Computational phenotyping of individual variation in latent-state learning, generalization and attention across the autism spectrum

  • Awarded: 2022
  • Award Type: Human Cognitive and Behavioral Science Award
  • Award #: 988485

People with autism spectrum disorder (ASD) often exhibit rigid patterns of behavior, manifesting as difficulty generalizing when encountering a new example of something known or something known in a novel context. Heterogeneity along the spectrum suggests there are multiple potential causes of failure in generalization. Indeed, ASD demonstrates high co-occurrence with attention-deficit hyperactivity disorder (ADHD) and phenotypic overlap with schizotypy.

Computational psychiatry is a new discipline that utilizes mathematical methods to rigorously describe the computational processes that give rise to clinically-observed phenomena. In latent-state reinforcement learning, agents learn “states,” which organize sensory information and action-outcomes.

John Murray and colleagues recently developed a latent-state computational model that provides insight into how top-down attention is deployed as internal states are learned and maintained over time, as well as how internal states are generalized to new examples or contexts. Utilizing the model, they developed novel behavioral tasks that validated its use of specific attention and state-formation mechanisms in large populations of neurotypical individuals1.

In the current project, Murray and Suma Jacob’s team at the University of Minnesota plan to use a combination of online behavioral tasks and computational modeling to study heterogeneity in learning, attention and generalization along the autism spectrum. In Aim 1a, they plan to recruit autistic and neurotypical adults from a general online forum and use these tasks to investigate population-level ASD heterogeneity in attention and generalization of learned states, as well as administer standard clinical questionnaires to assess the relationship between causes of behavioral rigidity and clinical traits. This will leverage existing tasks to test general hypotheses regarding ASD behavioral heterogeneity. In Aim 1b, they plan to benchmark a novel computational phenotyping task against the existing task in the same individuals. In Aim 2, they aim to study the test-retest reliability of the computational phenotyping task in neurotypical adults, which will allow the research team to characterize the stability in measures over time. In Aim 3, they plan to deploy that task with matched autistic and neurotypical adults enrolled in SPARK. The overall goal of the project is to provide insight into the causes of behavioral rigidity in ASD.


  1. Pettine W.W. et al. Nat. Hum. Behav. 7, 442-463 (2023) PubMed
Subscribe to our newsletter and receive SFARI funding announcements and news