Regional Meeting on Mathematics, Computation and Biology

Hewlett Packard Laboratories

24 June 2008

---

Conference Homepage

Keynote:

• Crisantha Fernando "Is there a neurome?"

Keynote

 

"Is there a Neurome?"

Chrisantha Fernando
MRC National Institute for Medical Research

Session 1: Organisation and Learning in Social Systems

 

1 The importance of individual differences in the interaction between animals  

John McNamara
University of Bristol

When the fitness of an organism depends on the behavioural strategies of other population members evolutionary endpoints can be characterised using game theory. However, uses of game theory in behavioural ecology often ignore differences between individuals. I contend that such differences are not innocuous noise, but can fundamentally change the nature of a game. To demonstrate this I present an example in which differences in personality promote the need to be socially sensitive. Social sensitivity changes the selection pressure acting on individuals who are observed, and can lead to the maintenance of differences. In other words the existence of differences can change the fitness landscape so that differences are maintained.

2 The Limitations of the Baldwin Effect (Presentation)

Marios Richards, Joanna Bryson
University of Bath

The Baldwin Effect was originally proposed [Baldwin 1896] as a mechanism by which general learning ability can be selected for in favour of 'directly' genetically coded traits, but is currently more often used to describe the assimilation of learnt behaviours into genetically coded traits within a genepool. Hinton and Nowlan [1987,"How learning can guide evolution"] developed a simple quantitative model showing that the Baldwin Effect can be viewed as 'accelerating evolution' for agents engaging in individual learning in the context of a needle-in-a-haystack fitness distribution. We have replicated their model and engaged in some preliminary investigation of fitness distributions in addition to a NIAH delta function with a view to clarifying the space of evolutionary systems in which the Baldwin Effect may be significant. Understanding the limitations of the Baldwin Effect is a necessary precondition for extending its application to systems involving both individual and social learning and the subsequent meme-gene co-evolutionary dynamics arising. In the short term this work is intended to gain empirical salience by application to observations of primate social learning and open questions as to whether periods of perceived bias (learning from mother/peer-group) might represent adaptations to the limited fidelity of transmission or the advantages of being an early adopter of beneficial innovations rather than the direct Machiavellian advantages of constructing and maintaining social coalitions through partner-preferential social learning. In the longer term this work is intended to create a basis from which to begin characterising the selective pressures for and against communicative traits in some model primate society (e.g. Cebus capucinus).  

3 Collective Protection by Individuals Expediting or Delaying Resource Transmission 

Ana Sendova-Franks, Benjamin Wulf, Nigel Franks 
UWE and University of Bristol

Resource distribution is a fundamental aspect of social organisation, but it poses a dilemma. How to facilitate the spread of useful resources but restrict harmful substances? This dilemma reaches a zenith in famine relief. Survival depends on distributing food fast but that increases vulnerability to poisons. Ant colonies are a prime model system for addressing these questions. We studied individual foraging behaviour in Temnothorax albipennis colonies where benign or noxious food was introduced for 30 min after 72 h of starvation. Each of seven colonies underwent three treatments: (a) control –regular feeding followed by fresh honey solution; (b) benign - starvation followed by fresh honey solution (c) noxious - starvation followed by fresh honey solution with 40% alcohol. We found that the amount of time foragers spent self-grooming significantly decreased when the food was benign compared to the control. By contrast, when the food was noxious self-grooming time increased significantly compared to the control. Furthermore, while in the control, 33% of foragers returned to the nest and donated food to their nestmantes, following starvation, 89% did so when the food was benign and 54% when the food was noxious. We found no significant difference in the total time foragers spent drinking from the benign or noxious food source after starvation, but each was significantly longer than total drinking time in the control. Our results have important implications for the speed of famine relief and the networks of resource distribution within these decentralised biological social systems.

