A Study On Neural Ensemble Sparse Coding For Working Memory Event In Rat Prefrontal Cortex

ObjectiveNervous system codes information in the activities of neural ensemble consisting of several neurons. It is significant to expressing neural ensemble coding external stimuli effectively in the heart of neuroscience. In this paper, we investigate two different neural ensemble sparse coding to address this issue by recording the activities of neurons in the rat prefrontal cortex during the working memory task:(1) neural ensemble rate sparse coding; (2) neural ensemble entropy sparse coding. And this would be helpful to analyze the neural coding mechanisms of working memory.Methods1. Experimental dataExperimental data come from Neural Engineering lab in Tianjin Medical University.16 channels microelectrode array were planted in prefrontal cortex of 3 SD rats.50 raw data were collected during working memory task in Y-maze based on multi channel recording technology. The firing times of individual neurons were determined using spike detection and sorting algorithm. Effective period of 7 seconds were selected, enough to represent the entire working memory proceeding.2. Neural ensemble sparse coding(1) Continuous process for discrete point potential:as the neuronal activity is discrete, how to translate the neuronal activity into the input data of sparse coding is one of the key points in this paper. We use two methodologies to solve the problem:①Neural ensemble rate sparse codingPoint potential is quantified based neural firing rate in selected window. From the zero point, the neuronal firing times are binned in windows of 200 ms with 50 ms overlapping. Then the neuronal firing rate is normalized during the whole session.②Neural ensemble entropy sparse codingPoint potential is quantified based neural firing entropy in selected window. The neural firing entropy is obtained in the 200 ms window with 50 ms overlapping. Then the neuronal firing entropy is normalized during the whole session.(2) Non-negative sparse coding of the neuronal activity matrix:we set the number of the sources by hand by repeated experiments; the neuronal activity matrix is decomposed by NNSC into overcomplete weighted matrix and source component matrix with non-negative constraints.(3) Selection of sparse feature component:the components, in which the coefficients increase suddenly during the time according to neural ensemble, are extracted as feature components.(4) Sparse reconstruction and coding of the input neuronal activity:the neural population firing dynamic spatiotemporal mapping is obtained by an inverse of sparsifying transform of sparse feature components.ResultsIn this article, we study the neural ensemble sparse coding in rat prefrontal lobe during the working memory task. The neuron number corresponding to neural ensemble and the time neural ensemble lasts averagely are as follows:1. Neural ensemble rate sparse coding(1) Rat-1①Rate sparse coding:9,10,11 (removing 2 neurons), (0.8480±0.2897)s.②Rate coding:9,10,11 (removing 6 neurons), (2.9780±1.0031)s.The neural ensemble via rate sparse coding lasts longer than that via rate coding and T test shows remarkably difference (p<0.01).(2) Rat-2①Rate sparse coding:day 1,7,8 (removing 1 neuron), (0.9540±0.3736)s; day 2, 7,8,9 (removing 2 neurons), (1.2180±0.5966)s.②Rate coding:day 1,7,8,9 (removing 3 neurons), (2.7210±1.1983)s; day 2, 7,8,9,10 (removing 2 neurons), (3.4380±0.7899)s.The neural ensemble via rate sparse coding lasts longer than that via rate coding and T test shows remarkably difference (p1<0.01; p2<0.01).(3) Rat-3①Rate sparse coding:day 1,13,14,15 (removing 3 neurons), (1.1890±0.4418)s; day 2,5,6,7 (removing 2 neurons), (1.1700±0.3348)s.②Rate coding:day 1,13,14,15,17 (removing 8 neurons), (2.7560±1.1124)s; day 2, 5,6,7,8 (removing 10 neurons), (3.2150±0.8827)s. 2. Neural ensemble entropy sparse coding(1) Rat-1①Entropy sparse coding:9,10,11 (removing 2 neurons), (0.8478±0.2896)s.②Entropy coding:9,10,11 (removing 5 neurons), (2.9781±1.0031)s.The neural ensemble via entropy sparse coding lasts longer than that via entropy coding and T test shows remarkably difference (p<0.01).(2) Rat-2①Entropy sparse coding:day 1,7,8 (removing 1 neuron), (0.9540±0.3736)s; day 2,7,8,9 (removing 2 neurons), (1.2179±0.5965)s.②Entropy coding:day 1,7,8,9 (removing 3 neurons), (2.7209±1.1982)s; day 2, 7,8,9,10 (removing 2 neurons), (3.4380±0.7899)s.The neural ensemble via entropy sparse coding lasts longer than that via entropy coding and T test shows remarkably difference (p1<0.01; p2<0.01).(3) Rat-3①Entropy sparse coding:day 1,13,14,15 (removing 2 neurons), (1.1890±0.4418)s; day 2,5,6,7 (removing 3 neurons), (1.1701±0.3348)s.②Entropy coding:day 1,13,14,15,17 (removing 11 neurons), (2.7558±1.1125)s; day 2,5,6,7,8 (removing 8 neurons), (3.2151±0.8828)s.The neural ensemble via entropy sparse coding lasts longer than that via entropy coding and T test shows remarkably difference (p1<0.01; p2<0.01).Conclusions(1) Comparing neural ensemble rate coding, neural ensemble rate sparse coding can express neural ensemble coding working memory event more effectively.(2) Comparing neural ensemble entropy coding, neural ensemble entropy sparse coding can express neural ensemble coding working memory event more effectively.(3) Neural ensemble rate sparse and neural ensemble entropy sparse coding express the neural ensemble coding pattern effectively from two different aspects:neural firing rate and firing probability.