News and Events
Applications for 2015/16 entry are now open
International students wishing to apply for funding via the UCL Overseas Research Scholarship (ORS) must submit an application by January 16th 2015.
Details of how to apply can be found on the 'Prospective Students' page
Professor John O'Keefe wins Nobel Prize for Physiology or Medicine
CoMPLEX supervisor, John O'Keefe has been awarded the 2014 Nobel Prize in Physiology or Medicine for the discovery of cells that constitute a positioning system in the brain - an 'inner GPS' - that enables us to orient ourselves.
10 October 2014
Student and Alumni Profiles
Research topics of current CoMPLEX students
Visual signals for navigation in the cerebral cortex
A fundamental question in computational, systems, and cognitive neuroscience concerns the way the brain is able to perform coordinate transformations. A striking example of such transformations is the one that goes from the representation of the visual field present in primary visual cortex (V1) to the map of the local environment that is provided by the place fields in the CA1 region of the hippocampus. For the purposes of navigation, the information provided by V1 in retinal coordinates needs to be transformed by CA1 to a system of geographic coordinates. We know that this transformation occurs, but we don't know how or where. The first place to look for it is in the system of 9 higher visual areas that process information present in V1. Do they perform at least part of the coordinate transformations, and if so, how do they do so?
This project will attack this problem through a combination of advanced experimental approaches (supervised by Carandini) and computational approaches (supervised by Harris). The experimental techniques include: (1) widefield imaging of cortical activity at the mesoscopic scale, using Calcium sensitive or Voltage-sensitive fluorescent proteins; (2) two-photon microscopy of cortical activity at the cellular scale, using Calcium sensitive fluorescent proteins; (3) if necessary to reach areas below the surface of the brain, multi-electrode recordings of population activity. These experiments will be carried out in head-fixed mice that are trained to perform a simple navigation task in virtual reality. These experiments will provide a large amount of neurophysiological data, which will be analysed using advanced methods of data analysis tailored to decoding the activity of neuronal populations.
Interactions between environmental inputs and grid cell firings
Grid cells are found in the medial entorhinal cortex (mEC) [Hafting et al., 2005]. These cells fire at constant spatial intervals within the rodent's environment, mapping out a regular hexagonal pattern. Grid cells appear to be arranged in functional modules - cells that are proximate in the brain share a common scale and orientation, but differ in their relative spatial offset [Barry et al., 2007; Stensola et al., 2012]. The location of grid firing fields are constant between visits to an environment and, interestingly, the orientation of grid firing fields also appear to maintain a constant orientation relative to the boundaries of different square environments [Stensola et al., 2013].
It has been suggested that grid cells form part of a broader neural circuit that supports path integration - updating an internal estimate of position on the basis of self-motion information [McNaughton et al., 2006]. Importantly, path integration systems accrue error over time, and this is reflected in the drift or complete degeneration of grid firing field patterns in the absence of sensory inputs [Hafting et al., 2005, Bonnevie et al., 2013]. It has also been observed that grid cells emerge later in developmental timelines than either place, boundary or head direction cells, suggesting that input from one or more of these cell types contributes to the observed stability of grid firing fields in familiar environments [Wills et al., 2010; Langston et al., 2010; Bjerknes et al., 2014].
Taken together, these experimental results suggest that grid cells receive sensory inputs that stabilize their firing patterns in familiar environments, produce a specific grid field orientation relative to the boundaries of square environments and reduce accumulating path integration error in the grid cell network [Bush and Burgess 2014]. A likely candidate to provide this input are boundary cells [Barry et al., 2006; Solstad et al., 2008; Savelli et al., 2008], which fire at a specific distance and direction from environmental boundaries, independent of orientation.
The aim of this project is to examine the interaction of border and grid cells in the mEC through a combination of computational modeling and rodent in vivo electrophysiology. This includes the development of neurobiologically plausible models of the integrated grid cell circuit that can account for the experimental observations described above, and make further predictions that can be validated through the analysis of electrophysiological studies in rodents.
Computational techniques will include: simulations of neural firing rates contingent on the animal's location using ODEs and real behavioural data (e.g. Bush & Burgess, 2014); simulation of modification of synaptic strengths between neuronal populations using unsupervised learning rules; maximum likelihood modelling of neuronal firing assuming spatial models for the mean rate of each spatial cell type (boundary cells, grid cells etc), behavioural occupancy data and Poisson firing statistics (e.g. Burgess et al., 2005). Experimental techniques will include participation in in-vivo electrophysiological recording of spatial cells in the hippocampal formation of foraging rodents during manipulations of environmental and proprioceptive input, including the use of virtual reality (e.g. Chen et al., 2013), and investigation of neuronal recordings in developing rat pups (e.g. Wills et al., 2010).
Synthetic biological engineering of commensal E Coli for intestinal microbiome therapeutics
The human gastrointestinal (GI) tract and the wide range of prokaryotic and organisms it supports - here referred to as the intestinal microbiome - form a mutualistic host-microbe symbiotic system crucial for many processes including the breakdown of plant polysaccharides, microbial fermentation and metabolite production. Variation and disruption of this natural ecosystem is linked to a diverse array of disorders including infectious disease, autoimmunity, obesity and cancer. Probiotic treatments aim to alleviate some of these disorders by recolonising and restoring the normal intestinal microbiome.
