Autumn 2014

3rd Oct: Colin Campbell

Large scale data integration in the context of biomedical datasets

Many problems in the biomedical sciences require the integration of large and disparate types of data. We consider three recently completed projects involving data integration. First, we consider the prediction of sequence variants in both the coding and non-coding regions of the human genome, to predict if the variant is functional in disease. Utilising multiple kernel learning, the constructed predictor gives state-of-the-art performance in predicting the status of non-coding variants. Second, we consider a model for the prediction of breast cancer progression using a weighted combination of a variety of different data sources, from genomic data through to clinical measures. Apart from building a model for predicting disease progression, an accurate predictor implicitly indicates those features which most heavily influence disease progression and mortality risk. Next, we consider a variant of Canonical Correlation Analysis for finding correlated linear combinations of features in paired datasets (in this case SNP and disease phenotype data from the British Women’s Heart Survey). This lead to the discovery of novel single nucleotide variants apparently associated with several cardiovascular disease phenotypes. We then discuss future directions.


10th Oct: Conor Houghton

Calculating mutual information for spike trains and other data with distances but no coordinates

Many important data types, such as the spike trains recorded from neurons in typical electrophysiological experiments, have a natural notion of distance or similarity between data points, even though there is no obvious coordinate system. In this talk a simple estimator is presented for calculating the mutual information between data sets of this type.


17th Oct: Sergey Kasparov

Glio-centric view of the brain

24th Oct: Thelma Lovick

To pee or not to pee? – investigating a midbrain network that controls urinary voiding

For successful micturition (urinary voiding) the bladder must contract whilst the urethral sphincter simultaneously relaxes to enable urine to be expelled.  A spino-midbrain-spinal network is engaged to co-originate the event, which occurs only when the individual judges it is safe and socially acceptable to do so, implying that the controlling network can be switched on and off.  Using a rat model we will show that the functional integrity of the midbrain periaqueductal grey (PAG) is key to successful voiding and that the PAG contains a number of neuronal cell types whose firing is synchronised with different components of a void.  By considering the patterned activity of these neurones, it may be possible to model a functional in vivo network that integrates and controls a distinct physiological event.


31st Oct: 2 talks

James Hodge, Edgar Buhl and Krasi Tsaneva-Atanasova

Turning back the hands of time

Animals contain molecular clocks keeping time in dedicated clock neural circuits in their brains. These encode time of day information by changing action potential firing and membrane currents. When animals get old their circadian (~24 hour) rhythms become weaker, with sleep moving earlier in the day and becoming fragmented. These senescence changes and the mechanisms of circadian rhythms are well conserved between flies and humans, but occur in about a month in flies. We will compare fly clock electrophysiology and behavior in young and old flies and determine which properties and currents change and develop a computational model.

Risto Kauppinen

MRI in the context of ‘neural dynamics’

MRI is the gold-standard imaging technique both in basic and clinical brain research. Water, the ubiquitous molecule that samples ‘magnetic environments’ in vivo, is exploited by MRI to form images. MRI directly probes ‘dynamics of water’ in the brain through relaxation (T1 and T2) and thermal translations (diffusion) to generate images from brain macro- and micro-anatomy. Instead, ‘brain activity’ is revealed indirectly by MRI through mechanisms that are only partially characterized. Mathematical modeling of MRI signals yields both anatomical and functional connectivity maps with unprecedented value for understanding functional organization of the brain. This NDF presentation summarizes on-going MRI projects at CRIC with strong link to mathematical modeling


7th Nov: Dave Lester, University of Manchester

How to build an exascale supercomputer for computational neuroscience

In this short talk I will outline the challenges faced by chip designers when they contemplate the next generation of supercomputers: the most pressing of which is how to keep the energy budget manageable. The SpiNNaker team at Manchester has taken its inspiration from the proven energy efficiency of the brain, and has already produced chips and systems for neuroscience and robotic applications. These systems are now undergoing thorough testing and evaluation before the next generation system is produced as part of the HBP project. I will discuss the conclusions so far.


