Honours Projects 2008

Projects supervised by Albert Zomaya

A Biomimetic View of Large-scale Sensor Network Systems
The massive initial deployment of sensor nodes provides the basics necessary for the organization of a primitive community. The sensors then mimic births and deaths in a biological system as they are activated (born) in waves, or generations, each generation adding its learned collective experience to the genetic base and passing the enhanced genetic information along to the next generation. The new genetic material will be local in scope allowing regional properties in the evolutionary process. Thus, sensors belonging to the new generation and situated in different locales of the community will be “born” with differing genetic materials. This will allow the sensor net to perform efficiently in a non-uniform environment. The regimen of updating the genetic material from one generation to the next ensures that, just as a biological system, the sensor network evolves and, in the process, may change its strategy, priorities, and methods. Evolution is game-theoretic and is guided by the maximization of a global objective function, based on local data only. Learning is key to ensuring that individual sensors, or groups of sensors specialize in a way that furthers the interests of the sensor network community.

At one level of abstraction, collective intelligence can be defined as the ability of the sensor network community, given a specific current network state, to effect a transition to a next state that better serves the overall community goal. The localized nature of communication in sensor networks dictates a decentralized approach towards the development (and evolution) of collective intelligence. In a decentralized approach, neighboring sensors evolve a local collective intelligence. The composition of local collective intelligence is what gives rise to a communal collective intelligence. One fundamental problem is how to guarantee the correctness of communal collective intelligence, up to community goals, given the relatively autonomous nature of local collective intelligence evolution? To this end, one scenario is that an arbitrary group of neighboring sensors would initiate a locale, and negotiate, as a single entity and as warranted, local goals for their locale for x amount of time. These local goals would be, of course, a function of the overall community goal. During the negotiated time, the sensors effect state transitions that ‘better’ serve their negotiated locale goal(s), triggered by input stimuli from both inside and outside the locale.

Distributed Coalition Planning and Decision Making
Our research in this particular area proceeds along two distinct directions. First, we are interested in one specific aspect of federating resources namely the establishment of coalitions both in a game-theoretic sense and from a generational-learning and service-centric perspective. We view coalitions as being avenues for maximizing a given (often global) objective function subject to (mostly local) constraints. Our work focuses mostly on wireless sensor networks and coalitions of networks subject to functional mobility as opposed to physical mobility. Indeed, in sensor networks populated by fixed sensors, one can define strategies for functional migration very much akin to physical mobility, except that it is not visible to the adversary. In a military environment this translates into low detection probability.

Scheduling in Distributed Systems and Multi-Server HPC Environments
Distributed systems provide high performance capabilities to support a wide range of applications. These applications normally have different, and sometimes conflicting, requirements. This will necessitate the development of more flexible scheduling techniques. Another factor which is detrimental to the performance of such grids is the dynamic nature of such combination of heterogeneous resources that are, for most of the time, located in disparate locations. In addition, the availability of resources (e.g. computational, storage, etc) for some of the time does not mean that such resources will be available all the time. Such conditions will add more complexity to the design of grid schedulers. This also suggests the need to suites of schedulers that can be used in different operating scenarios. This project deals with the study and development of a variety of scheduling scenarios and algorithms that can help in achieving the ultimate goal of furthering our understanding of grid scheduling.

Simulation of Modular Robotic Systems
The project will focus on simulation of a multi-segment snake-like (salamander) organism, with actuators (“muscles”) attached to individual segments (“vertebrae”), investigating emergent locomotion. For example, a particular side-winding locomotion emerges as a result of individual control actions when the actuators are coupled within the system and follow specific evolved rules. There is no global coordinator component in the evolved system, and the amount of mutual information between groups of actuators grows as the distributed actuators become more coupled.

Algorithms for Protein Folding
Protein folding algorithms aim to understand how the amino acid sequence of a protein determines its unique native conformation. Experimental methods, such as X-ray crystallography, determine the native structure of a protein with a great accuracy. However, there is need for more powerful and “intelligent” optimization algorithms to simulate the process of folding. This project deals with the development of parallel optimization algorithms based on traditional and artificial life paradigms to solve protein-folding problems. These techniques are deployed onto grid-enabled platforms to facilitate the execution of these compute- and data-intensive algorithms.

