The focus of my research is the modeling of biophysical properties of the cerebellar neurons. Computational models of neurons are mathematical descriptions used to describe and simulate biological processes of these cells and are a valuable means for the neuronal investigation. I will describe the conductance-based models, based on Hodgkin and Huxley (HH) mathematical description and spiking neuron models, based on integrate and fire (IF) theory. Then, I will focus on the models of the cerebellar neurons, being the area of interest of my research. The first part of the thesis examines the reconstruction of the multi-compartmental model of the cerebellar stellate cell (SC), modelled with the HH notation. The model reproduced the main electrophysiological properties of the SC, the gap junctions functioning and the synaptic activity, achieving satisfying results. However, the most advanced models, based on the HH theory, in order to reproduce the correct neuronal dynamics, require the fine-tuning of a large number of parameters, which are the maximum ionic conductance (Gi-max). The definition of appropriate Gi-max values, for each section of the morphology, is a complex task: they cannot be experimentally determined, so they must be assigned during the modeling phase and accurately validated. The manual calibration of Gi-max parameters for each channel is also a time-consuming and error-prone task. In order to explore the quite extensive parameter space of single-neuron models, it is possible to exploit numerical optimization techniques, able to, automatically, estimate the most fitting Gi-max values, obtaining neuron models that can reproduce the expected electro-physiological behavior, compared with the experimental results. Therefore, the second part of the thesis investigates an approach based on an automatic parameter estimation (PE) method that exploits the swarm intelligence (SI) technique known as particle swarm optimization (PSO). The investigation of this complex problem is made possible by the increase of the computational power and the high performing computing (HPC) techniques that allows scientist to develop specific procedures to calibrate automatically the parameter tuning. PSO was applied to the cerebellar SC neuron model for the first time and choosing the correct fitness functions, to quantify how well the optimization solutions compare with the target traces. The methodology proposed relies on the execution of a massive number of simulations, whose computational costs are relevant. To reduce the overall running time, the methodology was implemented on a parallel architecture. In this way, the fitness evaluations were accelerated with a computer cluster. The third part of the thesis reviews a challenging area of neuronal development, focusing on the cerebellar neuronal development. The cerebellar development is a complex biological process that requires a huge interaction between biochemical and biophysical mechanisms. Many aspects of development, such as neurons differentiation, proliferation and migration, axon and dendritic growth, synapses formation and stabilization, were extensively described with experiments in both the brain and the cerebellum, but only some of these aspects were described with computational models. Many models describe specific neurogenesis and axonal connectivity in the cerebral cortex, hippocampus, olfactory bulb and spinal cord. The frameworks, used to build these models, describe specifically several characteristics that could be applied to the cerebellum. I will review the available models and the latest tools to model the development of the cerebellar network, creating a new framework, which will explain all the specific properties of the cerebellum.

La principale area della mia ricerca è la modellizzazione delle proprietà biofisiche dei neuroni del cervelletto. I modelli computazionali di neuroni sono descrizioni matematiche del loro funzionamento e sono importanti mezzi per lo studio di queste cellule. Saranno presentati i modelli che seguono la notazione matematica di Hodgkin e Huxley (HH) e i modelli descritti secondo la teoria degli Integrate and Fire e infine mi concentrerò sui modelli del cervelletto, essendo questa regione del sistema nervoso centrale il principale interesse della mia ricerca. La prima parte della tesi esamina la ricostruzione di un modello multi-compartimentale di singolo neurone, e in particolare verrà approfondito lo studio dettagliato della cellula stellata del cervelletto. Il modello si basa sulla notazione matematica di HH. Il modello riproduce le principali caratteristiche elettrofisiologiche del neurone, il funzionamento delle gap junctions e la trasmissione sinpatica, raggiungendo risultati soddisfacenti. Tuttavia, i più avanzati modelli descritti secondo HH richiedono la calibrazione di un vasto numero di parametri, che sono le conduttaze ioniche massime (Gi-max). L’assegnazione di un appropriato valore a ciascuna Gi-max, presenti nelle diverse sezioni della morfologia del modello, è un compito molto complesso. Dal momento che non possono essere determinate sperimentalmente, devono essere assegnate durante la fase di creazone del modello e accuratamente validate. La calibrazione manuale di questi parametri richiede una grande quantità di tempo e un elevato rischio di errore durante la procedura, per questo motivo sono stati studiati metodi automatici per la stima di parametri in grado di essere applicati per lo studio e la calibrazione delle Gi-max neuronali, ottenendo modelli ottimizzati in grado di riprodurre risutati ben comparabili con i risultati sperimentali. La seconda parte del lavoro, approfondisce un metodo di stima di parametri che si basa sulle tecniche di Swarm Intelligence. L’algoritmo particle swarm optimization (PSO) è appliacto per la prima volta per la calibrazione dei valori di Gi-max, portando ottimi risultati. Lo studio di questi problemi complessi è anche possibile con la parallela crescita di tecniche di high performing computing (HPC), in grado di parallelizzare i calcoli su macchine parallele e distribuite, in modo da diminuire i costi computazionali richiesti per le simuazioni. La metodologia è stata quindi implementata per potere essere usata con macchine parallele, come ad esempio computer cluster. La terza parte del progetto prevede invece lo studio di un’importante area delle neuroscienze che si occupa dello sviluppo neurnale, dagli stadi pre-natali ai post-natali. Mi concentrerò sullo sviluppo del cervelletto e sulla creazione di un framework teorico per lo sviluppo di un modello in grado di ricostruire le diverse fasi dello sviluppo. Prenderò in rassegna i diversi modelli esistenti, che modellizzazone singoli processi, com ad esempio il differenziamento neuronale e la crescita degli assoni e dei dendriti durante lo sviluppo, e gli strumenti utilizzati per la loro modellizzazione, valutando l’applicabilità per ricostruire l’intero sviluppo cerebellare.

