Prenatal genomics and autism: detectable susceptibility without validated individual-level prediction

Autism spectrum disorder (ASD) is a behaviorally defined, biologically heterogeneous neurodevelopmental condition with high heritability, estimated at 0.83 (95% CI 0.79–0.87) in Swedish population data. Large-scale exome sequencing, copy-number variant analysis, and genome-wide association studies have implicated a growing number of genes and genomic loci in autism liability, and functional genomic work has repeatedly converged on mid-gestational cortical development. In parallel, prenatal genetics has moved from karyotyping to chromosomal microarray analysis, exome and genome sequencing, and screening based on cell-free DNA in maternal plasma. This review asks how far these developments allow genetic susceptibility to autism to be identified before birth, and what prevents such identification from constituting prediction or diagnosis of autism in an individual fetus. It organizes the evidence in four steps — analytic validity (can the variant be detected in fetal DNA)?, variant interpretation (can it be classified under ACMG/AMP and ACMG–ClinGen standards)?, clinical validity (how far does it shift the probability of autism)?, and clinical utility (does knowing it improve outcomes)? — which makes explicit the classification step within the conventional three-stage sequence and locates where translation breaks down. A second distinction, between molecular diagnosis of a genetic condition and diagnosis of autism itself, runs alongside it. Recurrent copy-number variants such as 16p11.2 and 15q11–q13 rearrangements, and pathogenic variants in genes such as CHD8, SCN2A, SHANK3, and PTEN, are technically detectable by currently available prenatal diagnostic genomic platforms when such testing is clinically indicated. Their interpretation is frequently incomplete, their clinical validity for autism specifically is limited by incomplete penetrance, variable expressivity, and pleiotropy, and their clinical utility for neurodevelopmental prediction is unestablished. Common polygenic variation contributes substantially to population-level liability, but individual-level discrimination is poor and ancestry portability is limited. Prenatal diagnostic genomic testing is additionally largely indication-driven rather than a population screen for neurodevelopmental risk. Prenatal sequencing may therefore establish a molecular diagnosis of a condition associated with an increased likelihood of autism; this is not prenatal diagnosis of autism. Progress will require ancestrally diverse cohorts, population-based penetrance estimates, longitudinal prenatal-to-childhood follow-up, and probabilistic multimodal models, alongside counseling standards that resist deterministic interpretation.