The engineering of commensal microbiota through synthetic biology approaches has the potential to create novel therapeutics and diagnostic tools, that can be triggered by external and environmental signals, and delivered to localised regions of the GI tract in a context dependent manner. Of paramount importance is that the disruption to the natural inhabitants of the niche is minimal, which can be achieved through engineered intelligent gene systems.
One aim of our group is to engineer the probiotic strain E coli NISSLE 1917 into a versatile therapeutic device that can deliver molecules directly into the GI tract. We will develop, characterise and mathematically model, a suite of inducible, remote control gene expression devices using plasmids implementing robustly designed gene networks. For application in both animals and humans, release of antibiotic resistance genes into the environment must be minimised. Our system will be retained in the host chassis without the need for antibiotic resistance using toxin-antitoxin (TA) systems. This enabling technology will provide a foundational platform for more advanced cellular nanomachines. Future targets could include alternatives to antimicrobials in animals and humans, prophylactics for infectious disease, and novel ways to tackle chronic diseases such as diabetes and obesity.
Super-Beacons and Beacon-STORM: a new generation of small tuneable photoswitching probes and super-resolution approaches
We aim to improve super-resolution localization microscopy, allowing access to a dynamically tuneable imaging framework for live- and fixed-cell high-speed multi-colour super-resolution. At its core, the project entails the development and validation of a novel type of modular super-resolution probe of simple synthesis - Super-Beacons, permitting to easily plugin most available synthetic fluorescent dyes into a linker scaffold that induces photoswitching by transient high-efficiency quenching. This feature avoids the need of non-physiologic (generally toxic) photoswitching inducing buffers, thus allowing live-cell super-resolution. The probe scaffold can be easily modified to generate light, chemical or thermal sensitive triggers that modulate the fluorescence pulsing rate or arrest the probe in either an ON-or-OFF state. A hardware-software imaging framework will be developed to optimize the probes switching kinetics while super-resolving biological structures. We aim to fix a major unbalance between the current properties of probes for super-resolution localization microscopy (bulky and nearly uncontrollable switching dynamics) and imaging apparatus whose performance depends on specific probe kinetics and size. These developments will be critical to address fundamental problems in cell biology and of interest to the life-sciences research community.
The Drosophila circadian clock: Experimental and computational analysis of (i) different sensory entrainment pathways and (ii) integration of sensory information from different modalities
All forms of life adjust themselves to the daily rhythms of their environments using endogenous oscillators commonly called circadian clocks. Peripheral and central body clocks exist, and both require extrinsic information (e.g. light or temperature changes) to keep in sync with the geophysical cycle (entrainment). In addition, intrinsic cues that can be externally measured (e.g. activity levels) have been linked to clock entrainment. Recently, we showed in the fruit fly Drosophila melanogaster, that activation of one class of proprioceptors (chordotonal organs or ChOs) is sufficient to entrain central brain clock neurons. Proprioceptors are mechanosensors that monitor the positions, and relative movements, of an animal's own body parts. Since activity is itself partly regulated by circadian clocks, the existence of proprioceptive entrainment pathways that must be integrated with other entraining Zeitgebers implies substantial richness in the animal's conceptions of time.
Using Drosophila, this project will investigate the relationship between sensory and circadian systems. The project's first arm is dedicated to the dissection of the neurobiological bases of sensory, and particularly mechanosensory, clock entrainment. It will e.g. aim to (i) identify the specific stimulus requirements for effective entrainment, (ii) determine the underlying mechanosensory pathways and, in a combined computational and experimental strategy, (iii) quantify the precise contributions of an animal's activity to its sense of time. Computational and experimental strategies will be used to analyse the nature and impact of activity feedback. A specific aim will be to disentangle the influences of extrinsic and intrinsic pathways. The combined experimental/computational part of the project will specifically focus on the question of neuronal integration, weighting and reliability of multi-sensory cues.
The project's second aim is to build computational accounts of these experimental results, based on an evolving understanding of the intra- and inter-cellular mechanisms involved in clocks. We will initially consider Weakly Coupled Oscillator models, which assume that interactions among the oscillatory variables are sufficiently puny that only the phases of the underlying oscillators need be represented. The structure of the underlying interactions (mutual entrainment, master-slave etc.) will then be inferred using Bayesian inference in tailored models. An alternative approach relaxes the assumption of weak interactions, allowing dependencies among both amplitudes and phases of the experimental variables. Here the project will apply a generic approach from machine learning based on Nonlinear Systems Identification. This will allow a more sophisticated treatment of the relative strengths of extrinsic inputs (light, temperature) versus intrinsic variables (e.g. locomotor, of ChO, activity) on the fruit fly's circadian rhythms.
We will apply various techniques from the modern Drosophila experimental toolbox, e.g. genetically encoded neuro-manipulation (e.g. silencing or hyperactivation of ChOs by means of thermo- and optogenetics) or neuropharmacological approaches (e.g. ChO-specific inhibitors) to tease apart and test the resulting models.
Whole-genome sequencing and mathematical modelling of isoniazid resistant tuberculosis in hard-to-reach populations
The UK sees around 8,500 cases of tuberculosis in a year, around which ~6.5% are resistant to the cheap, effective, first-line drug isoniazid. An outbreak of isoniazid resistant tuberculosis has been occurring in hard-to-reach populations in London since 1995 (Maguire Eurosurveillance 2011). Hard-to-reach groups have trouble engaging with traditional models of clinical practice and this outbreak has been difficult to contain.
