Are prediction error modifications domain specific in autism but domain general in ADHD?

Abstract

In the last decade, several theories have proposed explanations of autism spectrum disorder (ASD) based on the Bayesian Brain hypothesis. Research on repetition suppression suggests that neural correlates of prediction errors in ASD might be domain specific with larger differences for faces than for objects. Contrastingly, research assessing mismatch negativity in attention-deficit/hyperactivity disorder (ADHD) indicates domain-generally modified prediction errors. Therefore, we captured neural correlates of prediction errors to colours and emotions in adults with ADHD or ASD to assess domain specificity of prediction error modifications. We used a multi-feature roving paradigm where we assessed predictive processes regarding unattended, task-irrelevant faces. We extracted task specific precision-weighted prediction errors and prediction strength for emotions and colours of the faces separately using a generative model, specifically a Hierarchical Gaussian Filter. While we found neural correlates of colour precision-weighted prediction errors and prediction strength as well as emotion prediction strength in the pooled sample regardless of group, we did not find any differences in neural correlates of emotion or colour precision-weighted prediction errors or prediction strength between our groups. These results suggest preserved neural correlates of precision-weighted prediction errors in ASD and ADHD for low-level visual features and potentially complex social information.

Keywords: ADHD; Bayesian Brain; autism; emotion processing; face processing; prediction error; predictive coding.