Back

Predictive inference alterations in psychosis proneness are context-dependent

Tarasi, L.; Romei, V.

2025-05-20 neuroscience
10.1101/2025.05.17.654666 bioRxiv
Show abstract

BackgroundPerception depends on the dynamic integration of sensory inputs and prior expectations, shaped by recent experience. While disruptions in this inferential process have been proposed as core features of psychosis, whether such alterations can be observed in subclinical psychosis-like experiences (PLEs), and how they manifest across different forms of perceptual inference remains unclear. MethodsTwo independent samples of healthy adults participated in the study: 80 completed a visual detection task, and 62 performed a motion discrimination task. Data were analysed using mixed-effects models with a continuous PLE measure and supplemented by categorical comparisons between the lowest- and highest-PLEs terciles. ResultsPLEs predicted reduced sensitivity to sensory evidence and increased reliance on probabilistic cues across both tasks. Critically, history biases diverged across contexts. In the detection task, a repulsive bias from the previous stimulus was observed in the overall sample, but this effect was attenuated in high-PLEs. In contrast, in the discrimination task, high-PLEs exhibited a repulsive bias away from their previous choice, reversing the typical repetition tendency seen in low-PLEs. ConclusionsOur results suggest that psychosis-like traits alter how prior expectations interact with sensory input and recent perceptual history. High-PLEs showed reduced influence of recent sensory input in detection task, consistent with altered low-level adaptation. In contrast, in the discrimination task, they showed a shift away from previous choices, suggesting disruption of decision-level integration of past choices. This points to a task-dependent instability in the use of past information, reflecting a broader inferential vulnerability linked to the psychosis spectrum.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.