Whole genome sequencing, coupled with mathematical modelling and social-network analysis is known to be a powerful tool to capture the dynamics of tuberculosis outbreaks and therefore to identify routes of control (Gardy NEJM 2011). For the first time in the UK systematic whole genome sequencing of isoniazid resistant tuberculosis cases is being planned and funded through a UCL-Public Health England collaboration. Building on the demographic, clinical and contact tracing data available from Public Health England's Enhanced Tuberculosis System this project will therefore seek to model the dynamics of the London outbreak and investigate effective mechanisms for control using an individual based model.
By sequencing strains of TB that are isoniazid resistant, we would generate a unique dataset of WGS, contact tracing and epidemiological data. This will allow us to advance methods (based on Didelot et al MBE 2014 as a starting point) to infer transmission dynamics, for example comparing transmission in household versus transmission at the community level, and taking into account age mixing patterns and place of origin. This calls for novel statistical and modelling approaches. Specific biological questions to be addressed by the novel modelling approach will include the extent to which INH resistance is being acquired vs transmitted in London. Finally, we will model approaches to determine whether there likely points where we might have missed individuals. This will allow us to ask whether particular approaches would be better at containing the outbreak (e.g. respondent driven sampling), and ultimately if real time WGS and analysis might impact TB control. These findings will have application beyond the current outbreak and London and should advance both TB control and mathematical biology.
Advancing cochlear implants
Cochlear Implants (CIs) are the most successful sensory therapeutic technology, as judged by their ability to restore hearing function to the severe and profoundly deaf. CIs work by stimulating directly the auditory nerves, bypassing damaged or absent sensory hair cells in the inner ear. Today, nearly 400,000 people worldwide have a CI, and in UK children born deaf receive two implants, often in the first few months of life. Such is their success in restoring hearing function that individuals with some remaining (usually low-frequency) hearing are now candidates for implantation - implanting the many individuals with residual low-frequency hearing will combine the benefits of acoustic hearing at low frequencies (critical for distinguishing voices, sound localisation and ‘cocktail party listening’) with the benefits of electrical hearing for distinguishing speech sounds. The aim of this project is to develop combined Electrical and Acoustic Stimulation (EAS) stimulation strategies that optimise listening performance in deaf, implanted individuals with residual hearing function, and to test these strategies in implanted animal models of EAS. Understanding how the brain combines information about sound across very different forms of input (acoustical and electrical) will be a key outcome of this study.
Accoustic tracking to understand behavioural environment of wild animals
Understanding the activity of wild animals in their natural environment is of critical importance for their conservation. Monitoring habitat-use, movement and behaviour can aid understanding of how species may interact with their surroundings and the ecosystems to which they belong. Recent developments in technology have allowed GPS tracking of a wide range of animals, via animal-borne tags, however whilst GPS tracking can provide detailed locational data over time and additional environmental variables can be identified via remote sensing, it cannot yet provide concurrent information on specific events occurring in the animal's surroundings. In this PhD, we hope to deploy newly developed miniature animal-borne audio-recorders in conjunction with GPS tags, enabling acoustic as well as spatial monitoring of animals in the wild.
The additional deployment of an audio-recorder during tagging will produce a corresponding soundscape of an individual. Machine learning will then be used to identify particular events in this soundscape, for example noise caused by human activity, such as boat motors. Calls made by the focal animal or by nearby members of the same or another species will also be classified in the data, providing potential for increased understanding of intra- and inter-specific interactions. The extra dimension provided by acoustic data will also be used to build upon existing methods of behavioural identification, allowing more accurate exposition of the relationship between behaviour and an animal's environment.
Devices are planned for deployment on Black-browed and Wandering albatross (in collaboration with the British Antarctic Survey) this winter (Dec-Jan). While these species are subject to high mortality due to fishing industry by-catch, the frequency and spatial pattern of non-lethal interactions with smaller fishing vessels is currently unknown. Furthermore, the composition and potential function of mixed species rafting is unknown. Acoustic tracking to identify fishing vessel interactions and the composition of multi-species rafting events will provided unparalleled information on the foraging behaviour of these species. In addition to on-going opportunities to deploy these devices on UK species (Gannets) and Galapagos Tortoise, we are keen to explore opportunities to further miniaturise these devices for deployment on Bat species, where not only are acoustics actively used for foraging and communication, but where these species are critical indicators of ecosystem health.
This interdisciplinary PhD would bring together new developments in miniature tracking technology, novel statistical and mathematic analysis (to decompose and model spatiotemporal & acoustic behaviours) to determine the complex relationships between an individual, its local behavioural context, and the broader environment on which it relies.
New phylogenetic methods to resolve deep animal phylogenies
Reconstructing accurate phylogenies is a major pursuit in biology, yet despite the progress made in our understanding of deep metazoan phylogeny, important questions remain. These typically involve interesting cases of rapid radiation or of rapid divergence-two drivers of biodiversity change and biological innovation.
State-of-the-art species trees are inferred from concatenated, "carefully" selected marker gene. This is wasteful at two levels. First, most genes are simply not considered. Second, even for those that are, by "averaging" their sequence through concatenation, we stand to overlook interesting instances of rapid evolution having occurred in individual genes and branches.
This interdisciplinary project involves both experimental and computational biology. In the experimental part of the project, we will collect fresh material from multiple members of key taxa, extract RNA and use NGS to produce largely complete transcriptomes. In the computational part, we will implement a novel and efficient maximum likelihood tree inference method that relaxes the assumption of identical branch length across all orthologous genes and instead only assumes a joint tree topology (i.e. speciation order).
