Dr Stefano Panzeri

The overall aim of the research of the Laboratory of Neuroinformatics is to understand how sensory information is represented and transmitted in the central nervous system. We pursue this issue by developing new quantitative data analysis techniques based on the principles of Information Theory and by developing computational models of neural network function.

From the very early stages of sensory processing, signals from the external world, such as the presence of a face or an object in the visual field, are converted into a series of electrical pulses (called "spikes") emitted by individual neurons. At the cortical level, the neuronal population activated by the presentation of even the simplest stimulus can be rather large, involving several brain areas and thousands of neurons. Moreover, each neuron might emit several spikes over time in response to the stimulus. Therefore the "code" used by the cerebral cortex to represent the environment can be extremely complex. Nevertheless, the brain can interpret this electrical activity correctly and quickly every time a new stimulus is presented. Our major aim is to investigate and elucidate how the neuronal cortical code for sensory features originates, and how the brain is able to decode the sensory information embedded in the electrical activity of other neurons. The specific questions we try to address are:

• How does the brain decode sensory information from single stimulus presentation?
• What is the functional neuronal architecture underlying the neuronal code for sensory functions?
• Information theory and neuronal population coding

One way to understand the aspects of neuronal population responses that underlie rapid and reliable representation of sensory information is to think of the nervous system as a communication channel. Shannon's Information Theory can then be used to quantify how much information neuronal population responses carry about external stimuli. Shannon's information quantifies the stimulus discriminability achieved from a single observation of the neuronal responses. The information quantified in this way is thus accessible to "downstream" neurons that have to interpret pre-synaptic neuronal activity online.

Part of the work in our lab has focused on developing a method to break down the "code" into its elementary components, by quantifying separately the information contributed by the various "symbols" making up the population code [Neural Computation 2001; Physical Review Letters 2001; Network 2003]. An analysis into how neuronal populations transmit information about stimulus location in rat somatosensory cortex revealed that all information available from the population spike train is decodable in a very simple way by just considering the timing of the first spike emitted after the occurrence of a stimulus [Neuron 2001]. Thus, despite the potentially enormous complexity of the cortical code, most sensory information can be extracted by very simple and fast decoding rules based on considering only a few spikes.

Our laboratory is currently working to extend these analyses to more complex and naturalistic stimulus presentation protocols, and to incorporate these results into computational models of sensory processing. We are also trying to extend these "measures" of the neuronal code to data recorded with non-invasive techniques, such as Functional Imaging, with the aim of understanding information processing in the human brain [Neuroimage 2004].

An essential part of this work is the collaboration with neurophysiological laboratories that provide data and ask interesting neurophysiological questions. My main long-term collaborators in this work are Rasmus Petersen at Machester (somatosensory coding, natural stimuli), Mathew Diamond at SISSA (somatosensory coding), Huw Golledge and Paul Flecknell at Newcastle (neuronal correlates of pain, anaesthesia and coding), Alex Thiele at Newcastle (parallel recordings and functional imaging in the visual system), and Enrico Bracci at Manchester (cortico-striatal information processing).

Neuronal network models and the functional architecture of cortical and subcortical networks. The determination of the functional architecture underlying certain sensory functions is of fundamental importance for our understanding of how sensory processing is implemented by the brain. This knowledge has potentially far reaching practical implications for clinical and neuro-engineering applications, such as neuro-prosthetics. The wealth of anatomical, physiological and biophysical data that is being acquired can potentially lead to constrain the functional architecture of cortical and subcortical structures. However, given the intrinsic complexity and diversity of the data, it remains difficult to provide a comprehensive framework to bring these data together to characterize structure-function relationships. Our laboratory tries to address this problem by using biologically plausible neuronal network models as a tool to bring together anatomy and physiology.

In collaboration with the neurophysiolocial laboratory of Dr. Bracci at Manchester, we are currently addressing structure-function relationships in the cortico-striatal system. We are building detailed models of the network of striatal projection neurons that take into account the main biophysical properties of the individual neurons and of their lateral GABA-ergic connections, as well as the nature and topography of the cortical input. By exploring and manipulating this simulated network, we investigate the relationship between GABAergic inhibition and cortical input and its impact on striatal processing of information of cortical origin. This project is being carried out in a close collaboration between theory and experiment. We are using data acquired in the laboratory of Dr. Bracci to constraint the network parameter, and the neuronal network predictions are also tested experimentally by Dr. Bracci.

In collaboration with Prof. Rolls (Oxford) and Prof. Young (Newcastle), we have previously used biophysically and anatomically constrained models of neuronal networks to explore the dynamics of information processing in the visual cortex [Network 2001, Neuroreport 2001].