: combinatorial drug perturbations of cancer cells, Molecular Systems Biology, 4:216, link.

The model

A network model is built, with nodes representing some biological features that can be quantified, e.g. molecular concentration. The time evolution of this artifical system is modeled by differential equations and an interaction matrix. Optimization of the model focuses on prediction performance and model simplicity.

The interaction

Basically, the pure mathematical model needs at least some experimental data to train it. To build a network like this, knowledge from cell biology is also needed to initially establish relationships among elements in the model. The other way around, The model can help analyzing experimental data and can do prediction. In this paper, the model does a pretty good job on predicting the outcomes of drug pair treatments on a cancer cell. Further more, the model can hopefully be extended to predict multi-drug treatment, and may guide to discover new therapies.

Reviewer

Yutong Li

Response of C. elegans reproduction to chronic heat stress

This paper develops a macro-level model to describe how temperature stress affects reproduction in C. elegans. The results and model suggest that the behavior of complex biological systems may be determined by a small number of key components.

Reference

P.McMullen, E. Aprison, P. Winter, L. Amaral, R. Morimoto, I. Ruvinsky (2012) Macro-level Modeling of the Response of C. elegans Reproduction to Chronic Heat Stress, PLoS Computational Biology,8, 1-12., link.

The Model

A macro-level parsimonious model was developed that is sufficient to make quantitatively accurate predictions of the dynamics of reproduction under stress. The model uses detailed, time-resolved experimental data to predict reproductive dynamics of animals in a number of environmental and genetic backgrounds. This shows that a minimal model of a process can be sufficient for capturing dynamic systems.

The Interaction

The model is both simple and falsifiable. The authors conceptualized the reproductive system as a series of interconnected compartments. The process can be likened to a chemical reaction because the transitions between compartments can be modeled as a conversion of precursors to products. The assumptions made are that all gametes in the model are conserved and can be accounted for. Also, all transitions between states obey mass-action kinetics. This application can be used in other models that involve mass transfer.

Reviewer

Sheila Chandran

Cholesterol

This paper explores dynamical modeling of the cholesterol pathway. Cholesterol is an important molecule used by many pathways in the human body.

Reference

G., Kervizic, L.Corcos (2008) Dynamical modeling of the cholesterol regulatory pathway with Boolean networks, BMC Syst Biol. 99. [1].

The Model 

Boolean Analysis and Markov chains were used in modeling the cholesterol regulatory pathway. Boolean analysis was used for efficient computation of synchronous analysis and Markov chains were used for identifying the spurious cycles which result from synchronous analysis.

The Interaction 

The model accurately describes inhibitors and activators of the cholesterol pathway. In silico experiments were done to verify dynamical modeling of the cholesterol pathway.

Reviewer 

Barbara Sanchez-Neri

An Integrated Model of Epidermal Growth Factor Receptor Trafficking and Signal Transduction

This paper uses a mathematical model to show the signaling and trafficking activities of EGFR. The interplay of these two processes is critical to proper signal transduction by EGFR, and understanding them more fully may lead to a better understanding of where errors occur.

Reference

H. Resat, J. A. Ewald, D. A. Dixon, and H. S. Wiley (2003) An Integrated Model of Epidermal Growth Factor Receptor Trafficking and Signal Transduction. Biophysical Journal, 85(12): 730-743. [2]

The Model 

The authors used a novel probability – weighted dynamic Monte Carlo simulation method to predict the signaling and trafficking processes. Reaction rates were taken from previous experimental work. The novel simulation method served to effectively sample reactions that occur on vastly different time scales, accurately predicting outcomes that had posed previous computational problems.

The Interaction 

Previous experimental data was critical to building an effective model, through providing necessary reaction rates. Further experimental data was then used to validate the model. The model accurately predicted the location and concentration of activated receptors, during stimulation by different types and concentrations of ligands. The model also provided insight into how different ligands can affect downstream signaling events, as had been observed in previous experiments.

Reviewer 

Bobby Sheehan

Pou5f1-dependent Transcriptional Networks

This paper investigates the role of the transcription factor Pou5f1 in the timing of zebrafish development.

Reference 

D. Onichtchouk, F. Geier, B. Polok, D. Messerschmidt, R. Mossner, B. Wendik, S. Song, V. Taylor, J. Timmer, and W. Driever. (2010) Zebrafish Pou5f1-dependent transcriptional networks in temporal control of early development, Molecular Systems Biology, 6:354. link.

The Model 

The authors developed a system of ordinary differential equations to model the transcription rate of pou5f1 and its downstream targets in both wild-type and MZ (mutants lacking both maternal and zygotic Pou5f1) embryos. The model was used to predict the expression of these targets in M mutant embryos, which are characterized by a non-functional maternal Pou5f1.

