Has the IACC Finally Embraced Precision Medicine for Autism?

I have been experimenting for the last few days with uploading the EpiphanyASD knowledge base into AI and then asking it to answer questions with this accumulated context in mind.

I am interested in whether this can do something different from simply asking an AI a question about autism.

Instead of treating every question as if it were starting from scratch, can an AI use a large body of material collected over many years—research papers, treatment observations, genetic findings, mechanisms and individual responses—to look at new developments through the particular perspective that has developed on EpiphanyASD?

I am not American, but a major and increasingly controversial topic in the US is the new Interagency Autism Coordinating Committee (IACC) established under HHS Secretary Robert F. Kennedy Jr.

The newly constituted IACC has published its Working Draft Strategic Plan 2026–2028, a huge 336-page document setting out proposed priorities for US autism research, treatment and services.

The original public-comment period was only four days, which produced considerable criticism from autism organisations and other stakeholders. Seven major organisations—including Autism Speaks, the Autism Science Foundation, the Autism Society of America, the National Council on Severe Autism and the Autistic Self Advocacy Network—jointly called for a 90-day extension. The IACC subsequently extended the deadline to August 20, 2026.

So I decided to try something slightly different.

I asked the research copilot running on ChatGPT to read the entire 336-page document and then review it using the EpiphanyASD knowledge base as its background context.

The result was surprisingly interesting.

Rather than simply summarising what the IACC says, I asked it to look for convergence and divergence between the new federal research strategy and the ideas that have emerged from years of research and discussion on EpiphanyASD.

The question was essentially:

Does this new US autism strategy move toward the kind of precision-medicine approach that EpiphanyASD has been exploring for years?

The answer turned out to be yes, in some quite striking ways.

The IACC plan places considerable emphasis on biological heterogeneity, immune and inflammatory biology, folate transport, mitochondrial and redox function, GI and microbiome biology, regression, epilepsy, autonomic dysfunction and—perhaps most importantly—identifying treatment-responsive subgroups rather than assuming that a treatment should work for everyone with an autism diagnosis.

It even explicitly proposes a federal pathway for investigating repurposed medications and for designing clinical trials capable of detecting benefits that might otherwise be hidden in heterogeneous autism populations.

There are also some interesting areas where the EpiphanyASD research agenda goes beyond what is currently included in the IACC plan.

So this is not intended to be a review of whether the new IACC plan is politically good or bad, nor is it an attempt to claim that the IACC has validated treatments discussed on this blog.

Instead, I wanted to see what happens when a large accumulated autism knowledge base is used as the lens through which a major new autism research strategy is examined.

Here is the AI-generated review, with my own comments and editing where appropriate.

I did then take the logical next step and have the Copilot draft an email and pdf attachment with a full list of comments, which I then sent to the IACC. Some of the new members of the IACC are aware of EpiphanyASD already. At least making that 4 day deadline with the Copilot would not be troubling. 10 minutes would be plenty. 

Has the IACC Finally Embraced Precision Medicine for Autism?

An analysis using EpiphanyASD copilot AI of the IACC Strategic Plan 2026–2028

The new IACC plan

The Interagency Autism Coordinating Committee (IACC) has published its Working Draft Strategic Plan 2026–2028 for public review. At 336 pages, it is an enormous document covering autism research, clinical care, services and federal priorities.

I have gone through the therapeutic sections of the document in detail and compared them with the research themes that have repeatedly appeared on EpiphanyASD.

What struck me most was not any particular drug or biological theory. It was the change in the way autism treatment is being conceptualised.

The IACC is increasingly moving away from asking: “Does this treatment work for autism?” and toward: “Does this treatment work for a biologically or clinically identifiable subgroup of autistic people?”

That is a very important change.

Autism may contain many different treatment-responsive subgroups

The new plan explicitly recognises that a treatment can genuinely benefit a subgroup while producing a negative result when tested in an unselected autism population.

The document states that if a drug works for some people but not others, the benefit can be washed out in an all-comers trial. Rather than dismissing such signals, the IACC proposes trials specifically designed to identify and confirm treatment-responsive populations.

This is remarkably consistent with one of the central ideas that has developed on EpiphanyASD.