14th Nov: David Murphy

Transcriptomic approaches to understanding complex biological systems

The research interests of the Murphy lab are focused on the role of hypothalamic structures in the neurohumoral and behavioural control of salt and water homeostasis. We have used Affymetrix microarray gene profiling to catalogue gene expression in these brain regions in euhydrated and dehydrated male Sprague Dawley (SD) rats. These gene catalogues were then subject to robust statistical analysis to identify genes that are differentially regulated as a consequence of dehydration. We have exclusive access to ReLyter, a proprietary cutting-edge Java tool developed by Source BioScience LifeSciences to make use of the statistical and machine learning functions within WEKA. The resulting gene networks allow identification of nodal genes with multiple links. Analysis of our transcriptome data from dehydrated rats revealed a putative network around Gonadotrophin inducible transcription factor 1 (Giot1), which is robustly up-regulated in the dehydrated hypothalamus. Rats are normally averse to 2% (w/v) NaCl. However, this aversion is overcome if 2% (w/v) NaCl is their only fluid source. An initial decline in fluid consumption is followed by a progressive increase in drinking over the course of the 7-day stimulus, concurrent with an increase in the excretion of large volumes of urine. However, hypothalamic injection of a lentiviral vector that expresses a Giot1 shRNA completely blocked fluid intake following the onset of salt loading. These data suggest that Giot1 is a crucial component of the drain mechanisms that regulate salt and water balance.


21st Nov: Padraig Gleeson, UCL

The Open Source Brain Initiative, enabling collaborative model development in computational neuroscience

Computational modelling is important for understanding how brain function and dysfunction emerge from lower level neurophysiological mechanisms. However, computational neuroscience has been hampered by poor accessibility, transparency, validation and reuse of models. The Open Source Brain (OSB) initiative (http://www.opensourcebrain.org) has been created to address these issues. This aims to create a repository of neuronal and network models from multiple brain regions and species that will be in accessible, standardised formats and work across multiple simulators. OSB will create a collaborative space to facilitate model creation and sharing, where both computational and experimental researchers can contribute to their development. This talk will introduce the aims of the OSB initiative, describe the current functionality of the website and the range of models already available, and present future plans for the project.


28th Nov: Simon O’Connor, Biocomputations Group, University of Hertfordshire

Tools and Techniques for producing Detailed Biophysical Neuron Models

In this talk I will go through the software and techniques that were used to construct a gap junction connected olfactory bulb mitral cell model (O’Connor, Angelo and Jacob 2012). This will include the digitisation of morphology from fixed slice preparation slides; the fitting of passive parameters to multiple dual patch clamp recordings; and handling of ion channels in Genesis, Neuron and the move towards standardisation within the modelling community.


5th Dec: Tom Shimizu, FOM Institute AMOLF, Amsterdam

What can bacteria teach us about the motility of nematodes?

Motility provides a rich yet well-defined set of problems for studying the physiological bases of behavior. Given the inherently uncertain nature of natural environments, tasks such as ‘exploration’ and ‘exploitation’ of resources resemble computational search/optimization problems, and as such involve the generation and tuning of random variables. We know very little about how organisms with a nervous system generate and modulate random behavior, but much has been learned in recent years about how bacteria achieve this to optimize behavioral performance. In this talk, I will review some of the highlights in the bacterial arena, emphasizing what we have learned about their motile strategy and their mechanistic implementation in these simple cells. I will conclude with our nascent efforts to frame nematode motility as a similar problem – of generating and biasing a random walk.


12th Dec: Tim Vogels (Oxford)

The dance of excitation and inhibition (and some other interesting stories)

The first part of my talk will investigate the electrical filtering abilities of dendritic spine necks.
Most excitatory inputs in the mammalian brain are made on dendritic spines. Spines are thought to compartmentalize calcium gradients, and have been hypothesized to serve also as electrical compartments. The latter hypothesis necessitates relatively high spine neck resistances.  Due to its small size it is difficult to assess spine neck resistance directly in experiments, and has thus been discussed at somewhat above average temperatures in the field. I will show some modeling work that aims to deduce resistance estimates from two recent datasets showing negative correlations between spine neck length and somatically recorded EPSPs, and thus seem to imply high spine neck resistance. Using numerical simulations, we explore the parameter regimes for the spine neck resistance and other variables (such as synaptic conductance changes) necessary to explain these data sets. Since we use NEURON for the above mentioned simulations, and NEURON, and its accompanying MODELDB database for previously published models is (sometimes) notoriously difficult to get comfortable with,  I will show some recent meta-analysis on publicly available model ion channels. Our work visualizes the family relations of over 3000 unique models, and uses the similarity of their performance in standard protocols as a means to suggest a handful truly useful channels for everyday use.
The last part of my talk will visit the stabilizing performance of inhibitory synaptic plasticity in recurrent cortical networks and introduce a class of cortical architectures with very strong and random excitatory recurrence that is stabilized by intricate, fine-tuned inhibition. I will show that excitation and inhibition in such networks dance with each other to transiently amplify specific activity states that can be used to reliably execute multidimensional movement patterns. The intriguing similarity to recent experimental observations along with tightly balanced excitation and inhibition, suggest inhibitory control of complex excitatory recurrence as a generic organizational principle in cortex.

Leave a Reply

Your email address will not be published. Required fields are marked *