Parkinson's disease (PD) is a debilitating neurodegenerative disorder whose etiology is strongly linked to the G2019S mutation of the LRRK2 gene - or Dardarin. Although this causal relationship is well established, the underlying mechanisms are not fully understood. In this thesis, we use Microelectrode Array (MEA) technology to analyze in-vitro neural networks, in order to delineate the characteristic neural activity patterns of carriers of the G2019S mutation of the LRRK2 gene, associated with PD pathogenesis. This study utilizes recordings from both G2019S-mutated cortical neural networks and from healthy control networks in two phases: before (baseline) and after Kainic Acid stimulation. The chemical stimulation process aims to induce neurotoxic effects, and by incorporating post-stimulation data into the analysis, to record changes in neural activity. The work focuses on the development of a three-phase algorithm capable of receiving raw MEA recordings and recognizing the population to which they belong, distinguishing between healthy and mutated activity. Initially, Spike Sorting techniques are used to identify individual neuronal activities. Subsequently, Point Process Modeling technique is employed to characterize the parameters of the ISI curve distribution, through Maximum A Posteriori Estimation. Finally, a classification phase is performed, where the model is trained with the parameters obtained from the modeling, and the data are categorized initially into mutated and control neurons, and subsequently, in a multi-class problem, both in terms of type of population and baseline vs. post-stimulation activity. The results obtained from binary classification demonstrate an excellent accuracy of 99% (mutated vs. control) using a Support Vector Machine model. Furthermore, in a more nuanced four-class classification scenario (mutated vs. control and pre- vs. post-stimulation), a Random Forest classifier achieves an accuracy of 77%, showcasing its efficacy in identifying both the genetic mutation and the stimulation-induced changes in neural activity. The research sheds light on the complex interplay between the LRRK2 mutation and neuronal activity dynamics, providing valuable insights into the pathophysiological mechanisms underlying Parkinson's disease, as confirmed by the obtained results. Moreover, the computational framework developed in this study offers a novel approach for studying neurodegenerative disorders at the cellular level, with implications for future diagnostic and therapeutic strategies.
Il morbo di Parkinson è un disturbo neurodegenerativo debilitante la cui eziologia è fortemente legata alla mutazione G2019S del gene LRRK2 (o Dardarina). Sebbene questa relazione causale sia ben stabilita, i meccanismi sottostanti non sono del tutto noti. In questa tesi, utilizziamo la tecnologia di Microelectrode Array (MEA) per analizzare reti neurali in vitro, al fine di delineare i modelli di attività neurale caratteristica dei portatori della mutazione G2019S del gene LRRK2, associata alla patogenesi del morbo. Questo studio utilizza registrazioni sia da reti neurali corticali portatrici della mutazione G2019S, che da reti di controllo sane e in due fasi: baseline e dopo una fase di stimolazione con Acido Kainico. Il processo di stimolazione chimica ha lo scopo di indurre effetti neurotossici, e incorporando nell'analisi i dati in post-stimolazione, si vuole osservare i cambiamenti nell'attività neurale. Il lavoro si concentra sullo sviluppo di un algoritmo a tre fasi, in grado di ricevere registrazioni MEA grezze e riconoscere la popolazione di appartenenza, distinguendo tra attività sana e mutata. Inizialmente, vengono utilizzate tecniche di Spike Sorting per identificare le attività neuronali individuali. Successivamente, viene impiegata la tecnica di Point Process Modeling per caratterizzare i parametri di distribuzione della curva ISI (intervalli temporali tra gli spike), tramite la Stima del Massimo A Posteriori. Infine, viene eseguita una fase di classificazione, in cui il modello viene addestrato con i parametri ottenuti dalla modellizzazione. Inizialmente, i dati vengono categorizzati in neuroni mutati e di controllo, e successivamente, in un problema a multi classe, sia in popolazione di appartenenza, che in attività di baseline e post-stimolazione. I risultati ottenuti dalla classificazione binaria dimostrano un'alta accuratezza del 99% (mutato vs. controllo) utilizzando un modello Support Vector Machine. Inoltre, in uno scenario di classificazione a quattro classi più complesso (mutato vs. controllo e pre- vs. post-stimolazione), un classificatore Random Forest raggiunge un'accuratezza del 77%, dimostrando la sua efficacia nell'identificare sia la mutazione genetica che i cambiamenti nell'attività neurale indotti dalla stimolazione. La ricerca fornisce una migliore comprensione dell'interazione complessa tra la mutazione LRRK2 e la dinamica dell'attività neuronale, fornendo preziosi spunti sui meccanismi fisiopatologici alla base della malattia di Parkinson, riscontrabili nei risultati ottenuti. Inoltre, il framework computazionale sviluppato in questo studio offre un approccio innovativo per lo studio dei disturbi neurodegenerativi a livello cellulare, con implicazioni per future strategie diagnostiche e terapeutiche.
Analysis of neural activity changes associated with LRRK2 Gene mutation: a computational approach for neuronal activity classification using in-vitro networks
Vettori, Gaia
2022/2023
Abstract
Parkinson's disease (PD) is a debilitating neurodegenerative disorder whose etiology is strongly linked to the G2019S mutation of the LRRK2 gene - or Dardarin. Although this causal relationship is well established, the underlying mechanisms are not fully understood. In this thesis, we use Microelectrode Array (MEA) technology to analyze in-vitro neural networks, in order to delineate the characteristic neural activity patterns of carriers of the G2019S mutation of the LRRK2 gene, associated with PD pathogenesis. This study utilizes recordings from both G2019S-mutated cortical neural networks and from healthy control networks in two phases: before (baseline) and after Kainic Acid stimulation. The chemical stimulation process aims to induce neurotoxic effects, and by incorporating post-stimulation data into the analysis, to record changes in neural activity. The work focuses on the development of a three-phase algorithm capable of receiving raw MEA recordings and recognizing the population to which they belong, distinguishing between healthy and mutated activity. Initially, Spike Sorting techniques are used to identify individual neuronal activities. Subsequently, Point Process Modeling technique is employed to characterize the parameters of the ISI curve distribution, through Maximum A Posteriori Estimation. Finally, a classification phase is performed, where the model is trained with the parameters obtained from the modeling, and the data are categorized initially into mutated and control neurons, and subsequently, in a multi-class problem, both in terms of type of population and baseline vs. post-stimulation activity. The results obtained from binary classification demonstrate an excellent accuracy of 99% (mutated vs. control) using a Support Vector Machine model. Furthermore, in a more nuanced four-class classification scenario (mutated vs. control and pre- vs. post-stimulation), a Random Forest classifier achieves an accuracy of 77%, showcasing its efficacy in identifying both the genetic mutation and the stimulation-induced changes in neural activity. The research sheds light on the complex interplay between the LRRK2 mutation and neuronal activity dynamics, providing valuable insights into the pathophysiological mechanisms underlying Parkinson's disease, as confirmed by the obtained results. Moreover, the computational framework developed in this study offers a novel approach for studying neurodegenerative disorders at the cellular level, with implications for future diagnostic and therapeutic strategies.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Vettori_Tesi_01.pdf
Open Access dal 19/03/2025
Descrizione: Tesi
Dimensione
4.12 MB
Formato
Adobe PDF
|
4.12 MB | Adobe PDF | Visualizza/Apri |
2024_04_Vettori_ExecutiveSummary_02.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
524.5 kB
Formato
Adobe PDF
|
524.5 kB | 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/10589/219275