Leveraging computational modeling and data-driven clustering approaches to constrain the heterogeneity of autism phenotypes
- Awarded: 2022
- Award Type: Human Cognitive and Behavioral Science Award
- Award #: 986832
In this project, Gabriela Rosenblau and colleagues aim to systematically investigate and constrain the heterogeneity of autism spectrum disorder (ASD) by providing a fine-grained mechanistic characterization of the relationship between two defining symptom dimensions of ASD: (i) restricted or highly-focused interests and (ii) challenges in social cognition.
To achieve this, they plan to combine hypothesis-driven, carefully designed assessments of behavior, computational modeling and data-driven clustering approaches. Leveraging the SPARK cohort and the Research Match program, the proposed work aims to recruit 1,000 children with ASD. Rosenblau’s team will assess the self-preferences of these children for a large number of activities and foods. They will then characterize the self-preference distributions in order to differentiate between individuals with broader preference profiles and those with restricted, highly focused interests. In addition, the project will characterize social cognition via computational modeling, by describing the strategies of children with ASD when learning about the preferences of peers that are typically developing (TD) and peers with an ASD diagnosis.
The social learning task and associated computational models used in this proposal have yielded robust and replicable results across different adolescent populations of TD individuals and neurotypical adults1,2,3. Moreover, this social learning framework has been successfully used to identify differences in social learning strategies and their neural underpinnings in autistic adolescents3.
To assess how typical or atypical self-preference distributions and social learning strategies of individuals with ASD are, Rosenblau and colleagues plan to cluster individuals based on core dimensions of social learning strategies and rigidity/flexibility of self-preferences. They will also investigate a multi-dimensional classification based on core task-relevant dimensions identified using principal component analysis. An important additional goal is to explore sex differences in learning strategies, self-preferences and their relationship. To test the external validity of the clustering approach, the team will use individuals’ cluster memberships and their position on the two- and multi-dimensional space to predict individuals’ clinical and behavioral profiles (i.e., symptom severity, cognitive flexibility, overall social cognitive functioning and adaptive living skills).
Rosenblau and colleagues hypothesize that individuals with more narrowly defined self-preferences will rely less on rich social knowledge structures during learning and this will scale with lower social functioning and adaptive living skills in real life. Findings from this project will inform future studies examining associations between behaviorally derived clusters and ASD genetic risk factors.