4 BitTorrent and Vampire Bats (Presentation)

Miranda Mowbray
Hewlett-Packard

BitTorrent is a popular peer-to-peer tool for file distribution which makes use of bandwidth-sharing by the downloading peers. I will talk about sharing behaviours by some animals - including vampire bats - and ideas for improvements to the BitTorrent client inspired by these behaviours.

5 A model for infection spread on graphs  (Presentation)

AJ Ganesh (joint work with Moez Draief)
University of Bristol

We consider a model in which agents perform independent random walks on a graph. An infected individual can infect a susceptible one only if they are at the same site, and the infection probability depends on the total time that they spend at the same site. This model motivates us to consider the coincidence time of independent random walks on a graph. We obtain results for the coincidence time on general graphs. When these results are specialised to scale-free graphs, they demonstrate a threshold phenomenon at an exponent of 3, corresponding to the transition between finite and infinite variance of the node degree. They show that epidemics spread much more quickly on networks having infinite variance of the degree distribution.  

Session 2: Information Processing and Decision Making in the Mammalian Brain

 

1 Reinforcement learning with indefinite state signals (Presentation)

David Leslie (Joint work with Tobias Larsen, Rafal Bogacz and Sean Collins)
University of Bristol

In a learning problem, one often assumes that the learner has perfect information about the state of the world, and in this case it is easy to learn the value of the different states. However when the learner only knows a probability distribution over states of the world it is more difficult to learn the value of each state, since it is not clear to which state an observed reward should be credited. The simplest models, which assign the reward based solely on the prior distribution of the state, can easily be seen to learn incorrect values. A more sophisticated approach that decides on allocation after the reward has been observed is shown to be more successful; in fact it can be shown to be a version of an online EM algorithm. The algorithm generally works well, but can get stuck in local minima, especially when the reward distributions change during the course of the learning episode.

2 Optimal integration of biologically plausible decision making and reinforcement learning

Tobias Larsen  (Work in collaboration with David Leslie, Sean Collins and Rafal Bogacz)
University of Bristol

When attempting to combine decision making and reinforcement learning models there are several sources of uncertainty to deal with, uncertainty in the stimulus, uncertainty in the rewards obtained and uncertainty in which action to take for a given stimulus. Recent work (presented by David Leslie) shows that it is possible to learn the correct reward values in a classical conditioning experiment in spite of uncertainties in both reward and stimulus. In this work we look at how to best use the learned reward estimates to improve the performance in an operand conditioning task, and we develop a biologically realistic model combining decision making and reinforcement learning.

3 Mammalian decision making

Pete Trimmer
University of Bristol

Empirical findings suggest that the mammalian brain has two decision-making systems which act at different speeds. We represent the faster system using signal-detection theory. We represent the slower (but more accurate) cortical system as the integration of sensory evidence over time until a certain level of confidence is reached. We then consider how two such systems should be combined optimally for a range of information-linkage mechanisms. We conclude with some performance predictions which will hold if our representation is realistic.

4 The study of the basal ganglia as a dynamical system and its relation with Parkinson's disease

Alejo Nevado Holgado, Rafal Bogacz
University of Bristol  

The basal ganglia are a group of subcortical nuclei interconnected with the cerebral cortex, thalamus and brainstem. It has been proposed that the basal ganglia are involved in action selection and they resolve the competition between cortical regions vying for behavioural expression. Dysfunctions of basal ganglia results in different impairments including Parkinson's disease which is characterized by difficulties in executing voluntary movements and tremor at rest. Several mathematical models of Parkinson's disease have been proposed but they include many free parameters difficult to estimate from experimental data. Here we present a model firmly supported by experimental data and constrained by proposed basal ganglia function. We use dynamical systems analysis techniques to understand the behavior of the model and clarify some standing questions in Parkinson's research.

5 Relevancy markers: a context free toolkit for task resolution

Martin Groen, Jan Noyes
University of Bristol

When humans operate in their environment to realise their objectives, they often need to consult external information sources (e.g., pictorial or textual information about foraging patches, advice from someone knowledgeable, and so on). However, it is well documented that people are poor at finding relevant information. Moreover, the increase in available information compounds this problem by extending the sheer amount of plausible possibilities that needs to be considered for realising the task at hand.