The project will exploit the recent availability of efficient software libraries for tree likelihood computation. The new method will be validated using a broad variety of simulated sequence data.
Computational components of the project:
- Bioinformatics: transcriptome assembly and annotation, orthology inference.
- Computer science: software engineering, code profiling and optimisation, numerical analysis, high-performance computing
- Statistics and Modelling: Markovian models of sequence evolution, phylogenetics
Life Science components of the project:
- Experimental biology: Specimen collection of principally marine taxa from Kristineberg marine station, Sweden as well as via collaborators. Nucleic acids extraction, NGS short read transcriptome sequencing (funded by existing grants).
- Evolutionary Biology: interpretation of evolutionary trees
Sensor networks and control system of WAM exoskeleton project
Orthotics and prosthetics are becoming increasingly advanced, however they still remain heavy and bulky with subjects at risk of developing pressure sores . This project will aim to develop a flexible sensor skin system to be in contact with the user's body, which will have the ability to flex and articulate. The device must be able to provide information about forces applied related to articulating joints of the body or prosthetic, and also warning of possible damage to skin/bone of user, in affect joint proprioception and must be biocompatible.
The most common transduction techniques for wearable sensors are based on capacitive, piezoresistive, thermoresistive, inductive, piezoelectric, magnetic and optical methods . The intrinsic principles associated with these techniques are well established but none have been used successfully as yet. The optical systems have the advantage of very low hysteresis, at a very high frequency response. Piezoelectric systems are easy to fabricate, and are generally stiff and unaccommodating, hysteresis is also a problem affecting frequency response. Developing a new sensor system for use in orthotics and prosthetics, modelling their ability to monitor the movement of the human body, and testing their performance is the aim of this PhD project.
Investigating viral evolution using phylogenetic models
Viruses undergo host shift events upon transmission from one host species to another. These shift events are accompanied by a change in the host specific selective constraints active on the virus. In the case of influenza, the 2009 human H1N1 pandemic virus was the product of multiple host shift events, from waterfowl to pigs and subsequently pigs to humans. The 1918 'Spanish flu' pandemic was far more devastating than the 2009 pandemic, but the transmission route this virus underwent remains unclear. The Goldstein group has developed computational models to investigate the selective constraints active upon genes of interest, and characterising how these constraints change. By applying these models to investigate the changes in selective constraints acting on influenza, we aim to elucidate the host shift events that preceded the 1918 influenza pandemic. Understanding such past transmission histories is vital to understanding the process that leads to a pandemic outbreak and therefore to preventing future outbreaks.
It is unclear whether the virus which participates in the host shift has evolved traits that bias the virus for transmission or whether selection of the virus which will be transmitted is a purely opportunistic process. Substitutions have been identified which characterise the pig-human transmission of influenza. However, the avian-pig host shift is likely to present a greater challenge to the virus as it must move from an avian to a mammalian host. The changes a virus needs to undergo to permit this transmission are less well defined. At such transmissions, directional selection is predicted to act on viral proteins as the virus needs to adapt to a new situation. We aim to apply our evolutionary models to identify directional selection acting on influenza between avian and swine hosts, and using ancestral construction techniques, determine how 'typical' the host shift virus was. Identifying protein sites which are under such selection is vital in surveillance efforts to identify viruses with the ability to change host.
Current phylogenetic models are unable to deal with polymorphisms as they treat all changes as fixed changes. We aim to extend our current phylogenetic models to incorporate population genetics and therefore polymorphisms. We will then apply these population genetic models to situations in which multiple strains of a single virus exist within a human population, such as influenza and norovirus. These models will also be applicable to examine intrahost HIV evolution in deep sequencing datasets when such datasets become available.
Understand the reciprocal interactions between the Gut Microbiomes and Paediatric Immune System
At Great Ormond Street Hospital for Children, we have many children who have an underlying congenital immune defect, such as missing T cells or B cells. These patients usually require a Bone Marrow Transplant or Gene Therapy to correct the defect and survive. It has become increasingly clear that during the recovery phase in patients who survive, the short-term and long-term outcomes, such as how well they grow and how many infections they get, are influenced by environmental factors. The gastrointestinal tract (GIT) (from the mouth to the anus) has co-evolved with microbial communities that are specific to a particular mucosal ecological site. These microbiomes, as they are called, contains more than 10 times the total number of cells in the human body and 100 times more genes. It is likely that changes in the GIT microbiome are important determinants of outcome. We are investigating the short and longterm impact of host immunity and antibiotics on the composition and complex structure of this GIT microbiome. We are then examining these data with indices of immune function to examine how the GIT microbiome may subsequently influence immune function. At present we are using a 16S rDNA gene-based approach, which allows us to sequence 16S rDNA that all bacteria possess to identify bacteria from the GIT sample. Sequences are clustered into operational taxonomic units against a reference database. This produces a list of species and their relative abundance for each sample, as well as species richness and phylogenetic diversity indicies, which can then be compared group to group. We are now also generating metagenomic data, which will enable us to define the microbiome at higher resolution and provide data on viral and fungal organisms. A combination of bioinformatic methodologies will allow us to delineate the composition of microbial communities and their relationship with the human immune system. This will provide us with the opportunity to derive predictive models for future disease outcomes and ultimately give insight into novel forms of therapy.
