No single blood test, brain scan, or genetic screen can yet diagnose autism, but researchers are closing in on a range of biological signals that appear months or even years before behavioral signs become obvious. The gap between what families experience and what science can offer remains wide: studies of real-world health care data show an average delay of more than two years from a child’s first developmental screening to an autism diagnosis. That delay matters because early intervention consistently produces better outcomes. The push to find reliable biomarkers is, at its core, an effort to shrink that gap and catch autism earlier, more objectively, and with fewer children slipping through the cracks.
Why Current Screening Falls Short
Autism is still diagnosed by observing behavior. Clinicians look for differences in social communication, restricted interests, and repetitive actions, but these behaviors often don’t fully emerge until a child is two or three years old. The most widely used screening tool, the Modified Checklist for Autism in Toddlers with Follow-Up (M-CHAT-R/F), has sensitivity and specificity above 90%, which sounds impressive. The catch is that autism’s prevalence in the general population sits around 2%, and that low base rate drags the tool’s positive predictive value down to roughly 33%. In practical terms, only about one in three children flagged by the screener actually has autism, meaning two out of three are false positives who then undergo unnecessary evaluation and parental anxiety.1PubMed. Early Detection and Diagnosis of Autism Spectrum Disorder: Why Is It So Difficult?
Meanwhile, pediatrician compliance with the American Academy of Pediatrics recommendation to screen at 18 and 24 months remains patchy, and the U.S. Preventive Services Task Force has declined to endorse universal screening for autism in toddlers, citing insufficient evidence that it improves outcomes at a population level. The result is a system that leans heavily on parent concern and clinician judgment, both of which vary enormously. Data from a large national health research network found an average delay of over two years from first screening to diagnosis, with no meaningful differences by sex, race, or ethnicity.2PubMed Central. Delay from Screening to Diagnosis in Autism Spectrum Disorder: Results from a Large National Health Research Network Biomarkers could, in theory, offer an objective shortcut through this bottleneck.
Where Babies Look
Eye tracking is among the most developed biomarker approaches. Infants can sit in a parent’s lap while a camera-equipped screen records where they look, for how long, and how quickly they shift attention. Researchers have been studying this in babies who have an older sibling with autism and are therefore at higher genetic risk. A meta-analysis synthesizing 35 studies of eye tracking in high-risk infants under two found that certain stimulus features reliably distinguished babies who later received an autism diagnosis. Dynamic stimuli and socially relevant regions of a scene, like the target of another person’s gaze or the object someone is pointing to, were particularly sensitive markers.3PubMed. Evaluating the validity of eye-tracking tasks and stimuli in detecting high-risk infants later diagnosed with autism: A meta-analysis
One pattern that stands out is attention disengagement: how quickly a baby moves their gaze from one stimulus to another. Infants who later develop autism tend to have more difficulty disengaging attention, and tasks measuring this could also pick up babies with broader autism-related traits who don’t end up meeting full diagnostic criteria. A multi-method comparison study examining infant social attention found that eye tracking showed promise for identifying very early markers associated with elevated autism likelihood, even before behavioral signs become clear to a parent or clinician.4PubMed Central. Infant Social Attention Associated with Elevated Likelihood for Autism Spectrum Disorder: A Multi-Method Comparison The appeal of eye tracking is that it’s passive, fast, and doesn’t require any cooperation from the infant beyond looking at a screen.
Brain Imaging in the First Year of Life
Some of the most striking biomarker findings come from MRI studies of infants. A prospective study of 148 infants, 106 at high familial risk and 42 at low risk, revealed that the brains of babies later diagnosed with autism showed unusual cortical surface area expansion between six and twelve months, followed by brain volume overgrowth from twelve to twenty-four months. A deep learning algorithm trained on surface area data from MRI scans at six and twelve months predicted individual autism diagnoses at twenty-four months with a positive predictive value of 81% and sensitivity of 88%.5PubMed Central. Early brain development in infants at high risk for autism spectrum disorder Those are numbers that would represent a substantial improvement over current behavioral screening, though they come from a study of high-risk infants, not the general population.
Another imaging finding involves extra-axial fluid, the cerebrospinal fluid sitting in the space over the brain’s surface, especially the frontal lobes. Infants who later developed autism had significantly more of this fluid as early as six to nine months, and the amount at that age predicted symptom severity at the time of diagnosis.6Brain. Early brain enlargement and elevated extra-axial fluid in infants who develop autism spectrum disorder These brain differences are appearing months before parents or doctors typically notice anything unusual.
