LEARNING FLEXIBLE DECISION-MAKING IN RAT MODEL OF FRAGILE X SYNDROME AND IN RECURRENT NEURAL NETWORKS

Flexible decision-making, the ability to rapidly switch thinking and actions according to context, is a fundamental component of higher cognition, but our understanding of the neural circuit implementation remains incomplete. Cognitive flexibility is compromised in neurodevelopmental conditions, such as Fragile X Syndrome (FXS). We investigate the neural basis of this process and its possible alteration by training a rat model of FXS, Fmr1-knock-out (-KO), on a flexible decision-making task. The task requires rats to select and accumulate relevant sensory evidence depending on the current context (Pagan et al., 2025). Both wild-type and Fmr1-KO rats show dramatic individual variability, with individuals in both groups progressing at varying speeds through training stages of increasing task complexity and showing heterogeneous performance at each stage. We use modelling frameworks, such as GLM-HMM (Ashwood et al., 2022), to characterise individual behaviour over learning in detail, including reliance on irrelevant factors when forming decisions. To better understand individual variability in learning, we examine a large sample of previously-trained WT rats (Pagan et al., 2025). We find a consistent order in which rats learnt to accumulate evidence and select the relevant feature over months of training. We also train recurrent neural networks on the task, which show random variability in learning speed and give us access to the dynamics underlying task computations. Networks models which learn at different speeds nonetheless recapitulate the consistent structure in their learning dynamics found in rats. This work will inform efforts to improve learning of flexible decisions on an individual basis.