Speaker and Talk Information
Benjamin Amor: Exploring Allostery with Complex Network Theory
In this talk I will show how we can apply network-theoretic methods to uncover information about how signals propagate through proteins. Allostery is a classic biochemical phenomenon in which a perturbation at an `allosteric site’ on a protein has a functional effect at the distant active-site. Classical models of allostery provide phenomenological descriptions of this behaviour, but are not able to predict new allosteric sites or uncover the specific pathways connecting the two sites at an atomistic level. By representing an individual protein as a network of atoms and studying the behaviour of random-walks on this network, we have shown that it is possible to correctly predict the location of known allosteric sites and uncover information about the pathways linking them to other regions of the protein. This is of great interest in the field of drug discovery, since drugs which target allosteric sites offer a number of advantages over traditional active-site targeted drugs. I will give a general introduction to some of the network-theoretic techniques that have been developed in our group and present a number of interesting proteins to which we have successfully applied these methods.
Elias Bamis: Combining Gravity and Intervening Opportunities in Models of Spatial Interactions
Human spatial interactions such as commuting, migration etc. have recently been approached with the hope of uncovering universal mobility patterns. The standard model in transportation analysis and spatial econometrics is the gravity model, estimated most often in a generalized linear regression setting. Simini et al. (Nature, 2012) motivated by the concept of intervening opportunities proposed an essentially parameter-free model, the radiation model. However in the presence of observed data the model does not usually outperform gravity models. In this talk we propose an Absorbing Markov Chain model for spatial interactions. With assumptions similar to the radiation model’s, it can take a parameter-free form, while on the other hand it offers the modelling flexibility and performance of gravity models. We use Poisson Regression to fit and compare the models on USA Commuting-to-work data.
Adam Gosztolai: Integrating Signals of Nitrogen and Carbon Status in E. coli
Ammonium assimilation in Escherichia coli is a model system of environmental sensing and adaptation in bacteria. At the heart of it lies a signal transduction cascade whose role is to facilitate a balanced uptake of carbon and nitrogen. It achieves this by sensing signals of nitrogen and carbon state and controlling the activity of the key enzyme glutamine synthetase (GS) and the transcription of nitrogen-regulatory genes. We use theory and, for the first time, in vivo data about the post-translational modification state of proteins to investigate the response of E. coli to the run-out followed by a sudden upshift of external ammonium. Since long-term regulation is of interest, the variation in total protein levels is considered; a feature that has been ignored in previous studies. We validate our model by predicting the response to gene knockouts. We argue that the response of the system to nitrogen perturbation is determined by the subtle interplay between signalling, transport and gene regulation.
Juan Kuntz: Bounding Steady State Averages of Stochastic Reaction Networks via Semidefinite Programming
Stochastic reaction networks (SRNs) describe situations in which the populations of a finite number of species evolve through finitely many different predefined interactions. These continuous time Markov chains are popularly used as models in biological sciences such as systems biology, epidemiology, systems biology, and ecology. However, the development of analysis tools of these networks still remains in its infancy. The majority of the currently available tools involve either simulation using a Monte-Carlo scheme or the approximation of the SRN by a simpler process. In this talk, I discuss a computational tool that applies to networks with mass action kinetics and that does not involve either simulations or approximations. The method produces hard upper and lower bounds on each of the moments of any invariant measure of the network. We are interested in the invariant measures of a SRN because, under some mild stability conditions, they determine the long-term behaviour of the SRN. If the network only has a single invariant measure, then the upper and lower bounds are often close to each other. In this case, we can use the bounds to compute further bounds on other quantities of interest such as the average long-term molecule counts and reaction rates, their coefficients of variation, and the correlation coefficients between them. Our approach can also be extended to compute bounds on probabilities of “nice-enough” events, for example the event that the number of molecules of a given species is in some interval.
The method combines the definition of an invariant measure and some results from the theory of moment problems to construct semidefinite programs (SDPs) whose solutions are the desired bounds. SDPs are tractable convex optimisation problems that can be efficiently solved using high-quality packages freely available online. The method can be automatised so that all the input necessary from the user is specifying the reaction constants and stoichiometry matrix of the SRN, and the state space the user is interested in. This said, before applying our approach the “mild stability conditions” mentioned above must be verified and this is often not a simple task. However, some progress in developing computational techniques to establish this type condition holds has already been made elsewhere and we expect more will follow in the near future.
Refreshments will be provided after the talks.