For years I have argued that autism is unlikely to be a single biological disorder. The behavioural diagnosis may describe a common phenotype while the underlying biology differs considerably between individuals.

One person might have a prominent immune phenotype. Another might have mitochondrial or redox abnormalities. Another might have abnormal folate transport. Another might have significant GI pathology. Another might have epilepsy or abnormal network excitability.

If that is true, then testing a drug across everyone with an autism diagnosis may be a remarkably inefficient way of finding out whether it works.

The 11 therapeutic domains

The plan establishes a series of therapeutic domains:

1. Neurotransmission, neural circuits and neuropsychopharmacology
2. Immune, autoimmune and inflammatory biology
3. Folate metabolism and one-carbon biology
4. Gastrointestinal biology, microbiome function and nutrition
5. Neurodevelopmental regression and functional trajectory
6. Mitochondrial, redox, metabolic and endocrine biology
7. Autonomic dysfunction
8. Sleep and circadian regulation
9. Epilepsy and network excitability
10. Motor function
11. Adaptive trial design and repurposed medications

The important point is that these are not presented as competing explanations for autism. The plan repeatedly emphasises that these biological abnormalities may occur in defined subgroups, and that evidence needs to establish which abnormalities are clinically meaningful and treatment-relevant.

1. Immune and inflammatory biology

The IACC gives immune biology an entire therapeutic domain.

It discusses autoimmune disease, allergic disease, immunodeficiency, inflammatory GI disease, mast-cell disease, post-infectious neuroimmune presentations, neuroinflammation, microglia and astrocytes, cytokines and chemokines, maternal immune biology, autoantibodies and immune-associated regression.

The document does not support the idea that autism is universally an immune disorder. Instead, it says that immune findings may occur in particular clinical subgroups and that the research priority should be identifying those groups.

The important question is no longer simply: “Are cytokines abnormal in autism?”

It is: “Which autistic people have reproducible immune abnormalities, what clinical phenotype do they produce, and do those abnormalities predict treatment response?”

That is a much more useful question.

2. Folate receptor-alpha antibodies and leucovorin

The IACC identifies cerebral folate transport as one of its near-term translational priorities, together with harmonisation of folate receptor-alpha autoantibody assays and confirmatory leucovorin trials in defined subgroups.

The plan distinguishes people with FOLR1 variants and cerebral folate transport deficiency, people with cerebral folate deficiency from other causes, and people with folate receptor-alpha autoantibodies.

The third group remains a candidate subgroup because the assays are not yet sufficiently harmonised and the relationship between antibody status and treatment response requires prospective confirmation.

There is enough evidence to justify a properly designed confirmatory trial, but not enough to say that leucovorin is a treatment for autism generally.

The IACC therefore proposes a sequence of assay harmonisation, subgroup definition, adequately powered multicentre trials, validation of response, and treatment guidance if successful.

This is exactly the sort of progression that precision medicine requires.

3. Mitochondrial and redox biology

The IACC has created a substantial domain covering mitochondrial function, redox biology, metabolism and endocrine physiology.

It discusses oxidative phosphorylation, ATP production, lactate and pyruvate, fatty-acid oxidation, acylcarnitines, carnitine, glutathione, oxidative stress, metabolomics, mitochondrial stress and metabolic vulnerability during physiological stress.

The document distinguishes primary mitochondrial disease from secondary mitochondrial dysfunction or abnormal metabolic stress responses.

This is important because mitochondrial abnormalities should not automatically be interpreted as proof that mitochondria are the cause of someone's autism. They may instead be a susceptibility factor, a consequence of another biological process, a marker of physiological stress, a contributor to particular symptoms, or a predictor of response to a particular intervention.

Determining which of these is true is the research challenge.

NAC and taurine are actually named

There is a particularly interesting detail in this section.

The IACC explicitly lists N-acetylcysteine (NAC) and taurine, alongside L-carnitine, CoQ10, glutathione-directed strategies and riboflavin, as interventions with plausible rationales in selected contexts.

This certainly does not mean that the IACC has concluded that NAC or taurine treat autism. The document says that the evidence is uneven and asks a much more precise question: Which intervention helps which subgroup, based on which biochemical profile, at what dose, and with what measurable improvement?

