Assessing the impact of auto-correlations on the information represented by neurons in the central nervous system

To understand how the brain processes and represents information is a fundamental question in neuroscience. A complete understanding of this phenomenon would provide important insights into the mechanisms underlying neurological disorders and diseases. However, our understanding of how to decode neural signals is limited, in part due to our lack of understanding of the basic mechanisms used to encode the information. In the last century, neuroscientists have established that information is conveyed in the brain by activity of cells in the central nervous system. At the same time that investigators have been elucidating the mechanisms used by the brain to represent information, there has been an implicit assumption that variability present in the responses of the neurons is noise that does not represent information. Recently it has been shown that a stimulus can also alter the variability of firing rate, thus suggesting that this variability could contribute to the information about the stimulus. Therefore, there was a critical need to develop a framework that could rigorously measure the contribution of variability to the total information conveyed by neurons in response to a stimulus. In this work, a general formalism based on information theory has been developed that allows a rigorous quantification of the impact of variability, as measured by auto-correlations, on the representation of information. To further establish a role for auto-correlations in conveying information, the approach developed in this work was used to study the role of auto-correlations in the information encoded by cells about stimulus location. The results from this study demonstrated that up to 30% of the information conveyed by cells is carried by auto-correlations and this contribution is dependent on the time-scale of the response window. The results of this work show that auto-correlations play a major role in representing information and, therefore, variability can be used by the nervous system to convey information. This result is critically important to our understanding of the transformation of information from the impact of the stimulus on mechanoreceptors in the periphery to the cortex. Moreover, the tools derived in this work provide a general framework that can be used to develop decoding algorithms for brain machine interfaces.