The 12 Faces of Autism
A Gene-Phenotype Atlas
Research Data Only - Not For Clinical Use
This resource contains literature-derived associations, NOT clinical data. It cannot provide penetrance estimates, severity predictions, or diagnostic guidance. Absence of a gene-phenotype link means "not reported in literature," not "clinically absent." See Methods → Critical Limitations before interpreting any data.
Autism is not one condition. It's twelve.
AI analysis of 1,000+ papers reveals distinct genetic subtypes with unique symptom profiles
What We Did: A Plain-Language Summary
Purpose
Autism spectrum disorder (ASD) is incredibly diverse - no two individuals present exactly the same way. We wanted to see if we could identify distinct subtypes of autism based on which symptoms tend to occur together in people with specific gene variants.
Methods
We used AI to read over 1,000 scientific papers about genes from the SFARI autism database, extracting which symptoms (like seizures, language delay, or intellectual disability) were reported for each gene. Then we used a clustering algorithm to group the 241 genes with sufficient data by their symptom profiles.
Results
We found 12 distinct clusters - 6 major and 6 minor subtypes. For example, one cluster features mostly intellectual disability, another shows seizures with language problems, and another has behavioral/anxiety features. Machine learning correctly predicted cluster membership 86% of the time, suggesting these are real patterns.
Conclusion
Autism isn't one condition - it's many. Different genes tend to cause different constellations of symptoms. This could eventually help doctors predict what to watch for based on a patient's genetic results, though much more research is needed before clinical use.