This thesis explores a computational model suitable for understanding synaptic-based working memory in human brains. The starting point of the work is based on the publication by Taher H., Torcini A., and Olmi S. "Exact neural mass model for synaptic-based working memory" (PLOS Computational Biology, 2020). The goal is to computationally reproduce the biological dynamics recorded on human prefrontal cortex, a brain region important for executive functions and working memory. We developed an analysis starting from the single QIF neuron dynamics. With Quadratic Integrate-and-Fire neuron we refer to a mathematical model that simulates the procession and transmission of electrical signal and that incorporates realistic membrane dynamics. Through the mean-field work, we studied in detail the exact neural mass model that is derived and that describes the collective dynamics of infinite populations of QIF neurons synaptically coupled. Unlike traditional neural mass models, the proposed model comes from a new approach that replaces approximations with ordinary differential equations derived directly from the single-neuron biophysics. We examined how incorporating short-term synaptic plasticity into the neural mass model, so including synaptic depression and facilitation mechanisms, results in a model for synaptic-based working memory described by a system of four ordinary differential equations that govern the temporal evolution of firing rate r(t), mean membrane potential v(t), available synaptic resources x(t), and synaptic utilisation factor u(t). This approach allowed, then, to conduct some simulations to observe how memory items can be maintained through dynamic changes in synaptic strength, rather than through persistent neural activity. From a computational perspective, the model is first implemented and validated through a comparative simulation between a mesoscopic network of excitatory neurons with short-term plasticity and the corresponding neural mass model, composed of one excitatory node. With short-term synaptic plasticity we refer to the temporary changes in synaptic strength that occur over timescales of milliseconds to minutes (in contrast to long-term plasticity which can last hours to years). Next, we further investigated three specific working memory experiments, conducted on the developed neural mass model. These experiments include stimulations that demonstrate the system’s ability to encode, maintain, and retrieve information through short-term synaptic plasticity mechanisms. We observed that the model can maintain memory traces for extended periods through brief activation episodes, restoring facilitated synaptic strength. This analysis are, so, based on a model that correctly reproduced biological dynamics recorded on human prefrontal cortex, a brain region important for executive functions and working memory. The proposed model, therefore, represents a promising biophysical description of how information can be temporarily stored and manipulated through synaptic mechanisms in our brain, and provides a computationally efficient alternative to traditional models based on persistent activity. This work in conclusion, contributes to the theoretical understanding of neural mechanisms, specifically underlying working memory, but provides also a robust mathematical tool for future investigations concerning neural population modelling and cortical oscillations in clinical and functional analysis, from the generation of functional connectivity in the whole brain network to seizure propagation.
Questa tesi esplora un modello computazionale adatto per comprendere la memoria di lavoro basata su sinapsi nei cervelli umani. Il punto di partenza del lavoro si basa sulla pubblicazione di Taher H., Torcini A., e Olmi S. "Exact neural mass model for synaptic-based working memory" (PLOS Computational Biology, 2020).L'obiettivo è riprodurre computazionalmente le dinamiche biologiche registrate nella corteccia prefrontale umana, una regione cerebrale importante per le funzioni esecutive e la memoria di lavoro. Abbiamo sviluppato un'analisi partendo dalle dinamiche del singolo neurone QIF. Con neurone Quadratic Integrate-and-Fire ci riferiamo a un modello matematico che simula la processazione e trasmissione del segnale elettrico e che incorpora dinamiche realistiche di membrana. Attraverso il lavoro di campo medio, abbiamo studiato in dettaglio il modello esatto di massa neurale che ne deriva e che descrive le dinamiche collettive di popolazioni infinite di neuroni QIF accoppiati sinapticamente.A differenza dei modelli tradizionali di massa neurale, il modello proposto deriva da un nuovo approccio che sostituisce le approssimazioni con equazioni differenziali ordinarie derivate direttamente dalla biofisica del singolo neurone. Abbiamo esaminato come l'incorporazione della plasticità sinaptica a breve termine nel modello di massa neurale, includendo quindi i meccanismi di depressione e facilitazione sinaptica, risulti in un modello per la memoria di lavoro basata su sinapsi descritto da un sistema di quattro equazioni differenziali ordinarie che governano l'evoluzione temporale della frequenza di scarica r(t), del potenziale medio di membrana v(t), delle risorse sinaptiche disponibili x(t), e del fattore di utilizzazione sinaptica u(t).Questo approccio ha permesso, quindi, di condurre alcune simulazioni per osservare come gli elementi della memoria possano essere mantenuti attraverso cambiamenti dinamici nella forza sinaptica, piuttosto che attraverso attività neurale persistente.Dal punto di vista computazionale, il modello viene prima implementato e validato attraverso una simulazione comparativa tra una rete mesoscopica di neuroni eccitatori con plasticità a breve termine e il corrispondente modello di massa neurale, composto