Vertebrate neuronal networks: efficient coding and compressibility of interactions in the retina and the hippocampus

• Autor(a) principal: Luisa Fernanda Ramirez Ochoa
• Data de Publicação: 2022
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/57796; https://orcid.org/0000-0002-8052-3260
• topic: Neuronal code; Visual system; Retina; Efficient coding; Retina; Hipocampo

Our success in understanding the retina is partially due to its layered structure, that facilitates the study of circuit motifs and neuronal function. We can understand retinal information processing in three broad stages: I. encoding of light stimuli via electrical signals; II. signal processing by retinal circuits; and III. generation of the retinal code. In this thesis, I focus in the study of the first and third stages from an information-theory perspective. On the first stage, we investigate color coding in zebrafish retinal circuits based on recent experimental findings showing evidence of efficient coding. We propose a theoretical framework to study the encoding performance of different types of outer retinal networks, contrasting the role of excitation and inhibition. More specifically, we use a neuronal population model with chromatic stimulation to study the dynamical properties of such networks. Our findings suggest that inhibition plays a key role in encoding color information reliably, which is not guaranteed in networks with strong excitatory inter-cone couplings. Similarly, we find that networks optimized to encode aquatic spectral information are similar to that observed in zebrafish, providing more general understanding of zebrafish-like retinal circuits of color coding. These results provide quantitative evidence that the zebrafish retina is adapted to efficiently encode information from the environment, enhancing this animal's color vision capabilities. Studies in other species show that animals can adopt different strategies to improve color vision. For instance, in birds and turtles oil droplets serve as a filter to provide a plethora of distinguishable colors. Oil droplets and adapted retinal circuits have been investigated separately. Nevertheless, studies on the combination of both remain unknown. We implement a light transmission model of droplets to investigate the encoding performance of zebrafish-like retinal circuits exhibiting efficient coding. Our findings suggest that introducing droplets in a circuit for chromatic efficient coding creates a trade-off between coding efficiency and color-space gamut. That is, droplets decrease the network encoding performance while increasing the number of distinguishable colors. Regarding the third stage of processing, we focus on the theoretical study of neuronal interactions in the ganglion layer and their compressibility as a path to building simpler models of neuronal activity. Conventional models of neuronal activity introduce assumptions about neural interactions inspired in condensed-matter systems. But these models fail when the number of neurons increases, leading to an exponential explosion in the number of parameters. Here, we implement information theory and renormalization group ideas to explore efficient descriptions of neuronal activity. More specifically, we apply the compression-bottleneck formalism to a population of ganglion cells in the salamander retina. We find that compression leads to a vast simplification in the description of neuronal activity, outperforming conventional pairwise-interaction models. As a generalization, we implement this approach in a population of hippocampus neurons, yielding broadly similar results, suggesting that compressibility is a general feature of spiking neuronal networks.