The Interaction 

The authors utilized time-series microarray experiments to identify the downstream targets of Pou5f1. The targets were further validated through whole-mount in situ hybridization and real-time PCR. This helped to elucidate the wiring of the gene regulatory network that was modeled. Furthermore, model parameters were estimated based on the temporal expression profiles revealed by these experiments.

Reviewer 

Kristina Buschur

Probabilistic model for Protein-Protein Interaction Prediction

Reference 

Daniel R Rhodes et al, Probabilistic model of the human protein-protein interaction network, Nature Biotechnology volume 23, number 8, August 2005.[3]

The Model 

The author’s initially discusses how protein-protein interaction can be predicted from different biological datasets. They describe four different biological datasets in this context each providing specific information for predicting protein-protein interactions. Model organism interactions suggest interactions among orthologous human proteins while similar gene expression profiles helps in identifying the interacting protein products. Similarly protein domain pairs enriched among known human protein-protein interactions may suggest novel interactions and shared functional annotations from Gene Ontology may suggest physical interactions. Firstly, the likelihood ratios were calculated for groups of predictions for each data type. Then data types that contain redundant information were identified and likelihood ratios were calculated for predictions with both data types. Then for each dataset the maximum likelihood ratios were identified. The goal was to capture the protein interaction information from different independent data sources by using a probability based Naïve Bayes mode which multiplicatively combines the dataset specific likelihood ratios.

The Interaction 

Calculating likelihood ratios for each dataset and then identifying the maximum likelihood ratios will help in capturing the protein interaction identifying information from each individual dataset. However, this results in predicting a low number of very high confidence predictions or a very high number of very low confidence predictions. In order to overcome this, protein prediction information from different datasets was combined using a naive bayes probability model which captures information from all the datasets by obtaining composite likelihood ratios. The multiplicative nature of the model does not overestimate the likelihood of interaction in the training set. Implementing this model on all the available human protein pairs led to the identification of a large number of protein-protein interactions with good confidence values. The authors then validated the predicted interactions using coimmunoprecipitation assays. Ex: confirmation of interactions between BUB3 and ZNF207 and between RSU1 and LIMS1 proteins. The probability based naïve bayes model and its implementation in combination with different biological datasets used in this paper provides a very good example of how new computational biology approaches can be developed by combining different probability models with biological information and it also describes how effective these methods can be in working on complex problems like protein-protein prediction at the system’s level.

Reviewer 

Sirisha Penumetcha

Predicting HIV-1 drug resistance

This paper explores the understanding of HIV-1 drug resistance genetic properties and creating a model to lead the discovery of new antiretroviral drugs and to enhance the potency of pre-existing drugs.

Reference 

S. Rhee, J. Taylor, G. Wadhera, A. Ben-Hur, D. L. Brutla, and R. W. Shafer (2006) Genotypic predictors of human immunodeficiency virus type 1 drug resistance, PNAS, 103, 17355-17360. link.

The Model 

They used various computational methods involving decision trees, neural networks, support vector regression, least-squares and least angle regression to model the susceptibility of HIV-1 protease and reverse transcriptase mutations to various antiretroviral drugs. The model would be able to predict phenotypic characteristics of polymorphic proteins and unveil genetic mechanisms of HIV-1 antiretroviral cross-resistance.

The Interaction 

Each of the machine learning methods contained four mutation sets which were used as inputs. These four mutation sets each played an important role in discovering drug resistance properties. The first one contained a set of all mutations present in at least 2 sequences, the second had an expert panel mutation set, the third had a set of treatment-selected mutations, and the fourth contained a control set of the most common mutations. When the tests ran, the non-polymorphic treatment-selected mutations performed the best at predictions. The other methods also verified, quantitatively, the effect of mutations on drug susceptibility. In addition, it also confirmed previously reported drug resistant genes.

Reviewer 

Tim Song

Rapid Evolution in Predator Prey Dynamics

This paper explores the role of rapid evolution in predator-prey systems.In order to better understand ecological and evolutionary dynamics in predator-prey systems, models were created for and tested with a rotifer-algal laboratory microcosm.