That is exactly the question that should be asked.

4. Purinergic signalling

One of the more surprising sections concerns purinergic signalling.

The plan discusses extracellular ATP as a danger signal and describes pathways involving P2X and P2Y receptors, P2X7, microglia, calcium signalling, inflammasome activation, CD39, CD73 and adenosine.

It proposes measuring these pathways in defined populations.

This is particularly interesting from an EpiphanyASD perspective because purinergic signalling and the Cell Danger Response have been subjects of discussion on this blog for years.

But the IACC gets the distinction right. Purinergic signalling is presented as an investigational mechanism, not an established explanation for autism and not an immediate indication for antipurinergic treatment.

A biological mechanism can be sufficiently interesting to justify research without being sufficiently proven to justify clinical treatment.

5. Gastrointestinal disease and the microbiome

The IACC has a dedicated domain for Gastrointestinal Biology, Microbiome Function and Nutrition.

The plan recognises that GI problems can manifest as changes in behaviour, sleep or overall functioning and calls for better recognition of ordinary treatable GI disease.

At the same time, it proposes research into mucosal biology, microbial metabolites, enzyme function, microbiome composition, pain, feeding, nutrition, immune-metabolic interactions and diet-responsive interventions.

Importantly, the plan does not recommend treating the microbiome of autistic people indiscriminately. Instead, it proposes precision GI and microbiome trials in defined subgroups.

6. Regression is being treated as a clinical event

Perhaps the most important clinical change in the document concerns regression.

The IACC proposes a national clinical pathway for neurodevelopmental regression and substantial functional loss.

If someone loses previously acquired language, communication, adaptive skills, motor abilities, feeding ability, sleep stability, continence or behavioural regulation, the change should not automatically be attributed to autism. It should trigger appropriate evaluation.

The IACC lists possible contributors including seizures, metabolic problems, mitochondrial vulnerability, immune disease, infection, sleep disruption, GI disease, medication effects, psychiatric deterioration and pain.

Regression is therefore being treated as a clinical phenomenon, not a diagnosis of its cause. There may be multiple types of regression with different biological mechanisms.

The practical consequence is that doctors should investigate the change rather than assume that the underlying autism has simply become more severe.

7. EEG, epilepsy and network excitability

The epilepsy domain goes beyond conventional seizure treatment.

The IACC specifically identifies subclinical epileptiform activity, sleep-potentiated epileptiform discharges, electrical status epilepticus in sleep, altered network excitability and seizure-associated developmental or functional decline.

This is important because some changes that appear behavioural may actually reflect altered brain physiology.

It also reinforces the importance of considering sleep EEG and network activity in appropriate cases rather than assuming that the absence of obvious clinical seizures means that abnormal electrical activity cannot be relevant.

8. Autonomic dysfunction

The IACC identifies dysautonomia, orthostatic intolerance, POTS, abnormal thermoregulation, sweating abnormalities, GI motility abnormalities and abnormal stress-recovery responses.

This is interesting because autonomic physiology provides a possible bridge between several biological systems:

autonomic function ↔ GI function ↔ sleep ↔ immune function ↔ mitochondrial energy metabolism ↔ stress responses.

The plan increasingly treats these systems as interconnected rather than isolated.

9. What about the drugs discussed on EpiphanyASD?

The new IACC plan does not mention every treatment that has been discussed on EpiphanyASD.

For example, searches of the 336-page document did not find bumetanide, NKCC1, CACNA or “calcium channel” under those terms.

That means we should not claim that the IACC has endorsed those particular approaches.

But the methodology of the plan is highly relevant to them.

Bumetanide is a particularly good example. The evidence remains mixed. Some trials and meta-analyses show signals of benefit, while larger or later studies have failed to demonstrate convincing benefit across broad populations.

The obvious question is therefore not necessarily: “Does bumetanide work?”

but: “Is there a subgroup in which bumetanide works substantially better than in the general autism population, and can that subgroup be identified beforehand?”

That is precisely the type of question the IACC's new trial-design framework is intended to address.

