Multi-Electrode Array (MEA) neural activity recording has become a consolidated technique in studying in vivo and in vitro neural networks. The analysis of external electrical activity of neurons is key in understanding the neurophysiology of neurodegenerative diseases. Several studies have identified the G2019S mutation of the LRRK2 gene in sporadic and familial forms of Parkinson’s disease. Therefore, the study of LRRK2- mutated neuron’s behavior could lead to innovative clinical and diagnostic approaches. In this thesis’ work, structured multi-nodal human neural networks carrying the G2019S mutation have been grown, and they are compared with control in vitro neural networks by using custom-designed microfluidic chips coupled to MEA. The electrode recordings are digitized and processed by a spike sorting algorithm specifically tailored to suit the electrophysiological data and correctly extract the single neuron’s behavior. To this extent, a specific clustering algorithm (DBSCAN) has been chosen to automatically detect the number of neurons recorded by each electrode and remove noisy spikes. Spike trains have been modeled with a bayesian Dirichlet mixture point process model in order to describe the statistical behavior of neurons and goodness-of-fit of the devised models were assessed by a Kolmogorv- Smirnov test. Model’s features were estimated with Hamiltonian Markov Chains, which highlight a good convergence of all the features. Finally, a wide range of classification machine learning models have been devised and compared in order to distinguish healthy neurons from LRRK2-mutated ones. Among all, the Random Forest model reported the highest level of accuracy both in the two class in the Baseline (93.10) and in the four-class After 24 hours phase (68.60), with ROC areas under the curve of 92.37 forBaseline and 83.73 After 24 hours respectively. In this thesis’ work, an innovative approach in the study of in-vitro neurons was presented. Based on these conclusions, subsequent studies could pave the way for a deeper understanding of the influence of the G2019S mutation in Parkinson’s disease.
La registrazione dell'attività neurale Multi-Electrode Array (MEA) è diventata una tecnica consolidata nello studio delle reti neurali in vivo e in vitro. L'analisi dell'attività elettrica dei neuroni è fondamentale per comprendere la neurofisiologia delle malattie neurodegenerative. Diversi studi hanno identificato la presenza della mutazione G2019S del gene LRRK2 in forme sporadiche e familiari del morbo di Parkinson. Di conseguenza, lo studio del comportamento del neurone con mutazione del gene LRRK2 potrebbe portare alla scoperta di nuovi approcci clinici e diagnostici. In questo lavoro di tesi, reti neurali umane strutturate con mutazione G2019S sono state analizzate e confrontate con le reti neurali controllo in vitro utilizzando microfluidic chips progettati su misura per i MEA. Le registrazioni degli elettrodi sono state digitalizzate ed elaborate da un algoritmo di spike sorting, specificamente adattato ai dati elettrofisiologici al fine di estrarre correttamente il comportamento del singolo neurone. A tal fine, è stato scelto uno specifico algoritmo di clustering (DBSCAN) per rilevare automaticamente il numero di neuroni registrati da ciascun elettrodo e rimuovere gli spike non rilevati correttamente. L’attività elettrica è stata modellata con un modello bayesiano di point process basato su una mistura di Dirichlet per descrivere il comportamento statistico dei neuroni. Inoltre, la bontà di moderazione è stata valutata con il test di Kolmogorv- Smirnov. Le variabili del modello sono state stimate con Hamiltonian Markov Chains, che evidenziano una buona convergenza di tutte le caratteristiche. Infine, un'ampia gamma di modelli di machine learning di classificazione è stata valutata al fine di distinguere i neuroni sani da quelli portatori della mutazione LRRK2. Tra tutti, il modello Random Forest ha riportato il più alto livello di accuratezza sia nella classificazione binaria nella fase di Baseline (93,10) sia nella classificazione 4-classi della fase After 24 hours(68,60), con aree ROC sotto la curva di 92,37 per Baseline e 83,73 After 24 hours rispettivamente. In questa tesi è stato presentato un approccio innovativo nello studio dei neuroni in vitro. Sulla base di queste conclusioni, studi successivi potrebbero aprire la strada a una comprensione più profonda dell'influenza della mutazione G2019S nella malattia di Parkinson.
Characterization of spiking activity in structured in-vitro human neural networks through point process statistical modelling approaches
Levi, Riccardo
2019/2020
Abstract
Multi-Electrode Array (MEA) neural activity recording has become a consolidated technique in studying in vivo and in vitro neural networks. The analysis of external electrical activity of neurons is key in understanding the neurophysiology of neurodegenerative diseases. Several studies have identified the G2019S mutation of the LRRK2 gene in sporadic and familial forms of Parkinson’s disease. Therefore, the study of LRRK2- mutated neuron’s behavior could lead to innovative clinical and diagnostic approaches. In this thesis’ work, structured multi-nodal human neural networks carrying the G2019S mutation have been grown, and they are compared with control in vitro neural networks by using custom-designed microfluidic chips coupled to MEA. The electrode recordings are digitized and processed by a spike sorting algorithm specifically tailored to suit the electrophysiological data and correctly extract the single neuron’s behavior. To this extent, a specific clustering algorithm (DBSCAN) has been chosen to automatically detect the number of neurons recorded by each electrode and remove noisy spikes. Spike trains have been modeled with a bayesian Dirichlet mixture point process model in order to describe the statistical behavior of neurons and goodness-of-fit of the devised models were assessed by a Kolmogorv- Smirnov test. Model’s features were estimated with Hamiltonian Markov Chains, which highlight a good convergence of all the features. Finally, a wide range of classification machine learning models have been devised and compared in order to distinguish healthy neurons from LRRK2-mutated ones. Among all, the Random Forest model reported the highest level of accuracy both in the two class in the Baseline (93.10) and in the four-class After 24 hours phase (68.60), with ROC areas under the curve of 92.37 forBaseline and 83.73 After 24 hours respectively. In this thesis’ work, an innovative approach in the study of in-vitro neurons was presented. Based on these conclusions, subsequent studies could pave the way for a deeper understanding of the influence of the G2019S mutation in Parkinson’s disease.File | Dimensione | Formato | |
---|---|---|---|
Msc_Thesis_Levi_2020_10.pdf
solo utenti autorizzati dal 15/09/2021
Dimensione
5.34 MB
Formato
Adobe PDF
|
5.34 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/169848