Memorias de investigación
Statistical and optimization methods for spatial data analysis applied to neuroscience

Research Areas
  • Artificial intelligence,
  • Statistics

Neuroscience has undergone great development in recent decades, making it one of the most relevant biomedical disciplines today. The development of new technologies and in particular the recent technical advances in microscopy make it possible to have a great amount of data that collect the nature and the spatial distribution of some neuronal elements that form the brain. In the current state of development of neuroscience, the need of new computational techniques is becoming more evident, and in this thesis it is carried out developing statistical and optimization methods for data analysis giving explicit consideration to spatial characteristics such as location, spatial organization or distance between elements. The work developed in this thesis is mainly applied to the study of neuronal morphology. Despite the numerous efforts to better understand the brain, current knowledge about the neuron structure is still incomplete. Neuronal morphology reflects the organization of synaptic inputs and the way in which a neuron expands plays an important role in its functional and computational characteristics. Therefore, taking into account the inherent spatiality in neuronal morphology, key features can be revealed in the design of brain circuits. This thesis focuses on the modeling of the spatial distribution of different neuronal structures in order to discover specific patterns and rules in their spatial organizations. To do this, we develop spatial point process methods for 3D spatial modeling, in particular, using replicated point patterns. In addition, considering neuronal arborizations as networks connecting the points where the synapses are located, we use graph theory and evolutionary computational techniques with a reverse engineering approach, to analyze if these networks follow principles of optimality in their design. Regarding spatial point processes, the 3D spatial distribution of synapses is modeled in the six layers of the rat somatosensory cortex. Because several samples are available from each layer, replicated spatial patterns are used to detect similarities and differences between layers. Then, the existing 2D methodology for network spatial analysis is extended to 3D space. In addition, replicated spatial patterns are applied for the first time in this context. These methods are used to model the distribution of spines along the dendritic arborizations of human pyramidal neurons in both basal and apical dendrites. Next, the hypothesis of optimal wiring in neuronal circuits is used in conjunction with the analysis of the spatial distribution of branching and terminal points of dendritic arbors, using a measure related to the distance to the nearest neighbour to quantify how a set of points are distributed in space. Regarding network optimization, a new way of representing and solving the structural constraints that commonly limit network design problems is proposed, namely, restrictions on the maximum number of edges incident on a node and establishing a priori the roles of the nodes in the network (root, intermediate or leaf node). Then, using graph theory and the proposed representation it is analyzed if individual neurons optimize brain connectivity in terms of wiring length. The analysis is carried out to the dendritic and axonal wiring of interneurons with very different morphology and to the dendritic wiring of a homogeneous population of pyramidal neurons, also studying in the latter case if there are differences between cortical layers.
Mark Rating
Sobresaliente cum laude

Research Group, Departaments and Institutes related
  • Creador: Grupo de Investigación: Computational Intelligence Group
  • Departamento: Inteligencia Artificial