The PhD student will work on the bioinformatic components of this project. They will work within a team which includes scientists who are performing DNA sequencing, Cell Biologists, other Mathematicians, Statisticians and Clinicians. The team includes other PhD students and Post Doctoral Scientists. This multidisciplinary team is based at UCL. We have developed a highly successful model of supervision that includes experts pertinent to the project. In this case the PhD student will be supervised by experts in Statistics and Mathematics, Modelling, Microbial Genetics and Medicine.
Gene Sequence data will be analysed to obtain Taxonomic information. Alpha (Chao, Shannon index, coverage) and beta diversity (weighted and unweighted unifrac distances) for each sample will be calculated and used for analysis in R. Microbiome data from various patient groups will be integrated with data generated from immunological studies. A range of bioinformatics techniques will be used such as exploratory (hypothesis generating) data analysis to best describe each group and to ascertain differences between groups. We will employ supervised pattern recognition techniques including linear and nonlinear latent variable regression methods such as OPLS-DA (orthogonal partial least squares (PLS) discriminant analysis) and O2-PLS-DA to detect associations between the microbial ecology and immunological profiles.
The limits of Deep Learning for Vision
This PhD project would be conducted within CS and be primarily computational in nature, though psychophysical experiments will be used as appropriate. Alan Johnston will provide input on human visual mechanisms, making the research outputs accessible to psychologists, and design of psychophysical experiments. A developing collaboration with Pieter Roelfsema (Amsterdam) will provide relevant neurophysiological input to the project, with the student making annual visits to his lab.
Bag-of-Visual-Words and (more advanced) Deep Learning approaches have made a huge impact in Computer Vision, delivering performance on many categorization tasks that appears to approach human levels. The approaches also have neural plausibility, so stand as candidate models of biological vision. This project is concerned with challenging these models by discovering gaps between machine and human capabilities when applied to non-natural images. The goal of the project is to advance understanding of human vision, a secondary goal is to develop improved computer vision methods that can be applied to the problem domains (Biomedical, Security, Forensic and Geo Sciences) studied in the Griffin Lab. Three work packages are planned.
1. Understanding adversaries
Adversaries are images, very close to training images, that are incorrectly categorized by Deep Learning approaches. Adversaries seem unlikely to be a phenomenon of human vision. We hypothesize that they exist because DL approaches use many weak cues (which can be broken with the correct small pixel changes) rather than strong cues of human vision (which cannot be so easily negated). The student will continue work started during a CoMPLEX MRes summer project on demonstrating that adversaries exist even for Bag-of-Words type approaches. The properties of these adversaries will be determined, and the weak-cue explanation developed. The student will develop algorithms that find adversaries which are similar to training images, according to a perceptually-based metric, rather than merely pixel similar.
2. Topological categorization
B-o-W and DL approaches are based on local analysis, though repeated and iterated across position and possible scale. A good candidate for discriminations that they cannot make are topologically-based ones. In particular distinguishing between classes of patterns that differ in connectivity or enclosure, though are composed from the same local pieces. Having demonstrated maximally simple classes of such patterns that cannot be discriminated we will design extensions of BoW and DL that can. To inspire these extensions we will look to recent neurophysiology on propagating lateral processes within visual areas, as well as recurrent intra-areal processes.
3. Global shape
We will design classes of images which require global shape to be used for categorization, local weak cues having been removed. Such images can be expected to be a sore challenge for BoW and DL, while being easy for human vision. One approach to producing such images is to generate sparse line drawings, another is to mosaic images out of mis-leading small fragments (e.g. a car image made from horse details).
Investigating the role of human dopamine in encoding uncertainty in decision making
Substantial evidence suggests that the phasic activity of dopamine (DA) neurons encodes disparities between predictions of reward and actual rewards received (Montague et al, 1996). This signal plays a critical role in the acquisition and expression of decision-making, and thus in a wide range of neurological and psychiatric conditions.
However, with two exceptions (Montague et al, 2011, Kahana et al, 2009), all the experimental work on phasic aspects of dopamine has been conducted in non-human animals. Through solving a set of technological and ethical issues, the first supervisor's lab is now able to use cyclic voltammetry to measure phasic dopamine levels in vivo in human patients performing learning and decision-making tasks. The aim of the project is first to validate the findings from rodents and primates in humans, and then to exploit the vastly greater flexibility and sophistication of human decision-making to probe much more general aspects of the nature and role of dopamine.
The life-science content of the project concerns the psychology and neuroscience of affective decision-making. We will design and execute experiments looking at the impact of different sorts of uncertainty (for instance, data are being acquired using a variant of the paradigm used in macaques by Fiorillo et al, 2003), different aspects of positive and negative affect, and also forms of decision that are known from behavioural and neuroimaging analyses to engage structurally different neural mechanisms (Daw et al, 2005; Daw et al, 2012).
The maths/physics/computational content of the project concerns (a) methods for extracting as high quality information as possible from the cyclic voltammetric measurements -- we will create Bayesian methods that better reflect our uncertainty about the neuromodulatory nature of the signal (and thus also estimate serotonin as well as dopamine); and (b) computational modelling of the components of the decision tasks to make and refine clear quantitative hypotheses about the signals we are measuring. This sort of modelling is routine in neural reinforcement learning, and dovetails perfectly with non-invasive work in healthy volunteers ongoing in the labs of both supervisors.