The same applies to the calcium-channel hypotheses discussed on EpiphanyASD, including CACNA1C/CACNA2D3 and the possibility of repurposing drugs such as verapamil.

These ideas are not currently represented as priorities in the IACC plan. That is worth saying explicitly. It is one of the areas where EpiphanyASD's research agenda currently goes beyond what appears in this particular federal strategy.

10. The really important section: repurposed drugs and adaptive trials

For me, the most important section of the entire document is Priority Therapeutic Domain 11.

The IACC explicitly recognises the problem of repurposed medications.

Many drugs are already being prescribed off-label. Some have small studies behind them. Some have impressive individual clinical responses. Others fail to replicate.

Yet there is no efficient system for determining whether a credible signal represents a genuine treatment-responsive subgroup.

The IACC proposes a federal pathway for doing precisely this.

The goal is to develop trial designs that can detect treatment effects in heterogeneous populations and provide a route from:

clinical signal → subgroup → trial → evidence → regulatory review.

This is potentially transformative for repurposed drugs.

Why small autism trials can be misleading

Suppose a drug produces 30% major responders, 20% moderate responders and 50% nonresponders.

The average effect may be unimpressive.

A conventional trial could therefore conclude that the drug does not work.

But if the 30% of major responders share a biological characteristic, the correct scientific conclusion might be completely different:

“The treatment doesn't work for everyone—but it may work very well for a particular subgroup.”

The IACC now explicitly recognises this problem.

That is a very important development.

Where EpiphanyASD fits into this

Over the years, EpiphanyASD has accumulated a large collection of observations involving medications, supplements, genetic variants, immune biology, mitochondrial dysfunction, redox biology, GI problems, microbiome changes, EEG abnormalities, regression and metabolic abnormalities.

Some of these observations are strong. Some are weak. Some are anecdotal. Some have subsequently acquired experimental support. Others have not.

They should not all be treated as equivalent evidence.

But the collection has value because it can generate hypotheses.

The challenge has always been converting:

“This unusual person responded dramatically to X”

into:

“What characteristic distinguished this person from the nonresponders?”

That is the step that precision medicine requires.

A new way of looking at the EpiphanyASD database

Rather than asking whether the knowledge base has identified the treatment for autism, it is more useful to ask whether it contains candidate responder phenotypes.

Response to bumetanide → Is there a GABA/ionic-homeostasis subgroup?

Response to leucovorin → Is there a cerebral folate/FRAA subgroup?

NAC response → Is there a redox/oxidative-stress subgroup?

Taurine response → Is there a metabolic/neurotransmission subgroup?

GI-linked behavioural changes → Is there a GI/microbiome/inflammatory subgroup?

Regression after illness → Is there a metabolic, immune or seizure-related subgroup?

EEG abnormalities → Does network excitability predict particular treatment responses?

Particular CACNA variants → Are calcium-channel abnormalities treatment-relevant in a defined genetic subgroup?

Autonomic abnormalities → Do physiological stress-response abnormalities define another subgroup?

These are research questions, not treatment recommendations.

What the IACC plan gets right

I think the strongest aspects of the new plan are fivefold.

1. It recognises heterogeneity. Autism is unlikely to have one biological mechanism.

2. It separates established medical care from experimental biology. An autistic person with epilepsy, inflammatory bowel disease, an endocrine disorder, immune deficiency or another established medical condition should receive ordinary medical care rather than having symptoms attributed automatically to autism.

3. It takes biological subgroups seriously. Immune, mitochondrial, folate, GI and metabolic abnormalities are being considered as potential subgroup characteristics, rather than universal explanations.

4. It recognises the weakness of conventional all-comers trials. A real responder subgroup can disappear statistically when mixed with large numbers of nonresponders.

5. It recognises the potential of repurposed medications. Existing drugs can provide a faster route to treatment than developing completely new molecules—but only if we develop a credible way of determining who benefits.

What is still missing?

The plan is very good at identifying domains, but much harder questions remain.

How do we actually define the subgroups?

A cytokine abnormality is not necessarily a subgroup.

An abnormal metabolite is not necessarily a subgroup.

A genetic variant is not necessarily a treatment biomarker.

A GI symptom is not necessarily a mechanistic subtype.