da un nodo eccitatorio. Con plasticità sinaptica a breve termine ci riferiamo ai cambiamenti temporanei nella forza sinaptica che avvengono su scale temporali da millisecondi a minuti (in contrasto con la plasticità a lungo termine che può durare da ore ad anni).Successivamente, abbiamo ulteriormente investigato tre specifici esperimenti di memoria di lavoro, condotti sul modello di massa neurale sviluppato. Questi esperimenti includono stimolazioni che dimostrano la capacità del sistema di codificare, mantenere e recuperare informazioni attraverso meccanismi di plasticità sinaptica a breve termine.Abbiamo osservato che il modello può mantenere tracce mnemoniche per periodi prolungati attraverso brevi episodi di attivazione, ripristinando la forza sinaptica facilitata.Queste analisi si basano, quindi, su un modello che ha correttamente riprodotto le dinamiche biologiche registrate nella corteccia prefrontale umana, una regione cerebrale importante per le funzioni esecutive e la memoria di lavoro. Il modello proposto rappresenta, pertanto, una descrizione biofisica promettente di come l'informazione possa essere temporaneamente immagazzinata e manipolata attraverso meccanismi sinaptici nel nostro cervello, e fornisce un'alternativa computazionalmente efficiente ai modelli tradizionali basati su attività persistente.Questo lavoro contribuisce, in conclusione, alla comprensione teorica dei meccanismi neurali, specificamente quelli sottostanti la memoria di lavoro, ma fornisce anche uno strumento matematico robusto per future investigazioni riguardanti la modellazione di popolazioni neurali e le oscillazioni corticali in analisi cliniche e funzionali, dalla generazione della connettività
Dinamiche della memoria di lavoro nella corteccia prefrontale: da spiking neural networks a modelli di massa neurale con la plasticità sinaptica a breve termine
TURBA CERETTI, CHIARA
2024/2025
Abstract
This thesis explores a computational model suitable for understanding synaptic-based working memory in human brains. The starting point of the work is based on the publication by Taher H., Torcini A., and Olmi S. "Exact neural mass model for synaptic-based working memory" (PLOS Computational Biology, 2020). The goal is to computationally reproduce the biological dynamics recorded on human prefrontal cortex, a brain region important for executive functions and working memory. We developed an analysis starting from the single QIF neuron dynamics. With Quadratic Integrate-and-Fire neuron we refer to a mathematical model that simulates the procession and transmission of electrical signal and that incorporates realistic membrane dynamics. Through the mean-field work, we studied in detail the exact neural mass model that is derived and that describes the collective dynamics of infinite populations of QIF neurons synaptically coupled. Unlike traditional neural mass models, the proposed model comes from a new approach that replaces approximations with ordinary differential equations derived directly from the single-neuron biophysics. We examined how incorporating short-term synaptic plasticity into the neural mass model, so including synaptic depression and facilitation mechanisms, results in a model for synaptic-based working memory described by a system of four ordinary differential equations that govern the temporal evolution of firing rate r(t), mean membrane potential v(t), available synaptic resources x(t), and synaptic utilisation factor u(t). This approach allowed, then, to conduct some simulations to observe how memory items can be maintained through dynamic changes in synaptic strength, rather than through persistent neural activity. From a computational perspective, the model is first implemented and validated through a comparative simulation between a mesoscopic network of excitatory neurons with short-term plasticity and the corresponding neural mass model, composed of one excitatory node. With short-term synaptic plasticity we refer to the temporary changes in synaptic strength that occur over timescales of milliseconds to minutes (in contrast to long-term plasticity which can last hours to years). Next, we further investigated three specific working memory experiments, conducted on the developed neural mass model. These experiments include stimulations that demonstrate the system’s ability to encode, maintain, and retrieve information through short-term synaptic plasticity mechanisms. We observed that the model can maintain memory traces for extended periods through brief activation episodes, restoring facilitated synaptic strength. This analysis are, so, based on a model that correctly reproduced biological dynamics recorded on human prefrontal cortex, a brain region important for executive functions and working memory. The proposed model, therefore, represents a promising biophysical description of how information can be temporarily stored and manipulated through synaptic mechanisms in our brain, and provides a computationally efficient alternative to traditional models based on persistent activity. This work in conclusion, contributes to the theoretical understanding of neural mechanisms, specifically underlying working memory, but provides also a robust mathematical tool for future investigations concerning neural population modelling and cortical oscillations in clinical and functional analysis, from the generation of functional connectivity in the whole brain network to seizure propagation.| File | Dimensione | Formato | |
|---|---|---|---|
|
Dynamics of working memory in the prefrontal cortex _ from spiking networks to neural mass model with short-term synaptic plasticity.pdf
accesso aperto
Dimensione
3.06 MB
Formato
Adobe PDF
|
3.06 MB | Adobe PDF | Visualizza/Apri |
È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/30258