Modelling of Complex Metabolic Pathways Genomics and Proteomics
Complex multicellular organisms contain large genomes in which each structural gene is associated with at least one regulatory element and each regulatory element integrates the activity of at least two other genes. The nature of such regulation started to be understood from the analysis of small prokaryotic regulation subsystems and the current picture indicates that the webs that shape cellular behaviour are very complex. This project deals with the development of algorithms and tools that can be used to analyse metabolic networks based on heuristics and meta-heuristics.

Properties of Enzymatic Signalling Systems
Signalling within and between cells is driven by a complex set of molecular interactions. One type of signal is generated by a ligand (for example erythropoietin, or insulin) hitting its receptor, and inducing a cellular event, often by an oligomerisation of the receptor. There are various cellular tricks by which the receptor-mediated signals can be made digital or analogue. Another situation is where an enzyme generates a biologically active molecule from either an inert substrate, or sometimes a substrate that has an opposing biological effect. In these systems it is not just the concentration of the cognate ligand of a receptor, but the balance of substrate vs. product, and the activity and location of the enzyme that catalyses this reaction are critical.

The latter systems have been described as biological ‘rheostats’ and thus might have fundamentally different utilities in driving cellular behaviour. For example it might be envisioned that enzymatic signalling systems have broader effects on protein behaviour and/ or gene expression than ligand driven ones, or that enzymatic systems have intrinsically greater subtlety than ligand driven ones (where often very low receptor occupancy is sufficient to drive a full biological effect.) An example of an enzyme driven signalling system has as its fulcrum sphingosine kinase, catalysing an anti-proliferative, pro-apoptotic molecule’s (sphingosine’s) phosphorylation into a pro-proliferative anti-apoptotic sphingosine-1-phosphate. Additional complexities lie in the regulators of the activity of the kinase, as well as substrate availability and degradation of the product.

A fuller understanding of the workings of these systems might allow more penetrating biological questions and ultimately drugs better designed to prevent harmful outcomes.

MicroRNAs as Regulators of Cellular Programs
The metaphor ‘leader of the orchestra’ has often been used in biological systems to describe master regulators of complex cellular events or of development. Transcription factors, which work alone or in combinatorial systems to promote or prevent gene expression, have often been thought of as key ‘conductors’.

More recently a new class of regulation has emerged by non-coding RNA molecules. There is only a limited number of these molecules (of the order of hundreds) identified, but each has targets in a number of genes, primarily regulating their translation. Thus, in our hands, transformation of epithelial cells to mesenchymal cells is associated with the loss of several miRNAs, that drive the changes in proteins at cell-cell junctions that allow sessile epithelial cells to change into motile mesenchymal ones.

Whilst transcription factors and miRNAs target different levels of regulation (transcription vs translation) models that examine the reasons why these different control systems evolved and the what utility they might have individually and collectively, would provide an useful framework for further biological enquiry.

Common Gene Networks: Finding Underlying Patterns of Alcohol Mediated Liver Injury Across Species
Alcoholic liver disease (ALD) is a complex multifactorial and multistep chronic disease process which typically progresses through stages of alcoholic steatosis (AS), alcoholic hepatitis (AH), alcoholic cirrhosis (AC) to end-stage (ES) liver disease. Alcohol initiates liver injury by generation of oxidative and non-oxidative alcohol metabolites. With continuing alcohol use, the disease progresses via continuing cellular injury, inflammation, impairment of hepatic regeneration, and increasing fibrogenesis leading to cirrhosis and its complications. Human studies are complicated due to many confounding factors associated with ALD. Hence experimental animal models can be used to determine the effects of alcohol under controlled environment. Of the various animal models of ALD, the baboon model has been shown to reproduce sequential development of all the liver lesions observed in human alcoholics without exposure to other hepatotoxins. Studies on rodent models of alcohol report liver injury but not to the extent found in human and baboon and require additional insult to develop cirrhosis.

We have recently described the global intrahepatic transcriptome in progressive stages of this disease in human and the baboon model of ALD. We have identified several molecules/pathways that are common between human and the baboon model suggesting that these are important in alcohol mediated pathophysiology over and above the confounding factors. Our collaborator has also described intrahepatic transcriptome profiles in rodents (rats and mice) given alcohol.

It is fortunate that data from various species are available and the strength of this project is to compare the underlying effects of alcohol that result in liver injury. It is also provides an opportunity to identify the molecules that prevent severe liver injury in the rodent models of alcohol. This will be the first time that such a comparison amongst species will be performed in alcoholic liver injury and identify molecules with diagnostic and therapeutic potential.