And an individual response is not proof that the proposed mechanism caused it.

The next step therefore has to be biomarker validation.

The IACC itself repeatedly emphasises assay harmonisation, prospective validation, standardised endpoints and replication before biomarkers are used clinically.

That is exactly right.

The opportunity for autism research

The most encouraging aspect of this document is that it potentially creates a bridge between two worlds that have historically been separated.

On one side we have large-scale academic research and federal funding.

On the other we have clinicians, families and individual treatment experiences.

There are thousands of observations of unusual responses to treatments in autism.

Most will turn out to be noise, placebo effects, regression to the mean, spontaneous fluctuation or coincidence.

But some will not.

The challenge is to develop a scientific system capable of finding the genuine signals.

The IACC's proposed framework—particularly its emphasis on biomarker-defined subgroups, adaptive trials, objective endpoints and repurposed medications—could provide exactly that mechanism.

The big question for EpiphanyASD

After reading all 336 pages, my conclusion is not that the IACC has validated the ideas discussed on EpiphanyASD.

It hasn't.

Nor has it validated NAC, taurine, bumetanide, leucovorin, antipurinergic therapy, mitochondrial treatments, microbiome interventions or any other particular treatment.

What it has done is potentially more important.

It has moved much closer to the research model that EpiphanyASD has been advocating:

Autism is heterogeneous. Identify biologically meaningful subgroups. Find measurable characteristics of those subgroups. Test treatments in the people most likely to respond. Measure meaningful outcomes. Replicate the result.

The convergence is particularly striking in immune biology, folate transport, mitochondrial/redox biology, GI/microbiome function, regression, epilepsy, autonomic dysfunction and purinergic signalling.

And the explicit inclusion of NAC and taurine as examples of mechanism-supported interventions requiring subgroup evidence is notable.

But perhaps the most important development is Domain 11.

For years, one of the biggest problems in autism treatment has been that potentially useful drugs can disappear into the statistical noise of broad clinical trials.

The new IACC plan explicitly acknowledges that problem and proposes developing trials capable of finding treatment-responsive populations.

That is exactly where I believe autism research needs to go next.

Not “What treats autism?”

But:

“Which treatment works for which autistic person—and how can we tell in advance?”

If the IACC can turn that principle into actual biomarker-defined trials, it could represent a genuine change in the way autism treatments are discovered.

And that, rather than any individual drug mentioned in the document, may ultimately be the most important development in this new Strategic Plan.

 

Source note: This article is based on the IACC Working Draft Strategic Plan 2026–2028 and an EpiphanyASD-oriented analysis of its therapeutic domains. It distinguishes research hypotheses from established clinical evidence; the IACC document is a working draft and not an adopted treatment guideline.

From Genes to Mini-Brains: The Future of Personalised Autism Therapy

 I did write about the recent Yale paper that used AI to predict what drugs might be effect in the types of autism they studied.

Epiphany: Looking at the Yale perspective on identifying therapies for the downstream effects of a spectrum of autism genes

The logical follow up is to look at how you can detect effective drugs that the Yale methods missed. That is the subject of today’s post.

Another open issue is to update our knowledge about the use of Rapamycin, highlighted in the Yale paper, and the subject of interesting research at UCLA. That will come in a later post.

If all this sounds complicated, an alternative strategy is the shoebox method. After diagnosis with level 3 autism, the parents can request a shoebox with a small amount of 150 drugs. After completing some scientific instruction and with medical support, they are able to investigate which handful of those drugs meaningfully improve their n = 1 case of autism.

This approach would be consistent to the parent-led drive in the US for the 'Right to Try 2.0' (The Right to Try for Individualized Treatments Act), a legislative framework designed to grant legal access to bespoke, hyper-personalized therapies for children with severe, untreatable neurodevelopmental conditions.

 

Last time I wrote about an important study from Yale University showing that hundreds of autism-associated genes converge on a surprisingly small number of biological pathways. Rather than viewing autism as hundreds of unrelated disorders, the study suggested that many mutations ultimately disturb the same cellular processes.

That naturally raises the next question.

Once we know which pathways are abnormal, how do we identify the best treatment for an individual patient?