It would be useful, therefore, if search support could be provided. However, providing support implies an understanding of the specific context in which a task and its setting is embedded. Determination of context, however, is one of the unsolved philosophical challenges. In our work, we try to avoid the support paradox by not focussing on the topical matter of what humans are doing in their tasks, but by concentrating on how we use the available information sources when conducting a task. To do this, we examine which structural elements in an information stream, called relevancy markers, humans use to orient on the task-relevancy of the consulted information source.

We employ a two-step approach. In the first step we observe how successful problem solvers process an information source by measuring their eye movements. In the second step, the eye movements of the successful problem solvers are used to inform the manipulation of the information source displays, highlighting the relevant information bit with the highest discriminability. And then the effect of this manipulation on task performance is measured. So far, we have found positive evidence of the suggested role of relevancy markers across different media (transcripts of dialogues, recordings of dialogues and diagrams of mechanical devices), languages (Mandarin-Chinese, English and Dutch) and task domains. 

Session 3: Evolutionary information dynamics in genes and proteins

 

1 How to model synthetic gene regulatory networks

A. Polynikis, M. di Bernardo, S.J. Hogan
University of Bristol

A fundamental step in synthetic and systems biology is to derive appropriate mathematical models of the systems of interest for analysis and design purposes. For example, to synthesize gene regulatory networks, an important stage is the derivation of their mathematical model in order to carry out in silico investigations of their dynamics, investigate parameter variations and robustness issues. In this presentation, we compare different mathematical modelling approaches and discuss their predicted dynamics for a representative gene regulatory network.  

2 Detecting distant homologies among protein sequences by comparison of profile hidden Markov models (Presentation)

Martin Madera
University of Bristol

Proteins descended from a common evolutionary ancestor (homologs) adopt similar 3D structures and typically perform similar or related biological functions. Determining whether two proteins are homologous from their amino-acid sequences alone remains one of the most important tasks in computational biology. State-of-the-art methods for answering this question construct profile hidden Markov models of known protein families and score unknown protein sequences against these models. Taking this a step further, we can score two models against each other. This approach is more powerful and can detect more distant relationships. In the talk I will briefly describe my method for doing this, Profile Comparer (PRC).

3 The evolution of the protein repertoire

J Gough
University of Bristol

Most proteins have been formed by gene duplication, recombination, and divergence. Proteins of known structure can be matched to about 50% of genome sequences, and these data provide a quantitative description and can suggest hypotheses about the origins of these processes.

4 Investigating primitive nervous system transitions in models of simple organisms

Ben Jones
Birmingham University

Symmetry breaking is the major evolutionary transition that saw an alteration in primitive body-plan morphology. The most basic of these body symmetries is radial since it is evidently conceded that hydrozoan-like organisms led to more advanced bilaterally symmetric descendants. Intertwined with this transition however is a change in nervous system morphology. Both are tightly coupled meaning that behaviour is the result of the nervous system and organism body-plan coexisting together, in a rich sensory-motor configuration. Behavioural traits are said to be embodied. In my ongoing research, I am developing computational models of primitive-like organisms, with a view to establishing how evolution 'shapes' particular body-plan nervous system couplings. Firstly, in a model of a radially symmetric organism, it is shown how a minimisation of energy directly contributes to the network's architectural features. Secondly, in a model of a bilaterally symmetric organism, it is shown how a neural network can configure itself (via artificial evolution), so that the coupling is functionally bilateral. Thirdly, the ongoing part of this research is to amalgamate both nodels so that the above transition can be more fully realised. The core hypothesis is that the transition is precluded by a need to minimise energy. The hope is to inform both computer scientists (from an information perspective) and biologists (in terms of shedding light on the transition).

---

page author:
last updated:  11 Jul 2008