Methods and practice of detecting selection in human cancers
Cancer progression is inherently an evolutionary process. Cancer is initiated when a cell acquires somatic mutations that confer a fitness advantage within the tissue ecosystem, and tumours then progress via the clonal expansion of this first cancer cell. Clones within the tumour population undergo mutation, selection and compete for resources, a process which inevitably results in a genetically diverse tumour. It is this genetic diversity within cancer that makes it a difficult disease to treat. From an evolutionary perspective, cancer treatments are in effect a source of artificial selection that select for variants within the population that are resistant to the therapy given to the patient. Such treatments leave behind the resistant cancer cells, leading to later relapse of a treatment-resistant cancer.
A better understanding of the intra-tumour clonal dynamics of cancer progression would therefore enable us to better comprehend the shortcomings of present treatment strategies and to develop newer more effective ones, where we might harness the evolutionary dynamics of cancer clones to our advantage. One important facet in the dynamics that is yet to be fully understood is the role that clonal selection plays.
The traditional view of selection within the clonal dynamics in cancer is that subclones with mutations that confer a high fitness compared to the rest of the population clonally expand and come to dominate the tumour. Successive clonal expansions of selectively advantageous mutants results in an increasingly fitter tumour population. More recently an alternative hypothesis has been put forward that suggests that selection plays little role in tumour progression, and that the time that mutants arise within the tumour lifetime is more important than any fitness advantages it may have. Mutants arising early in the tumour lifetime will according this alternative hypothesis be more dominant in a tumour than those arriving later on.
The project will thus focus on elucidating the role selection plays within cancer, and ultimately attempt to answer how important it is in tumour progression. Due to the nature of the disease, time series data is difficult to obtain, but multi-region data sampling of individual cancers is available. However, in practice, to infer evolutionary dynamics from these data requires making inferences from a relatively sparse sampling of a spatially-structured and highly heterogeneous population of tumour cells. Mathematical and computational modelling is required to relate underling clonal dynamics with the patterns of genetic alterations observed in the genomic analysis of cancers. The student will use both mathematical modelling together with bioinformatics analysis of next generation sequencing data from cancer to investigate these two alternate hypotheses and the role of selection in cancer more generally. The student will also be encouraged to spend time in the wet-lab, collecting the appropriate data on within-tumour diversity that can be used to test the hypothesis generated by their model.
Computational modelling will involve simulating tumour growth using standard non-spatial population genetics approaches such as infinite-allele Wright-Fisher type constructions and birth death processes, and then developing spatial models that are more appropriate to the laboratory data. Analytical approaches that may be investigated as appropriate include branching processes, Markovian stochastic processes and stochastic differential equations.
Pattern regularity and scale in biology
The near-synonyms – regularity, order, symmetry and pattern – are important aspects of image structure. Tools for their analysis are relatively undeveloped, the focus having generally been on limited conceptions of symmetry (e.g. global reflectional only) and patterns which are near perfect symmetry. We will working on definitions of local quasi-symmetry in natural images and how to quantify them.
We will also investigate the problem of scale inimage symmetries and hope to obtain a unique scale or a systematic way of obtaining it for any natural image. We will also produce computer software that autonomously computes local symmetry in natural images. By imaging developing embryos as they reach the end of development it is possible to see order emerging within tissues. This pattern refinement occurs at different temporal and spatial scales for different processes in the same tissue. To shed light on this process of refinement, the aim of this project is to develop algorithms that can automatically identify ordering events at different temporal and spatial scales in the developing fly.
Investigating the specificity of the cortical responses elicited by stimuli of different modalities in humans
Whether primary sensory cortices are essentially multisensory or whether they respond to only one sense is an emerging debate in neuroscience.
The group I am working with recently used multivariate pattern analysis (MVPA, or machine learning) of functional magnetic resonance imaging data in humans to
demonstrate that simple and isolated stimuli of one sense elicit distinguishable spatial patterns of neuronal responses, not only in their corresponding primary sensory cortex, but in other primary sensory cortices (Liang et al Nature Comms 2013). These results indicate that primary sensory cortices, traditionally regarded as unisensory, contain unique signatures of other senses, which prompts a reconsideration of how sensory information is coded in the human brain.
The planned research aims at further characterising the specificity of the response from different cortical areas to stimuli of different sensory modalities, and also exploring the amount of stimulus features that are encoded in the response elicited by stimuli of one given sensory modality in non-pertinent primary sensory cortices. The research will involve the collection of electrophysiological (EEG/MEG) and metabolic (fMRI) functional neuroimaging data, and their analysis using techniques like machine learning.
Dynamics of epitope presentation on MHC Class I molecules following viral infection
Major Histocompatibility Complex (MHC) molecules make up part of the cell-mediated adaptive immune system in which antigen-specific defence mechanisms are acquired against threats such as viruses and tumours. MHC-I alleles present short sequences of proteins, known as peptides, which are cleaved in the cytoplasm primarily by the proteasome. Some cell surface peptide-MHC complexes are able to activate cytotoxic T lymphocytes (CTLs) through interactions with T cell receptors (TCR). The probability of CTL activation is determined by the peptide cell surface abundance and the binding affinity between the presented peptide and the TCR. Those peptides which activate T lymphocytes are known as epitopes. In this project we will be focussing on the role of the timing of the presentation of viral epitopes on the cell surface in determining the immune response. This project aims to combine a kinetic model of the pathway of viral peptide presentation from supply to the Endoplasmic Reticulum (ER) to the peptide-MHC complex cell surface abundance, with the production of defective ribosomal proteins (DRiPs) which are implicated in many viral infections. The kinetics of DRiP production will be guided by experimental data and the epitope dynamics of multiple viruses will be compared. It is hoped that such a model will aid the understanding of epitope production and presentation and the role of timing in determining the immunodominance of certain few viral epitopes despite the large variability in peptides ligands at the surface. Understanding what leads some epitopes to become immunodominant may inform the design of peptide vaccines against viruses and other novel therapies.