Several recent studies suggest the answer may lie in combining multiple technologies rather than relying on genetics alone. Alongside transcriptomics, researchers are now using patient-derived brain organoids ("mini-brains"), multi-electrode arrays (MEAs), machine learning and, in the future, metabolomics to determine not only what has gone wrong, but which treatment is most likely to restore normal brain function.

If successful, this approach could fundamentally change how autism therapies are developed and prescribed.

 

Genes tell us where to look

The Yale study used transcriptomics to examine how different autism mutations alter gene expression.

This approach is extremely powerful because it identifies which biological pathways are disturbed. If a mutation overactivates the mTOR pathway, drugs such as rapamycin become obvious candidates. If calcium signalling is disrupted, calcium channel modulators become logical possibilities. If NMDA receptor signalling is abnormal, drugs affecting glutamate transmission deserve investigation.

Transcriptomics provides a molecular roadmap.

But it cannot tell us whether a drug actually restores normal brain function.

After all, the brain is not simply a collection of genes.

It is an electrical organ.

 

The brain is ultimately an electrical organ

Every thought, every memory and every movement depends upon billions of neurons communicating through electrical impulses.

A treatment might completely normalise gene expression while leaving neuronal circuits functioning abnormally.

Conversely, another drug might produce only modest changes in gene expression while restoring normal neuronal communication.

Ultimately, it is the latter that is far more likely to improve behaviour.

This is why several research groups have begun measuring brain function directly, rather than relying solely on molecular biology.

Researchers collect skin, blood or even urine cells from an individual with autism. These cells are reprogrammed into induced pluripotent stem cells (iPSCs), which are then used to grow tiny brain organoids.

Although these "mini-brains" are vastly simpler than a real human brain, they contain functioning neural networks that spontaneously generate electrical activity.

By placing these organoids onto multi-electrode arrays (MEAs), researchers can record how neurons communicate with one another.

The goal is not to reproduce the whole human brain.

The goal is to determine whether a potential treatment restores healthier neuronal communication.

 

Different levels of the same biological story

It is tempting to ask whether transcriptomics or electrophysiology is the better approach.

I think that is the wrong question.

They are measuring different levels of the same biological cascade.

 

 

The Yale study examines what happens near the top of this cascade.

The organoid studies examine what happens much further downstream.

Transcriptomics identifies candidate therapies.

Functional electrophysiology helps prioritise those candidates by determining which drugs actually restore neuronal network activity.

Clinical trials then determine whether those laboratory improvements translate into meaningful benefits for patients.

Neither approach is sufficient on its own.

Together they provide a much more complete picture.

 

 

Different autism genes produce different electrical fingerprints

A recent autism study compared brain organoids from patients with several syndromic forms of autism, including SHANK3, SCN2A, STXBP1, PPP2R5D and GRIN2B syndromes.

Rather than finding one common "autism signature", each syndrome showed its own pattern of neuronal firing, bursting, synaptic plasticity and network connectivity.

Perhaps even more interestingly, patients carrying mutations in the same gene did not always behave identically.

Different GRIN2B patients, for example, showed distinct electrophysiological profiles despite sharing the same genetic diagnosis.

This is exactly what many clinicians and parents have observed for years.

There is no such thing as a "typical" SCN2A child.

Or a "typical" GRIN2B child.

The specific mutation matters.

The rest of the genome matters.

Environmental influences matter.

All of these factors combine to produce an individual pattern of brain function.

This has profound implications for personalised medicine.

 

One gene may not mean one treatment

It may not be enough simply to say:

"This child has a SHANK3 mutation, therefore everyone with SHANK3 should receive Drug X."

Instead, the future may require testing each patient's own neurons.

Two children carrying mutations in the same autism gene could ultimately require different treatments because their neuronal networks behave differently.

This represents a significant shift in thinking.

Rather than treating the mutation, we may ultimately need to treat the biology of the individual patient.

 

Why transcriptomics alone may not be enough

The Yale approach is a major advance, but it is unlikely to identify every useful therapy.

One useful way to think about autism treatments is to divide them into three broad categories.

 

Bucket 1: Drugs that modify gene expression

These drugs work primarily by altering transcriptional programmes or signalling pathways.

