Q-bio Friday 08/12/11

Reka Albert, Pennsylvania State University, Discrete dynamic modeling of signal transduction networks

• Discrete dynamic modeling: boolean, arbitrary choice for times — sync update, random update, explicit delay update.
• Drought signaling in plants to change stomata size (via abscisic acid signaling)
• variability in timing and initial conditions doesn’t affect model.  most components can be removed without effect.
• topologically redundant paths (some paths require dual input to propagate signal)
• specific time scale ratios allow oscillations.  (seriously discrete fixed time steps not a good way to explore oscillations).
• Model 2: extra cyto-toxic T cell survival.  3 input states.
• Asynchronous boolean model.  ID minimum conditions (least activity of input) for cells to escape programmed death.
• State transition graph what states lead to disease condition vs. normal condition.
• Take away points: given the graph / hairball, which nodes are most important and which most dispensable.

David Klinke , West Virginia University, Quantifying cross-talk among Interferon-γ, Interleukin-12 and Tumor Necrosis Factor signaling pathways within a TH1 cell model: A model-based inference approach

• Making inference with limited knowledge of quantitative parameters
• Naive CD4+T helpers interact with “educated dendritic cells” interlukin/cytokines determine mature T-effector cell.
• Il-12 (link innate immunity with adaptive immunity) via JAK/STAT promotes TH1 polarization.
• plasticity / reversible path between effector and naive.
• Quantitative Data set: measure cell density, cell signal intensity (try to get copy number from intensity) expose to IL12 and observe.
• Empirical Bayesian Approach to model-based inference:  account for uncertainty in parameters.  (Klinke BMC Bioinformatics 2009).  Integrate using MCMC.
• Look at cloud in 32D parameter space which supports the data: some combinations of parameters are tightly correlated — free to choose one but that restricts others, some are narrowly constrained, and some are unconstrained.
• Can make probablistic statements that 95% confidence flow through this path is more important than flow through that path.
• What is the class of models you are inferring?  ODEs?
• Response to Il12: Stat1 turns on in a pulse, Stat4 turns on and stays on.
• Clearer illustration of what does the experimental setup look like and what data are you working with?  How does it relate to your predictions.

Mohammad Fallahi-Sichani, University of Michigan, The dynamics of TNF signaling control tuberculosis granuloma formation

• Tuberculous (TB) background:
• 5-10% of exposed people to TB are infected
• 90% latent TB, can become active later in life
• immune cell collections form granula — infected macrophages in center surrounded by latent macrophages and T cells.  Stabilized latent state: controlled but not cleared.
• Goal: how does TNF activity affect outcome of infection?
• TNF induced NFkB — related to bound receptors.
• TNF induced apotopsis – related to internalized bound receptors
• low internalization rate: clearance with very high TNF – excessive inflammation, very high macrophage levels.
• too high internalization: no clearance.
• experimental observation to explore further: stability of mRNA influences timing and order of gene induction.
• model wants lifetime TNF < CHEM < ACT (estimate 30 min to 1 hr)
• make some active predictions about mRNA stability
• Takehome: multiscale platform for immune response to TB.  (good integrator of existing knowledge of system?)
• experimental system for granuales: human data, monkey model, genetically modified mice to make mouse model of granules (not part of natural mouse response).

Huilei Xu, Mount Sinai School of Medicine, Functional Atlas of Mouse Embryonic Stem Cells Pluripotency and Early Differentiation

• ESCAPE: Database of embryonic stem cell
• learning boolean transition functions to match whole genome expression data.
• predict results of knockdown of individual genes: pretty good concordance
• predict combinitoral knockdown.  Tests of combinatorial knockdown?
• part 2: infer upstream regulators from differnetially expressed genes using differences in mRNA levels (chip) PWM analysis and ChiP-X enrichment.
• cluster 44 different cell lines (transcription factor pseudo activity).

Jakub Otwinowski, Emory University, Speeding up evolutionary search by small fitness fluctuations

• time dependent fitness landscape – fluctuating triangle wave.
• small peaks drive model around — enhances diffusive motion: poster 108.
• Q: plot correlations? is this2D  diffusive motion or something else?  Levy flights?

• ultrasensitive switches may hide in small networks and have deterimental behaviors.
• correlating gene expression reduces bursts.   Poster 81.  Also cool

Edward Stites, Translational Genomics Research Institute, Promiscuity quantitatively and qualitatively impacts early growth factor receptor signaling

•  Ras mutations strongly correlated with cancer.
• graded activation of pERK associated with increased proliferation
• transient models of Ras don’t provide insights into steady state oncogenic Ras signaling.
• little fitting required to build new steady state model.
• Most sensitive parameters from sensitivity analysis are indeed corresponding to common mutants.
• GAP insensitive RAS mutants and Fast cycling mutants.  Only former frequently found in cancer.  Effect consistent with model predictions.  test experimentally.
• Make new predictions about combined mutations.  predict less than additive, greater than additive, etc.
• Reducing GTPase activity previously considered unlikely to affect system, predict larger effects.  ?
• Predict cellular properties from molecular and biochemical properties (mutations etc)

Raymond Cheong, Johns Hopkins University, Advantages and limitations of network-based information processing in biological signaling systems.  Retitled: Information transduction capacity of noisy biochemical signaling networks.

• how much information can a noisy pathway transmit about signal strength?
• TNF -> NF-kB (signals: external concentration to nuclear concentration).
• P(R,S) = P(R|S)P(S).  Measure P(R|S).  Maximize mutual information over possible P(S) is “channel capacity”.
• I(NF-kB;TNF) < .92 +/- .01 bits (less than 1 bit, can’t always tell whether or not signal was present).
• Many other canonical systems close to 1 bit.   Torso signaling in drosophila 1.61 bits.
•  Ways to mitigate noise (presumably these are already in action though?):
• multiple pathways: (branching common motif).
• Bush network, branching at level of signal, information increase is unbounded.  Tree network is bounded by the capacity of trunk.  This is Trivial — adding extra links adds capacity.
• claim branched signaling from Receptor to NFKB and A2.
• short time removal of negative feedback increases information transmission,
• Time integration to reduce molecular noise / increase information.
• time of onset varies considerably, so time integration not especially helpful
• Multiple cells work together – arbitrary large gains.
• How much of this variation is due to the fact that population is heterogeneous.
• Presentation stimulates substantial discussion.

Christopher Zopf, Massachusetts Institute of Technology, Single promoters as regulatory network motifs. (Narendra Maheshri lab)

• chormatin adds complexity to single promoter kinetics — slow change to primed promoter.   If TF acts on primed promoter -> Tx.  has topology of coherent feedforward loop.
• Properties: Delay. Filtering (won’t reply to short bursts).  Memory — after being primed, will respond.  Q: Filtering only of signals that occur less frequently than the loss rate of chromatin state.
• Goal: probe state of promoter.  Method – infer from calculation of transcription rate.
• compare promoters with high affinity site exposed vs. low affinity site exposed.  see if difference in temporal response.
• single cell traces highly variable.  roughly shifted by fixed delay.  high response looks more noisy.  20 min vs. 40 min response.
• test for delays in cis or trans comparing CFP vs YFP.
• “less active Pho4 still in nucleus”.  Most of delay is occuring in trans — two separate loci have the same delay.
• localization is repeatedable on successive treatment, fraction of cells that respond each time is increasing.
• is this effect in cis or trans.  Memory is occuring at the level of promoter.