JMIR Publications

ABSTRACT

Autism encompasses heterogeneous genetic, neurobiological, developmental, and medical pathways. Artificial intelligence (AI) has advanced autism research through multimodal analysis of eye tracking, EEG, neuroimaging, and molecular data; beyond autism, AI is also accelerating therapeutic target identification and drug discovery. Yet translating these advances into individualized, biologically informed clinical decisions remains a major challenge. We propose biologically anchored AI as a decision-oriented framework linking genetic etiology, molecular and microbiome-associated pathways, neurophysiology, digital observations, and developmental context to explicit clinical questions. We distinguish etiological characterization, exploratory mechanistic stratification, and treatment-specific response prediction; these are not interchangeable forms of validation. Examples include individualized genetic therapies, SYNGAP1-directed drug development, heterogeneous responses in fragile X and oxytocin trials, stool metabolomics, sleep EEG, and prenatal MRI. Rather than requiring comprehensive testing for every person, the approach asks whether a selected measurement improves a defined decision beyond standard assessment. Candidate subgroups and biomarkers must be replicated, tested prospectively for incremental value and, where treatment selection is proposed, evaluated for differential benefit from a specified intervention. The goal is not a new universal autism subtype or an automated treatment recommendation, but clinically meaningful, equitable, and developmentally responsive precision care.