Brain imaging with functional near-infrared spectroscopy (fNIRS), which uses light to measure blood flow changes in the brain and is much more infant-friendly than MRI, has added to the picture. A study of five-month-old infants at elevated likelihood for autism found overgrown local brain connections compared to low-risk peers, suggesting inefficiencies in brain network organization within the first year of life.7PubMed. Atypical brain network development of infants at elevated likelihood for autism spectrum disorder during the first year of life Preliminary fNIRS work during naturalistic social interactions found that high-risk infants showed a drop in brain connectivity during social periods, the opposite pattern of what you’d expect for engaged social processing.8PubMed Central. Exploring cortical activation and connectivity in infants with and without familial risk for autism during naturalistic social interactions: A preliminary study
Electrical Brain Signatures
EEG, which measures brain electrical activity through sensors on the scalp, offers another non-invasive window. Resting-state EEG studies in people with autism suggest a distinctive pattern: excess power in both low-frequency and high-frequency brain wave bands (a U-shaped profile), abnormal connectivity between brain regions, and enhanced power in the left hemisphere.9PubMed Central. Resting state EEG abnormalities in autism spectrum disorders The Autism Biomarkers Consortium for Clinical Trials, a large multi-site effort, evaluated several EEG-based candidate biomarkers and found that three assays, including resting state, a faces task, and a visual evoked potential task, showed promising acquisition rates and construct performance. Several of these measures had moderate stability when tested again six weeks later, a basic requirement for any useful biomarker.10PubMed Central. The Autism Biomarkers Consortium for Clinical Trials: Initial Evaluation of a Battery of Candidate EEG Biomarkers
EEG is attractive for practical reasons: it’s portable, relatively cheap, and doesn’t require sedation. The challenge is that the brain signatures identified so far tend to overlap with other neurodevelopmental conditions, making it difficult to distinguish autism-specific patterns from broader developmental differences.
Molecular Clues in Blood, Saliva, and Cord Blood
Researchers are also hunting for molecular signals that could serve as early flags. One line of work has examined DNA methylation in cord blood, the chemical tags that sit on DNA and influence which genes are active. A study found a distinct methylation signature in cord blood collected at birth from babies who were later diagnosed with autism, with differences concentrated over genes involved in early brain development and X-linked genes.11PubMed Central. Cord blood DNA methylome in newborns later diagnosed with autism spectrum disorder reflects early dysregulation of neurodevelopmental and X-linked genes The fact that these marks are detectable at birth, years before diagnosis, is tantalizing.
Saliva offers another molecular window. Small RNA molecules called microRNAs circulate in saliva and can be measured non-invasively. A panel of four salivary microRNAs distinguished children with autism from peers with moderate accuracy, with area-under-the-curve values around 0.69 to 0.73.12PubMed Central. Saliva MicroRNA Differentiates Children With Autism From Peers With Typical and Atypical Development Earlier pilot work identified 14 differentially expressed salivary microRNAs in children with mild autism, many of which are expressed in the developing brain and correlated with adaptive behavior measures.13PubMed Central. Salivary miRNA profiles identify children with autism spectrum disorder, correlate with adaptive behavior, and implicate ASD candidate genes involved in neurodevelopment These findings have been partially replicated in different populations, including a study in Bosnia and Herzegovina suggesting that salivary microRNAs could complement standard developmental screening.14PLOS ONE. Identification of developmental disorders including autism spectrum disorder using salivary miRNAs in children from Bosnia and Herzegovina None of this is anywhere close to a clinical test yet, but the non-invasive collection method makes it appealing for very young children.
Metabolomics, the study of small molecules circulating in blood and urine, has revealed shifts in amino acid metabolism, lipid profiles, and markers of oxidative stress in people with autism.15PubMed Central. Profiles of urine and blood metabolomics in autism spectrum disorders A study comparing plasma and fecal metabolites found differences in amino acid, lipid, and xenobiotic metabolism that pointed to oxidative stress, mitochondrial dysfunction, and altered microbial metabolites.16PubMed Central. Plasma and Fecal Metabolite Profiles in Autism Spectrum Disorder These metabolic signatures are interesting because they may reflect underlying biological processes rather than just genetic risk, but they also overlap with profiles seen in other conditions.
