Researchers at Princeton University and the Simons Foundation have used artificial-intelligence modelling to analyse genetic and clinical data from more than 5,000 children in the SPARK autism cohort, identifying four biologically and clinically distinct subtypes of autism. The work, published 9 July in Nature Genetics, offers one of the clearest links yet between specific genetic patterns and the condition’s varied presentations. The subgroups—labelled Social and Behavioral Challenges, Mixed ASD with Developmental Delay, Moderate Challenges, and Broadly Affected—differ in developmental milestones, psychiatric co-occurring conditions and underlying mutations. For example, the Broadly Affected group shows the highest rate of damaging de-novo mutations, while the Mixed ASD with Developmental Delay subtype is more likely to carry rare inherited variants. Lead author Olga Troyanskaya said the findings could accelerate precision diagnosis and personalised interventions by enabling clinicians to anticipate developmental trajectories and tailor care. The framework may also aid gene-discovery efforts in other heterogeneous disorders, the team added.