(2017). Parameter estimation of cerebellar stellate neuron model. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2017).

Parameter estimation of cerebellar stellate neuron model

RIZZA, MARTINA FRANCESCA
2017

Abstract

The focus of my research is the modeling of biophysical properties of the cerebellar neurons. Computational models of neurons are mathematical descriptions used to describe and simulate biological processes of these cells and are a valuable means for the neuronal investigation. I will describe the conductance-based models, based on Hodgkin and Huxley (HH) mathematical description and spiking neuron models, based on integrate and fire (IF) theory. Then, I will focus on the models of the cerebellar neurons, being the area of interest of my research. The first part of the thesis examines the reconstruction of the multi-compartmental model of the cerebellar stellate cell (SC), modelled with the HH notation. The model reproduced the main electrophysiological properties of the SC, the gap junctions functioning and the synaptic activity, achieving satisfying results. However, the most advanced models, based on the HH theory, in order to reproduce the correct neuronal dynamics, require the fine-tuning of a large number of parameters, which are the maximum ionic conductance (Gi-max). The definition of appropriate Gi-max values, for each section of the morphology, is a complex task: they cannot be experimentally determined, so they must be assigned during the modeling phase and accurately validated. The manual calibration of Gi-max parameters for each channel is also a time-consuming and error-prone task. In order to explore the quite extensive parameter space of single-neuron models, it is possible to exploit numerical optimization techniques, able to, automatically, estimate the most fitting Gi-max values, obtaining neuron models that can reproduce the expected electro-physiological behavior, compared with the experimental results. Therefore, the second part of the thesis investigates an approach based on an automatic parameter estimation (PE) method that exploits the swarm intelligence (SI) technique known as particle swarm optimization (PSO). The investigation of this complex problem is made possible by the increase of the computational power and the high performing computing (HPC) techniques that allows scientist to develop specific procedures to calibrate automatically the parameter tuning. PSO was applied to the cerebellar SC neuron model for the first time and choosing the correct fitness functions, to quantify how well the optimization solutions compare with the target traces. The methodology proposed relies on the execution of a massive number of simulations, whose computational costs are relevant. To reduce the overall running time, the methodology was implemented on a parallel architecture. In this way, the fitness evaluations were accelerated with a computer cluster. The third part of the thesis reviews a challenging area of neuronal development, focusing on the cerebellar neuronal development. The cerebellar development is a complex biological process that requires a huge interaction between biochemical and biophysical mechanisms. Many aspects of development, such as neurons differentiation, proliferation and migration, axon and dendritic growth, synapses formation and stabilization, were extensively described with experiments in both the brain and the cerebellum, but only some of these aspects were described with computational models. Many models describe specific neurogenesis and axonal connectivity in the cerebral cortex, hippocampus, olfactory bulb and spinal cord. The frameworks, used to build these models, describe specifically several characteristics that could be applied to the cerebellum. I will review the available models and the latest tools to model the development of the cerebellar network, creating a new framework, which will explain all the specific properties of the cerebellum.
ANTONIOTTI, MARCO
D'ANGELO, EGIDIO
Stellate; models; particle; swarm; optimization
Stellate; models; particle; swarm; optimization
INF/01 - INFORMATICA
English
16-nov-2017
INFORMATICA - 87R
29
2015/2016
open
(2017). Parameter estimation of cerebellar stellate neuron model. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2017).
File in questo prodotto:
File Dimensione Formato  
phd_unimib_068430.pdf

accesso aperto

Descrizione: tesi di dottorato
Tipologia di allegato: Doctoral thesis
Dimensione 7.94 MB
Formato Adobe PDF
7.94 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/180709
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact