An information-theoretic study of neuronal spike correlations in the mammalian cerebral cortex

• Autores: Montani, Fernando Fabián
• Año de publicación: 2008
• Idioma: inglés
• Tipo de recurso: tesis doctoral
• Colaborador/a o director/a de tesis: Schultz, SImon R.
• Materia: Informática; Neural Coding; Computational Neuroscience; Neuronal Dynamics; Neurophysiology; Visual Cortex; Data Mining; Neuronal Networks; spiking neural networks; higher-order correlations; Information Geometry
• Institución: Universidad Nacional de La Plata

In chapter I of this thesis we present a review of the historical background of the previousspike correlation studies and current state of the problem. In the chapters II, III and IV ofthis thesis we have applied an information theoretic approach to study the role of correlationsin the neuronal code, using the responses of pairs of neurons to drifting sinusoidal gratingsof different orientations and contrasts recorded in the primary visual cortex of anesthetizedmacaque monkeys. In chapter V we investigate the effects of a focal stroke in a populationof neurons on information transmission using a computational and analytical approach tothe problem. Finally, in chapter VI we use a novel analytical approach to study effects ofhigher order correlations in a population of neurons.It has been proposed in neuroscientific literature that pooling can lead to a significant improvementin signal reliability, provided that the neurons being pooled are at most weaklycross-correlated. We have computed mutual information, and compared the informationavailable from pairs of cells with the sum of the single cell information values. This allowedus to assess the degree of synergy (or conversely, redundancy) in the coding. In chapter IIof this thesis, we show that due to a loss of information encoded in the neuronal identity ofthe cells, pooling spikes across neurons leads to a loss of a large fraction of the informationpresent in their spike trains.We have used information theory to examine whether stimulus-dependent correlation couldcontribute to the neural coding of orientation and contrast by pairs of V1 cells. To this end,in chapter III, we have used a modified version of the method of information components.This analysis revealed that although synchrony is prevalent and informative, the additionalinformation it provides is frequently offset by the redundancy arising from the similar tuningproperties of the two cells. Thus, coding is roughly independent with weak synergy orredundancy arising depending on the similarity in tuning and the temporal precision of theanalysis. Our findings suggest that this would allow cortical circuits to enjoy the stabilityprovided by having similarly tuned neurons without suffering the penalty of redundancyas the associated information transmission deficit is compensated by stimulus dependentsynchrony.In chapter IV, we present a discussion about different measures of correlations and in particularwe propose the Jensen-Shannon Divergence as a measure of the distance between thecorresponding probability distribution functions associated with each spikes fired observedpatterns. We applied this Divergence for fixed stimuli as a measure of discrimination betweencorrelated and independent firing of pairs of cells in the primary visual cortex. Thisprovides a new, information-theoretic measure of the strength of correlation. We found thatthe relative Jensen-Shannon Divergence (measured in relation to the case in which all cellsfired completely independently) decreases with respect to the difference in orientation preferencebetween the receptive field from each pair of cells. Our finding indicates that theJensen-Shannon Divergence can be used for characterizing the effective circuitry network ina population of neurons.The underlying origins of synchronized firing between cortical neurons are still under discussion.Inter-cellular communication through chemically mediated synaptic transmissionis considered a major contributor to the formation of neuronal synchrony. GABAergic inhibitoryneurons may be involved in the generation of oscillatory activity in the cortex andits synchronization. Specifically, reduction of GABAergic inhibition may favour corticalplasticity producing functional recovery following focal brain lesions. Research into neurotransmittersystems is therefore of paramount importance to understand the origins ofsynchronized spiking. However, it is necessary to understand first how simple focal abnormalitiesin GABAergic modulators can affect the information transmission in an impairedbrain tissue. In chapter V, we present a computational and analytical model of a topographicallymapped population code which includes a focal lesion as well as a process for receptivefield enlargement (plasticity). The model simulates the recovery processes in the brain, andallows us to investigate mechanisms which increase the ability of the cortex to restore lostbrain functions. We have estimated the Fisher Information carried by the topographic mapbefore and after the stroke. Our finding shows that by tuning the receptive field plasticity toa certain value, the information transfer through the cortex after stroke can be optimized.A widespread distribution of neuronal activity can generate higher-order stochastic interactions.In this case, pair-wise correlations do not uniquely determine synchronizing spiking ina population of neurons, and higher order interactions across neurons cannot be disregarded.We present a new statistical approach, using the information geometry framework, for analyzingthe probability distribution function (PDF) of spike firing patterns by consideringhigher order correlations in a neuronal pool. In chapter VI, we have studied the limit ofa large population of neurons and associated a deformation parameter to the higher ordercorrelations in the PDF. We have also performed an analytical estimation of the Fisher informationin order to evaluate the implications of higher order correlations between spikeson information transmission. This leads to a new procedure to study higher order stochasticinteractions.The overall findings of this thesis warn about making any extensive statement about therole of neuronal spike correlations without considering the general case inclusive of higherorder correlations, and suggest a need to reshape the current debate about the role of spikecorrelations across neurons.
PhD in Computational Neuroscience
Imperial College London
Imperial College London