The brain can be regarded as a highly sophisticated computational system: it receives sensory inputs, constructs internal representations of the external world, and performs transformations on these representations to execute tasks. Neurons—the fundamental computational units of the brain—carry out this process by integrating thousands of synaptic inputs across their dendritic arbor. A defining functional property of many cortical neurons is their selectivity for specific stimulus features, such as orientation in the visual cortex. This selectivity emerges and becomes more refined across successive processing stages. Understanding how dendritic geometry and synaptic organization contribute to such selectivity remains a central question in neuroscience. Recent computational studies, based on synthetic or ex vivo datasets, have suggested that dendritic distance and compartmental specialization play key roles in shaping neuronal selectivity. The main objective of this work is to test whether these findings hold for the experimentally derived dataset in vivo from the mouse primary visual cortex MICrONS. We begin by assessing the limitations introduced by incomplete synaptic reconstructions and by validating a simple linear summation model as a baseline. This analysis shows that accurate predictions require access to several hundred synapses—far more than what is available in the MICrONS dataset—but also that the model performs substantially better than chance. Moreover, prediction accuracy improves systematically as more synapses become accessible. We then evaluate the relative importance of apical and basal dendritic compartments by optimizing the weight assigned to each contribution, in order to assess their individual and combined predictive power. Although the literature often assumes a marginal role for apical inputs, our results show that the lowest prediction error is achieved when both compartments are included with appropriately optimized weights. Finally, we extend the model to incorporate distance-dependent attenuation, testing the hypothesis—supported by previous theoretical work—that dendrites act as filters that reduce synaptic efficacy as the synapse–soma distance increases. To examine this effect, we systematically explored several distance-based models, finding that the inclusion of dendritic filtering reduced the overall error in predicting preferred orientation. Overall, these results suggest that both dendritic location and distance play an essential role in shaping how neurons combine their inputs.
Il cervello può essere considerato un sistema computazionale altamente sofisticato: riceve input sensoriali, costruisce rappresentazioni interne del mondo esterno ed esegue trasformazioni su queste rappresentazioni per compiere compiti complessi. I neuroni—le unità computazionali fondamentali del cervello—realizzano questo processo integrando migliaia di input sinaptici distribuiti lungo il loro albero dendritico. Una proprietà funzionale caratteristica di molti neuroni corticali è la selettività per specifiche caratteristiche dello stimolo, come l’orientamento nella corteccia visiva. Tale selettività emerge e si raffina progressivamente lungo le diverse tappe della gerarchia di elaborazione. Comprendere in che modo la geometria dendritica e l’organizzazione sinaptica contribuiscano a questa selettività rappresenta una questione centrale nelle neuroscienze. Studi computazionali recenti, basati su dati sintetici o non in vivo, hanno suggerito che la distanza sinaptica dal soma e l'appartenenza a compartimenti dendritici apicale e basale svolgano un ruolo chiave nel modellare la selettività neuronale. L’obiettivo principale di questo lavoro è verificare se tali risultati siano confermati nel dataset sperimentale in vivo MICrONS, ottenuto dalla corteccia visiva primaria del topo. Iniziamo analizzando i limiti introdotti dalle ricostruzioni sinaptiche incomplete e validando un semplice modello lineare di sommazione come baseline. Questa analisi mostra che per ottenere predizioni accurate sono necessari diversi centinaia di sinapsi—molte più di quelle disponibili nel dataset MICrONS—ma evidenzia anche che il modello ottiene prestazioni significativamente migliori del caso. Inoltre, l’accuratezza predittiva migliora sistematicamente all’aumentare del numero di sinapsi disponibili. Valutiamo poi l’importanza relativa dei compartimenti dendritici apicale e basale ottimizzando il peso assegnato a ciascun contributo, allo scopo di analizzare il loro potere predittivo individuale e combinato. Sebbene la letteratura spesso attribuisca agli input apicali un ruolo marginale, i nostri risultati mostrano che l’errore di predizione minimo si ottiene quando entrambi i compartimenti vengono inclusi con pesi opportunamente ottimizzati. Infine, estendiamo il modello includendo un’attenuazione dipendente dalla distanza, testando l’ipotesi—supportata da studi teorici precedenti—che i dendriti agiscano come filtri che riducono l’efficacia sinaptica con l’aumentare della distanza dal soma. Per valutare questo effetto, esploriamo sistematicamente diversi modelli basati sulla distanza, osservando che l’inclusione del filtraggio dendritico riduce l’errore complessivo nella predizione dell’orientamento preferito. Nel complesso, questi risultati suggeriscono che sia la posizione dendritica sia la distanza della sinapsi dal soma svolgono un ruolo fondamentale nel modellare il modo in cui i neuroni combinano i propri input.
Modeling dendritic processing for orientation selectivity in the mouse visual cortex
RUSCITO, FIAMMA
2024/2025
Abstract
The brain can be regarded as a highly sophisticated computational system: it receives sensory inputs, constructs internal representations of the external world, and performs transformations on these representations to execute tasks. Neurons—the fundamental computational units of the brain—carry out this process by integrating thousands of synaptic inputs across their dendritic arbor. A defining functional property of many cortical neurons is their selectivity for specific stimulus features, such as orientation in the visual cortex. This selectivity emerges and becomes more refined across successive processing stages. Understanding how dendritic geometry and synaptic organization contribute to such selectivity remains a central question in neuroscience. Recent computational studies, based on synthetic or ex vivo datasets, have suggested that dendritic distance and compartmental specialization play key roles in shaping neuronal selectivity. The main objective of this work is to test whether these findings hold for the experimentally derived dataset in vivo from the mouse primary visual cortex MICrONS. We begin by assessing the limitations introduced by incomplete synaptic reconstructions and by validating a simple linear summation model as a baseline. This analysis shows that accurate predictions require access to several hundred synapses—far more than what is available in the MICrONS dataset—but also that the model performs substantially better than chance. Moreover, prediction accuracy improves systematically as more synapses become accessible. We then evaluate the relative importance of apical and basal dendritic compartments by optimizing the weight assigned to each contribution, in order to assess their individual and combined predictive power. Although the literature often assumes a marginal role for apical inputs, our results show that the lowest prediction error is achieved when both compartments are included with appropriately optimized weights. Finally, we extend the model to incorporate distance-dependent attenuation, testing the hypothesis—supported by previous theoretical work—that dendrites act as filters that reduce synaptic efficacy as the synapse–soma distance increases. To examine this effect, we systematically explored several distance-based models, finding that the inclusion of dendritic filtering reduced the overall error in predicting preferred orientation. Overall, these results suggest that both dendritic location and distance play an essential role in shaping how neurons combine their inputs.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Ruscito_Executive_Summary.pdf
accessibile in internet per tutti a partire dal 19/11/2026
Descrizione: executive summary
Dimensione
867.38 kB
Formato
Adobe PDF
|
867.38 kB | Adobe PDF | Visualizza/Apri |
|
2025_12_Ruscito_Thesis.pdf
accessibile in internet per tutti a partire dal 19/11/2026
Descrizione: Tesi
Dimensione
4.11 MB
Formato
Adobe PDF
|
4.11 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/10589/247181