Immune Signals and Inflammation
Immune system differences have been studied in autism for decades, and a large meta-analysis covering 54 cytokines found that autistic people had elevated levels of several immune signaling molecules, including IL-1β, IL-6, IL-8, TNF-α, and interferon-gamma, compared to controls.17Molecular Psychiatry. Atypical cytokine profiles in people on the autism spectrum: a comprehensive systematic review and meta-analysis including 54 cytokines However, individual studies sometimes produce conflicting results. One study of children with autism found that IL-10 and IL-8 were actually lower in the autism group, while several other cytokines showed no significant difference from controls.18PubMed Central. Correlation of biochemical markers and inflammatory cytokines in autism spectrum disorder (ASD) The inconsistency across studies illustrates a recurring challenge: autism is not one condition with one biology. What shows up in a meta-analysis as a group-level difference may reflect genuine variation across autistic subgroups rather than a universal marker.
Maternal Antibodies as a Prenatal Flag
One of the more unusual biomarker approaches focuses not on the child but on the mother. Maternal autoantibody-related autism (MAR-ASD) refers to a subtype in which the mother produces antibodies that react to proteins highly expressed in the developing fetal brain. Researchers developed a blood test for patterns of reactivity against eight specific brain proteins and found that certain antibody combinations were found exclusively in mothers of children with autism, never in mothers of typically developing children. The three main patterns were associated with substantially increased odds of an autism diagnosis in the child, and reactivity to one protein in particular (CRMP1) roughly doubled the odds that the child would have a higher severity score on a standard diagnostic measure.19Molecular Psychiatry. Risk assessment analysis for maternal autoantibody-related autism (MAR-ASD): a subtype of autism
A pilot study found that MAR-ASD was present in about 24% of the sample and that children of mothers with these autoantibodies had more severe autism symptoms.20PubMed Central. Pilot Study of Maternal Autoantibody Related Autism (MAR ASD) This wouldn’t be a universal test, since most autistic children don’t fall into the MAR-ASD subgroup, but it could eventually identify a subset of pregnancies at elevated risk, potentially even before birth.
Cry Acoustics, Movement, and Digital Tools
Some researchers are looking at the earliest sounds a baby makes. Preliminary evidence suggests that infants who later receive an autism diagnosis may produce cries with atypical acoustic features as early as six months.21PubMed Central. Atypical Cry Acoustics in 6-Month-Old Infants at Risk for Autism Spectrum Disorder Some researchers have gone further, suggesting that atypical crying may not just be an early marker but could itself influence development by altering the social feedback loop between infant and caregiver.22International Journal of Neuropsychopharmacology. Cry, Baby, Cry: Expression of Distress As a Biomarker and Modulator in Autism Spectrum Disorder This idea, that a biomarker might also be a contributing factor, makes the picture more complex but also more interesting from a therapeutic standpoint.
Computer vision and machine learning are being applied to video of infants, tracking facial features and body movements during clinical assessment tasks. Algorithms developed to measure infant responses during the Autism Observation Scale for Infants have shown the ability to capture critical behavioral observations that compare favorably with expert clinicians.23PubMed Central. Computer vision tools for low-cost and noninvasive measurement of autism-related behaviors in infants Separately, a scoping review examined video-based approaches to automating the General Movement Assessment, a technique used for very young infants in which trained observers watch for specific movement patterns.24PubMed Central. The future of General Movement Assessment: The role of computer vision and machine learning – A scoping review The practical advantage of digital tools is enormous: a smartphone video could theoretically be uploaded and analyzed anywhere in the world, bypassing the need for specialist clinicians entirely.
The Gut Microbiome as a Biomarker Source
Evidence increasingly links gut microbiota differences to autism through what’s called the microbiota-gut-brain axis. Children with autism often show gut bacterial profiles that differ from neurotypical peers, and these microbial signatures are being explored for potential roles in early diagnosis and even presymptomatic risk prediction.25Cell Reports Medicine. Microbiome biomarkers in autism spectrum disorder: Toward prediction, diagnosis, and prognosis Whether the gut differences cause, worsen, or merely accompany autism remains unclear. The clinical causal evidence is still indirect, and microbiome composition is notoriously sensitive to diet, medication, geography, and age, all of which could confound results. Still, the microbiome is one of the few biomarker sources that could theoretically be modified with interventions like dietary changes or probiotics, which is why it attracts so much attention.
