Neuronal spike-trains and various other signals in the central nervous system have a discrete, impulsive nature that is well characterized with point process statistical models. In several neuroscience applications, such impulsive signals are transformed upon interaction with biological processes or measurement artifacts, and are consequently observed as filtered point process data. The goal of this project is to develop a principled statistical signal processing framework for filtered point processes with models and algorithms for estimation and inference, and to apply these novel methodologies to experimental data from rodent brain calcium imaging data and human neuroendocrine data. Our approach centers on a unified framework for sparse representation and dynamical systems modeling of marked point process data arising in neuroscience analyses. In addition to its novel statistical methodology, another major strength of our proposal is the application of these methods to experimental data arising in fundamental neuroscience and clinical problems, both to validate the new methods with real data and to investigate basic science questions related to the central nervous system structural and functional organization. Large-scale two-photon calcium imaging, in conjunction with spike-train deconvolution, will allow us to study the activity of over a thousand identified neurons simultaneously with single-spike resolution in a behaving animal. This will allow us to elucidate with high accuracy how the magnitude and spatial structure of signal and noise correlations across neurons vary with stimuli or behavioral tasks. It will shed light on visual encoding in the rodent brain, and neuronal architectures underlying visual perception and cognition, at an unprecedented spatiotemporal scale. Further, our modeling of pulsatile hormone secretion will apply to the release of cortisol, gonadal steroids, insulin, thyroid and growth hormones. Diseases linked to abnormal cortisol secretion include diabetes, visceral obesity and osteoporosis, disturbed memory formation and life-threatening Addisonian crisis. Hence, understanding and modeling the underlying impulsive nature of normal hormone release will aid our understanding of pathological neuroendocrine states and improve the efficacy of drugs and other interventions for treatment of hormonal disorders. Additionally, this project will combine Brown Lab?s computational expertise in point process models with Sur Lab?s experimental expertise in neuronal calcium imaging, extending our ongoing collaboration under the NIH Brain Initiative to developing novel neural population analysis techniques with unprecedented detail at single-neuron, single-spike resolution. Our research is well poised to improve significantly the state of the art and in computational and systems neuroscience tools and bridge together components from the statistical learning, signal processing and computational neuroscience communities to produce a unifying analytical framework for neural data analysis.
In this project we will develop a filtered point process framework for neural data analysis, with models and algorithms for sparse recovery of spike-trains from time-series data comprising a series of pulsed waveforms. We will use this novel methodology to extract neuronal spike-trains from calcium imaging data of cortical tissue, and to estimate hormone secretion times and amounts from serum hormone levels. We will make our mathematical algorithms for neural data analysis and their software implementation freely and easily available to the neuroscience research community.