Evolutionary game theory models for the origins of agriculture and the rise of social inequality
The origin of agriculture in southwest Asia at the end of the last Ice Age and its subsequent spread as the main form of human subsistence was one of the most important transformations in human evolution and continues to have major consequences today for many aspects of human health and ways of life (e.g. Bocquet-Appel 2011, Laland et al. 2010). Recent work has challenged behavioural ecology explanations that have seen it as the endpoint of a long-term expansion of
human diet breadth to incorporate lower-ranked resources with increased processing costs (Bowles 2011). Instead Bowles and Choi (n.d.) have developed an evolutionary game theory model of the returns from hunting and gathering versus farming in the context of different systems of property rights by comparing the payoffs of three different strategies – Sharer, Civic and Bourgeois – in relation to the different subsistence systems and the returns expected from them in different climatic conditions. However, there are weaknesses in the Bowles-Choi model. The aim of this project is to develop alternative evolutionary game theory models for the origins of agriculture and go on to model the payoffs to different property inheritance systems and their effects on inequality (cf. Borgerhoff Mulder et al 2009).
Global imaging of physiological and pathophysiological angiogenesis in live zebrafish using optical projection tomography
Cancer research is in its prime, whereby in vivo studies are incredibly important due to global processes often being involved in key cancer hallmarks, which is particularly true for angiogenic diseases. The majority of previous in vivo studies of cancer have required the model organism to be sacrificed to determine the tumour pathology and molecular alterations.
Therefore, genetic variation between the organisms at each time point was problematic, and so multiple sacrifices were required. The interdisciplinary approach of Optical Projection Tomography can avoid this through observing the progression of individual tumours within live zebrafish over time, therefore providing immense accuracy.
Vascular development and inflammation are critical for progression of the disease state and so act as important targets for drug discovery. Developing this system towards a new stage that can characterise these events at a molecular and signalling level is likely to dramatically improve the way in which global cancer development and cancer cell signalling can accurately be determined. I aim to do this through developing appropriate 3-D imaging techniques including optical projection tomography and establishing methodologies, including generation of novel transgenic zebrafish lines, for the study of angiogenesis.
CD4 T cell immune reconstitution in HIV infected children in response to antiretroviral therapy
The number of T and B-lymphocytes in an individual increases in a predictable manner during childhood and then stays remarkably constant throughout adult life in spite of declining production, peripheral cell division and unpredictable responses to infection. The complex homeostatic mechanisms involved are beginning to be understood through combined experimental data, clinical trials and mathematical/statistical modelling approaches.
In HIV infection, CD4+ T cells are lost causing profound immunodeficiency, susceptibility to infection and death. In response to treatment with antiretroviral drugs, normal homeostatic mechanisms in the body strive to replace the lost CD4 T cells by production of new T cells in the thymus and ‘homeostatic’ T cell proliferation in the periphery (lymphoid tissues) but often do not reconstitute to normal levels.
The aim of this project is to develop mathematical models capable of predicting the levels of CD4 counts in children infected with HIV. Non-linear mixed effect models will be used to compensate for individual variability of CD4 counts seen amongst patients infected with HIV.
The following questions will be answered:
1. How good are the models generated at predicting CD4 counts when children become adults?
2. What is the impact of treatment interruption on long-term CD4 cell reconstitution?
Investigating cognitive control in mental illnesses
The nascent field of computational psychiatry stems from an effort to understand mental illnesses in a new way, through large scale mathematical and computational modelling. Modellisation can bear different levels of abstraction, from the algorithmic level borrowed by the Machine Learning framework, to the neural models belonging to computational neuroscience and Reinforcement Learning.
Here, we harness the Reinforcement Learning framework in order to explore cognitive control in mental illnesses. We focus on the way that an illness can alter the way we make decisions, which can be both endogenous (i.e. decisions on the way we use our working memory) or exogenous, decisions on actions. If the mechanisms and metrics which govern these choices are corrupted, the decisions taken will not be normative. This is important in a range of conditions which affect impaired inhibition and impulsivity and also the deployment of attention. Schizophrenia will most likely be our target.
Evolution of reciprocal sex from lateral gene transfer in early eukaryotic evolution
Evolutionary adaptation to new or changing environments cannot happen without genetic variation, which is the source of new, fitter phenotypes in populations of reproducing organisms. The division of higher organisms into two sexes is a strategy to ensure genetic diversity; mixing the genes of two
individuals during sexual reproduction produces offspring with more varied genotypes, increasing the chances of novel adaptive phenotypes. Prokaryotes reproduce asexually, but maintain genetic diversity by processes known collectively as lateral gene transfer, involving the direct movement of genetic material between unrelated organisms rather than the recombination of the parents’ genetic information in the offspring.
Eukaryotes, including all sexually recombining organisms, evolved from symbiotic partnerships between prokaryotes, so recombination presumably first originated in organisms that initially used lateral gene transfer to ensure genetic variation. Using mathematical and computational models, I will investigate how events in the early evolution of eukaryotes might have favoured recombination over lateral gene transfer, and why modern-day prokaryotes and eukaryotes use such different mechanisms to ensure genetic variation.
