Computational modeling of the impact of synaptic pruning defects on Working Memory in Autism Spectrum Disorder

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Hikaru Sasaki¹ (c0b2107561@edu.teu.ac.jp), Masayuki Kikuchi²; ¹Tokyo University of Technology

Synaptic pruning plays an important role in the maturation of neural networks in the brain and is also thought to help shape small-world properties for efficient information transmission. In autism spectrum disorder (ASD), genetic abnormalities linked to synaptic pruning defects may disrupt these properties. Although the mechanism is still unknown, ASD symptoms are thought to be caused by dysfunction of neural networks. Working memory (WM) deficits in ASD are associated with communication issues and stereotypical behavior. This study aimed to explore the previously unexamined link between synaptic pruning defects and reduced WM ability in ASD using a computational model. We used Echo State Networks (ESNs), consisting of an input, reservoir, and output layers. The structure of reservoir layer follows the Watts-Strogatz model, enabling small-world networks by rewiring edges with a given probability. We constructed two models: the "ASD Model" (rewiring probability = 0.007) in which majority of connections are local and the "TD Model" (0.07) in which a small number of connections are global. In reservoir layer, each node corresponds to a group of randomly connected spiking neurons whose behavior is described by Izhikevich model. The model performed a 2-back task, identifying grating patterns presented two steps earlier. Results showed that the TD model had a higher accuracy rate than the ASD model (p<0.05). This result suggests that the loss of small-world properties due to defective synaptic pruning may contribute to the decline in WM ability in ASD.

Topic Area: EXECUTIVE PROCESSES: Working memory