Examples include rapamycin, pioglitazone, statins, telmisartan and corticosteroids.

These are exactly the kinds of therapies that transcriptomic approaches are designed to identify.

 

Bucket 2: Drugs that alter neuronal electrophysiology

Many neurological drugs work very differently.

Rather than changing which genes are expressed, they change how neurons behave electrically.

Examples include bumetanide, calcium channel blockers, sodium channel blockers and GABA-A modulators.

These drugs alter neuronal excitability within seconds or minutes by changing ion movement across cell membranes.

Their principal mechanism is electrophysiological rather than transcriptional.

 

Bucket 3: Drugs that improve cellular metabolism

A third group works primarily through direct biochemistry.

Examples include N-acetylcysteine (NAC), alpha-lipoic acid, taurine and agmatine.

These compounds improve redox balance, mitochondrial function and cellular metabolism without necessarily producing major changes in gene expression.

The exact proportions remain unknown, but the principle is clear.

Transcriptomic screening is naturally best suited to discovering drugs whose primary mechanism involves altering gene expression.

It is less likely to identify therapies whose principal actions are electrical or metabolic.

That is precisely why these emerging technologies should be viewed as complementary rather than competing.

 

Stress-testing the network

Another intriguing finding comes from Johns Hopkins University.

Researchers studying schizophrenia and bipolar disorder found that brain organoids became much easier to distinguish after they applied gentle electrical stimulation.

Many neuronal abnormalities remained hidden while the network was resting.

Only when the network was challenged did disease-specific defects become obvious.

This is remarkably similar to a cardiac stress test, where exercise reveals abnormalities invisible on a resting ECG.

The autism organoid study reached a similar conclusion.

Responses to stimulation and measures of synaptic plasticity distinguished syndromes better than spontaneous firing alone.

Perhaps future drug screening will resemble a cardiac stress test more than a routine blood test.

 

Brain organoids are like ECGs

Brain organoids are undoubtedly a huge simplification of the human brain.

They cannot reproduce language, social interaction or higher cognition.

Nor do they capture the long-range communication between distant brain regions.

Yet perhaps that simplicity is also their strength.

An ECG is also an enormous simplification of the heart.

It tells us nothing about heart valves, coronary arteries or cardiac metabolism.

Yet it remains one of the most valuable investigations in medicine because it measures one of the heart's most fundamental properties—its electrical activity.

Brain organoids may eventually play a similar role.

They will never reproduce the complexity of the human brain, but they may capture enough of its fundamental electrical behaviour to guide personalised treatment.

 

Why autism may be the ideal place to prove the concept

Ironically, autism may be one of the best neurological conditions in which to demonstrate this new approach to precision medicine.

Unlike schizophrenia or bipolar disorder, many forms of autism already have:

• well-defined pathogenic mutations,
• relatively predictable developmental trajectories,
• measurable biomarkers,
• informative animal models,
• an increasing number of candidate therapies.

 

One can imagine a future clinical workflow like this:

1.     Clinical & Genomic Assessment – Diagnose autism and perform genomic analysis to identify pathogenic mutations where present, recognising that many individuals will have no single identifiable genetic cause.

2.     Biological Pathway Analysis – Use transcriptomics, where appropriate, to identify disrupted molecular pathways and suggest candidate therapeutic targets.

3.     Organoid Derivation – Generate patient-derived brain organoids from induced pluripotent stem cells (iPSCs).

4.     Electrophysiological Profiling – Record the organoid's unique neuronal network signature using multi-electrode arrays (MEAs).

5.     Drug Screening – Test a panel of candidate therapies, selected on the basis of genetics, transcriptomics, previous clinical evidence, or drug repurposing studies.

6.     Functional Selection – Identify the treatment that most effectively restores healthy neuronal network activity.

7.     Personalised Treatment – Treat the patient and determine whether clinical improvement correlates with the improvement observed in the patient's own organoids.

Importantly, this workflow does not depend on identifying a single causative mutation. For many autistic people, particularly those with idiopathic autism, genetics may provide only limited guidance. Brain organoids offer a different approach: they measure how an individual's neuronal networks actually function, regardless of whether the underlying cause is a rare mutation, a combination of common genetic variants, environmental influences, or some combination of all three. In that sense, organoids may be particularly valuable for the majority of autistic people who currently lack a clear molecular diagnosis.