Magneto-Mechanical Actuation of Muscle
Force produced by permanent magnets in close proximity to the body offers an array of potential biomedical applications. One aspect involves the combined use of permanent magnets to control the orientation, migration and growth of cells grown with magnetic materials via a process termed magneto-mechanical actuation.
My PhD project builds on preliminary data that has shown it is possible to measure magneto-mechanical actuation using a combination of cell culture, in vitro and in silico modelling, and image analysis techniques. It is anticipated we will elucidate the parameters needed to successfully achieve magneto-mechanical actuation for a variety of muscle groups found throughout the body whilst also investigating the mechanical properties of explanted muscles by means of uniaxial mechanical testing. The data we generate will be used for in in vitro and in silico modelling of force distributions across cell populations cultured with magnet materials and exposed to different combinations of permanent magnets.
Such work may well significantly impact the field of tissue engineering as well as potentially forming the foundations of novel muscular regeneration therapies.
Mitochondrial inheritance in the evolution of complex multicellular organisms with germline-soma differentiation
The emergence of multicellularity in eukaryotes transformed the mapping between mitochondrial fitness in germ line and adult viability by introducing the effects of mutation accumulation, segregation of organelles, division of labour and differentiation. These changes have strong effects on the favoured pattern of mitochondrial inheritance and introduce several costs and benefits of having an isolated germ line.
We propose, that mitochondrial fitness and mutation might be central processes in the development of complex multicellular features, including early germ line definition. The hypothesis can be tested by investigating mitochondrial evolution trends in basal metazoans, higher animals and plants. Results of our computational modelling will possibly lead to the theory accounting for the role of extra-nuclear genomes in evolutionary transitions following the emergence of complex cell and multicellularity, explaining the existence of two sexes, germ line/soma distinction and patterns of cellular ageing.
Developing Wearable Assistive Materials (WAM) for Orthopaedic Applications
My research concerns the development of a novel robotic exoskeleton technology, with the ultimate aim of providing an alternative to wheelchairs, walking sticks and other mobility aids. Using chemical actuators and magnetic gels currently in development by the WAM team, we intend to create a material that is thin and light enough to be worn comfortably, whilst its flexibility and shape can be rapidly adjusted with small ionic potentials.
I am investigating the various ways in which the aforementioned drivers could be incorporated into a wearable material, the nature of the control this would deliver, and its applicability to human physiology. Once a suitable control system has been designed, I will be prototyping a proof-of-concept device at the Institute of Making. Working closely with patients at the Institute of Orthopaedics and Musculoskeletal Science, the intention is to be able to produce patient specific Wearable Assistive Materials using an advanced bespoke 3D printer.
Elucidating the structure of the combined regulatory and signalling network controlling haematopoietic stem cell differentiation
Haematopoietic stem cells are perhaps the most studied of all stem cell types. Despite this, the underlying gene regulatory networks that control proliferation and differentiation, and the signalling systems that process the intracellular and environmental cues, remain to be elucidated. Understanding how to manipulate these processes has wide implications for therapeutics and also serves as a model for the control of other multi-potent stem cell types.
This project has two distinct parts. The first phase will involve the building, and refinement of, a mathematical (computational) model of aspects of the regulation of blood stem cell differentiation. The second phase requires experimentation to validate the new models.
Mechanosensory pathways as potential targets for the control of insect-borne diseases
Insect-borne infectious diseases are one of the major plagues of humanity, annually causing millions of deaths. It always starts the same way: a female mosquito bites a human, infecting them with a disease-causing pathogen. Controlling mosquito populations and preventing them from biting humans is therefore a primary goal in global disease management.
Disease transmission is inextricably linked to the biology of mosquitoes. My PhD project will study the mechanosensory bases of mosquito blood-feeding behaviour. It seems very likely that, in order to insert their proboscis into the human skin, mosquitoes rely on sensory feedback from mechanosensory organs, like e.g. chordotonal organs (ChOs). We will test this by ablating chordotonal organ function using chordotonal-specific insecticides.
Mechanosensory, and specifically chordotonal, signalling, however, is also a crucial component of the animals’ air-borne courtship, and copulatory, behaviour. The same strategy that impairs the animals’ ability to insert their probosces into human skin could thus also directly impact the animals’ reproduction rate.
Results and implications of these experiments will be analysed with newly devised computational models of disease transmission, using the degree of mechanosensory impairment as a novel model parameter.
We anticipate that our approach will lead the way to new strategies of vector control.
Investigating Processes Driving Genetic Diversity Among Human Populations
Although it is well established that DNA varies substantially among different world-wide human groups, the principal forces driving this genetic diversity are not well understood. This project aims to describe genetic patterns among a wide range of human groups and characterize the primary historical, anthropological and sociological factors that contribute to observed levels of genetic diversity among these groups.
A novel statistical methodology will be developed and applied to available data stored at UCL, which includes genome-wide DNA interrogated at hundreds of thousands of single-nucleotide-polymorphisms (SNPs) in hundreds of world-wide human subjects. These data resources also include detailed sociological and genetic information from the world's largest DNA collection of individuals from Ethiopia, an important country for studying the origins of anatomically modern humans and their subsequent dispersal from Africa throughout the rest of the world. Through merging these unique methods and diverse datasets, this project aims to answer a number of important questions about human evolution.
Page last modified on 10 nov 14 11:07