Reference

Rapid evolution drives ecological dynamics in a predator-prey system. Stephen P. Ellner, Gregor F. Fussmann, Nelson G. Hairston, Jr, Laura E. Jones, and Takehito Yoshida. Nature. 424.6946 (July 17, 2003) p303. Word Count: 3584.Author(s): Takehito Yoshida [1]; Laura E. Jones [1]; Stephen P. Ellner [1]; Gregor F. Fussmann [1, 2]; Nelson G. Hairston, Jr (corresponding author) [1] Ecological and evolutionary dynamics can occur on similar

The Model

Multiple models were made. The basic model was a simple differential equation model that predicted the transition between qualitatively different population dynamics as a function of nutrient cycles. The alternative models in turn took account of variables that might explain the basic models failure. These variable sincluded changes in algal carbon, nitrogen ratio vis-a-vis nutrietn availability, and rapid prey evolution resulting from an evolutionary tradeoff between algal competitive ability and defence against rotifer predation. The prediction of the rapid prey evolution adjusted model is that after the crash in rotifer population, it cannot recover until the algae acheives high population and selection favors clones with high-food value, but which are also better competitors.

The Interaction

With a single clone system, where genetic variance is absent, the base model accurately predicts predator-prey cycle lengths. However, with multiple clone models where variation is present, the base model fails, producing oscillations almost exactly out of phase. This led to the rapid-evolution model, which accurately predicted the cycles. This leads to the conclusion that an accurate model of populations and food webs needs to consider rapid prey evolution, in addition to evological dynamics.

Reviewer

Max Tomlinson

Environmental and genetic pertubations reveal different networks of metabolic regulation

This paper takes a look at the role of genetics and the environment in pertubation in the metabolic pathways of Drosophila melanogaster.

Reference 

Anthony J Greenberg, Sean R Hackett, Lawrence G Harshman & Andrew G Clark (2011) Environmental and genetic perturbations reveal different networks of metabolic regulation, Molecular Systems Biology 2011,link.

The Model 

The model used was a multivariate Bayesian model. The covarianence among different metabolic parameters was estimated and this information was used to build probabilistic models of network topology. The variance among enzyme activities was measured using perturbations, taking advantage of natural genetic variablity and also environmental sources of variation.

The Interaction 

Activites of enzymes and metabolite levels were directly measured through assays. These were used to develop the Bayesian model described, which in turn was used to build other, more precise probabilty models of the network topology.These models were used to estimate the effect of various environmental factors on the metabolic pathways and the role of genetic variation.

Reviewer 

Vincent Yu

Representation of morphed odors in the rat olfactory bulb

Reference

'Odor Representations in the Rat Olfactory Bulb Change Smoothly with Morphing Stimuli.' Adil G. Khan, Mukund Thattai, and Upinder S. Bhalla National Centre for Biological Sciences, Tata Institute of Fundamental Research, Bellary Road, Bangalore 560065, India

Model

Olfactory coding model using Attractor networks in rats.

Interaction

This paper tries to show the odor representation in the Mitral/Tufted cells of the olfactory bulb for different odors and mixture of odors. Using this model, the representation of odor identity, intensity and the summation of odor in mixtures are stated. Different odors can evoke unique patterns which are marked by excitation or inhibition. By recording the Olfactory Bulb neurons it’s seen that most OB neurons respond to odor mixtures as a weighted sum of individual odor responses rather than as distinct attractor states in double tracheotomized rats to remove respiration dependence of olfaction. The basic nature of an attractor network is to have abrubt transitions in representations when odors are morphed. But the transition in OB is smooth and not abrupt. Hence it is suggested that other brain areas be examined to strengthen the case of attractor dynamics in olfaction.

Reviewer

Sripradha Viswanathan

RANK-RANKL-OPG Feedback in Bone Remodeling

This paper explores the role of competitive binding of OPG and RANK to RANKL and it's subsequent effects on bone remodeling.

Reference 

Pivonka, P., J. Zimak, D.W. Smith, B.S. Gardiner, C.R. Dunstan, N.A. Sims, T.J. Martin, G.R. Mundy, Theoretical investigation of the role of the RANK-RANKL-OPG system in bone remodeling, Journal of Theoretical Biology 262: 306:316, 2010.link.

The Model 

A differential equation model of bone remodeling was developed and analyzed. The model uses a system of three ordinary differential equations corresponding to osteoclasts, osteoblasts and an osteoblast progenitor cells. A fourth differential equation is used to model total bone volume which is used for model comparison to experimental data. Parameters of the models are tuned by fitting system response with experimental data.

The Interaction 

The model reproduces the experimentally observed behaviors of bone remodeling as shown in tables two and three of the paper. Notably the model reproduces the changes in bone volume corresponding to changes in OPG, RANK and RANKL suggesting that the feed back is accurately modeled such that OPG blocks the activation of osteoclasts by the binding of RANK with RANKL leading to bone resorption. As a result, increased levels of OPG lead to increases in bone volume.

Reviewer 

Devin Sullivan