Precision profiling of sensory and cognitive processing in neurodevelopmental conditions: language-free digital assessment, EEG and computational modelling for stratification

Background

Precision medicine in neurodevelopmental conditions is limited by measurement. Diagnosis rests on observed behaviour, outcome measures in trials are parent- or clinician-rated questionnaires, and the sensory differences that dominate lived experience in FXS and autism have no objective, scalable measure. Standard neuropsychological tests require language and instruction-following and are administered in clinic, which systematically excludes people with intellectual disability, the minimally verbal, and families without the means to travel: an inequality built into the evidence base.

Two developments make a different approach possible. Language-free video games can measure specific computations (evidence accumulation, context-dependent choice, rule switching, sensory detection) in children who cannot do standard tests, remotely and repeatedly. And per-individual computational modelling can decompose performance into interpretable parameters and latent dimensions rather than a score. The lead supervisor's work established that individuals differ in the neural computation they use to solve the same task (Pagan et al. 2025, Nature); the co-supervisor's centre runs the clinical cohorts and is collaborating with the lead supervisor on the analyses of a tactile psychophysics and EEG dataset in FXS.

Aims

1. Deploy and validate a language-free sensory and cognitive game battery, together with research-grade tactile psychophysics and tactile-evoked EEG, in participants with FXS, autism and typically developing controls, in clinic and remotely.

2. Fit per-individual computational models (drift-diffusion and related accumulation models; psychometric and lapse models for detection) and identify the latent dimensions of sensory and cognitive processing across measures.

3. Test whether these dimensions stratify participants, predict clinical phenotype and EEG markers, and are reliable enough on repeat testing to serve as outcome measures for future trials.

4. (Exploratory) Apply the same models to the laboratory's existing rat behavioural datasets (Fmr1 knockout and wild type, matched tasks) to ask whether human stratifying dimensions have a mechanistic counterpart.

Training Outcomes

Design and ethics of human studies in vulnerable populations; psychophysics and EEG acquisition and preprocessing; software engineering for remote tablet-based data collection; Bayesian and likelihood model fitting, model comparison, dimensionality reduction and clustering; machine learning including LLM-based model discovery; reproducible pipelines (Python, git, NWB/BIDS standards); scientific communication to clinicians, families and computational audiences; an understanding of trial outcome-measure development. The integrated-study credits will be chosen to fill the student's gaps (statistics and machine learning for a clinically trained student; clinical neuroscience and ethics for a computational one).

References

Pagan M et al. (2025) Nature. Robertson CE, Baron-Cohen S (2017) Nat Rev Neurosci 18:671. Ratcliff R, McKoon G (2008) Neural Comput 20:873. Chakravarty S et al. (2026) Computational mechanisms of perception in autism revealed using games inspired by rodent operant tasks. Sci Adv.