Large-Scale Computational Phenotyping of Social Cognition in Autism

Author: Wu, Qianying

Year: 2026

Degree: Dissertation (Ph.D.)

Advisor: Adolphs, Ralph

Committee Members: Alvarez, R. Michael; Adolphs, Ralph; O'Doherty, John P.; Shimojo, Shinsuke; Charpentier, Caroline J.

Option: Social and Decision Neuroscience

DOI: 10.7907/6h98-8v12

Abstract

Autism Spectrum Disorder (ASD) is a highly heterogeneous neurodevelopmental condition characterized by differences in social cognition, motivation, and communication, alongside atypical sensory processing and rigid behaviors. The remarkable variability in its behavioral manifestations reflects diverse etiological pathways, underscoring the need to move beyond descriptive analysis toward mechanistic accounts of autism. Computational psychiatry offers a powerful framework for this goal, combining large-scale data with computational modeling to quantify individual differences in cognition and behavior. Social cognition - the processes by which individuals process, store and leverage social information - is central to multiple cognitive domains, including attention, perception, learning, and memory. The current thesis applies computational psychiatry approaches to develop reliable and scalable experimental paradigms and quantitative methods for studying social cognition across the autism spectrum. Chapter 1 introduces the opportunities and challenges in computational phenotyping for characterizing heterogeneity in social cognitive profiles in autism. Chapter 2 investigates whether social communication challenges in autism stem from atypical attention patterns during group interactions. We developed a remote eye-tracking protocol using videos of interacting speakers, and collected data from over 1,300 adult participants. Results revealed that autistic individuals allocated less attention to current speakers and more attention to irrelevant distractors, patterns not explained by reduced social motivation and not observed in psychiatric comparison groups. Notably, increased attention to distractors was context-sensitive and most pronounced among autistic women. Chapter 3 examines individual differences in observational learning strategies. Although autistic traits have been linked to social learning difficulties, the underlying computational mechanisms remained unclear. Using computational modeling in two large online general-population samples, we characterized arbitration between two social learning strategies: action imitation and goal emulation. Autistic traits were associated with lower performance in observational learning through reduced emulation but not imitation, revealing difficulties in social goal inference. This association was specifically related to autistic traits but not social anxiety traits. Chapter 4 addresses concern about the reliability and generalizability of online psychological assessments. We systematically characterized the variability of autistic traits across individuals in a large cross-sectional dataset and examined within-individual temporal reliability in a test-retest dataset, comparing online and in-lab samples. Finally, Chapter 5 synthesizes the empirical findings, discuss the social cognitive mechanisms underlying behavioral differences in autism, reflects on practical challenges in large-scale phenotyping, and outlines future directions. Overall, leveraging a computational psychiatry approach, this thesis advances large-scale computational phenotyping of social cognition in autism, providing novel insights into the sources and structure of heterogeneity across the spectrum.