Why No Biomarker Has Crossed the Finish Line
With so many promising leads, it’s fair to ask why none of them are in routine clinical use. A systematic review of candidate diagnostic biomarkers for neurodevelopmental disorders in children couldn’t find any single biomarker with evidence, from two or more independent research groups with consistent results, of both sensitivity and specificity reaching at least 80%.26PubMed Central. Candidate diagnostic biomarkers for neurodevelopmental disorders in children and adolescents: a systematic review That’s a sobering benchmark. Most of the individual findings described above come from studies of high-risk infants, relatively small samples, or single research groups. The jump from “this signal differs between groups on average” to “this signal reliably identifies autism in an individual child” is one of the hardest transitions in biomedical research.
Part of the difficulty is biological. Autism is genetically heterogeneous. Research on polygenic risk scores has found that common genetic variants contribute to autism risk in some populations, while rare de novo mutations may play a larger role in others, such as those born preterm.27PubMed Central. Polygenic risk for autism spectrum disorder based on four group comparison across term and preterm birth A large Nature study found that common genetic variants influenced the age at which autism was diagnosed, but rare de novo variants did not show the same association, possibly because co-occurring intellectual disability in some carriers led to later or differently-timed diagnosis.28Nature. Polygenic and developmental profiles of autism differ by age at diagnosis This genetic complexity means a one-size-fits-all biomarker may not exist. The most realistic path forward likely involves combining multiple biomarker types into panels, for instance pairing an eye-tracking measure with a blood-based molecular assay and family history, to improve accuracy beyond what any single signal can achieve.
Sex Differences and Missing Diagnoses
The well-known male skew in autism prevalence, currently around three to four boys for every girl, is itself partly a biomarker problem. Research suggests that females may require a higher burden of genetic risk variants to develop autism, a phenomenon called the female protective effect. This means that autistic girls and women often present differently, with fewer restricted and repetitive behaviors and fewer externalizing problems, making them harder to identify with tools calibrated primarily on male presentations.29PubMed Central. Sex differences in autism spectrum disorders
This has implications for biomarker development. One study examining movement patterns found that females with autism showed distinct movement signatures compared to males with autism, and that these differences were easier to spot when compared against typically developing females rather than a mixed reference group.30PubMed. Strategies to develop putative biomarkers to characterize the female phenotype with autism spectrum disorders If biomarker tools are developed and validated primarily on male-heavy samples, which most of the current research has been, they risk perpetuating the same detection gap that behavioral screening already has. Sex-specific reference ranges or separate validation in female populations will likely be necessary.
Ethical Terrain of Predicting Autism Before It Appears
The ability to predict autism before behavioral symptoms appear raises difficult questions. Because prediction is probabilistic, some children flagged by biomarkers won’t develop autism, and some with autism will be missed. False positives could trigger unnecessary surveillance, parental anxiety, and early labeling that follows a child through medical and educational systems.31PubMed Central. Ethical dimensions of translational developmental neuroscience research in autism False negatives are equally concerning, because early identification only helps if it catches the right children.
Genetic testing raises its own set of worries. If newborn genetic screening for autism risk becomes available, the knowledge of a predisposition could burden families with anxiety about a future that may never arrive, and could create problems with medical insurance, school placement, or social stigma.32Journal of Medical Ethics. Should newborn genetic testing for autism be introduced? Many in the autistic community have also raised concerns that framing autism purely as something to be detected and prevented undermines acceptance of neurological diversity. Any biomarker test that reaches the clinic will need to navigate not just technical validation but also these social and ethical considerations, and the answers won’t come from laboratories alone.
Economic Realities of Biomarker-Driven Diagnosis
Even when a biomarker works, it has to be practical and affordable enough to deploy at scale. Brain MRI on every infant is a non-starter: it’s expensive, often requires sedation, and isn’t available in most settings worldwide. Eye tracking hardware is getting cheaper, but still needs trained operators and standardized protocols. Blood and saliva tests could eventually be cheap, but the laboratory infrastructure for microRNA panels or metabolomics profiles doesn’t yet exist in routine pediatric settings. An economic analysis of genome and exome sequencing in children already diagnosed with autism found that strategic integration of sequencing could be cost-effective, but flagged long wait times for genetic services and uncertain clinical utility as practical barriers.33PubMed. Cost-effectiveness of Genome and Exome Sequencing in Children Diagnosed with Autism Spectrum Disorder The tools that ultimately reach widespread use will probably be the ones that can run in a standard pediatric office visit, at a cost insurance systems are willing to cover, with results interpretable by a general pediatrician rather than a specialist. Smartphone-based video analysis and point-of-care saliva tests are, for that reason, attracting as much interest as the flashier neuroimaging approaches.