If repeated studies showed that normalising the electrophysiology of a patient's own organoids consistently predicted clinical improvement, it would represent a landmark advance.

Patient-derived organoids would become biological avatars for personalised medicine.

 

Will this ever be practical?

At first sight, growing a personalised brain organoid for every autistic individual sounds unrealistic.

Today, it probably is.

Generating induced pluripotent stem cells, growing brain organoids and screening dozens of drugs requires specialist laboratories, takes weeks or even months, and is expensive.

It is difficult to imagine every autistic person undergoing this process in today's healthcare systems.

However, many revolutionary medical technologies began in exactly the same way.

Whole-genome sequencing once cost billions of dollars.

Today it costs only a few hundred dollars and is becoming routine clinical practice.

MRI scanners were once rare research instruments.

They are now standard equipment in hospitals worldwide.

Organoid technology is also likely to become faster, cheaper and increasingly automated.

Even then, personalised organoid testing may never be necessary for everyone.

Initially, it may be most valuable for individuals with rare genetic syndromes, severe neurodevelopmental disorders, or those who have not responded to conventional treatments.

There is another possibility.

As researchers study thousands of patients, machine learning may identify recurring electrophysiological subtypes of autism and link them to treatment responses.

Organoids could be used to discover these subtypes and validate therapies. Once these patterns are established, many future patients might be classified using simpler biomarkers, with personalised organoid testing reserved for only the most complex cases.

 

Bringing the technologies together

Biological layer	Technology	Primary question answered
Genome	DNA sequencing	What mutation is present?
Gene expression	Transcriptomics (Yale approach)	Which molecular pathways are disrupted?
Neuronal function	Brain organoids + MEAs	How are neuronal networks functioning, and does a drug restore normal activity?
Cellular metabolism	Metabolomics / high-content imaging	Has mitochondrial function and cellular metabolism recovered?
Clinical outcome	Patient	Does the treatment actually improve symptoms?

Rather than competing approaches, these technologies complement one another.

Each answers a different question.

Together they provide the first realistic framework for truly personalised autism therapy.

 

The final proof still remains

There is one critical experiment that has yet to be performed.

If a drug restores normal electrical activity in a patient's brain organoid...

does that same patient improve clinically?

Nobody yet knows the answer.

If the answer proves to be yes, we may look back on these studies as the beginning of a new era in autism research.

The Yale study showed us where to look.

Brain organoids may help us decide what to do.

Perhaps the most important lesson is that a genetic diagnosis is only the beginning.

Even two children carrying mutations in the same autism gene may have different neuronal network abnormalities and therefore require different treatments.

The future of precision medicine may become so personalised that we no longer ask:

"What drug works for SCN2A?"

Instead we ask:

"What drug restores healthy neuronal function in this particular child?"

If patient-derived brain organoids can one day answer that question before treatment even begins, they will have transformed not only autism research, but the practice of personalised medicine itself.

When might this become reality?

Whenever a new technology is discussed, the obvious question is:

"When might this actually become available?"

The answer depends on whether we are talking about an individual research project or a routine hospital service.

 

Today (2026–2030): Proof of Concept

In many ways, the first stage has already arrived.

Researchers can already:

• generate patient-derived brain organoids,
• record neuronal network activity using multi-electrode arrays,
• demonstrate that different autism syndromes produce distinct electrophysiological signatures,
• and test small numbers of candidate drugs on organoids from individual patients.

This is the classic "n = 1" approach.

Although still largely confined to research laboratories, the scientific foundations have now been established.

The challenge is no longer proving that the technology works—it is making it practical, reproducible and affordable.

 

The Next Decade (2030–2035): Early Clinical Translation

The next major milestone will be demonstrating that normalising a patient's brain organoid predicts clinical improvement.

If repeated studies consistently show that laboratory improvements correlate with patient outcomes, specialist centres such as Kennedy Krieger Institute, Boston Children's Hospital, Great Ormond Street Hospital and similar academic centres could begin offering organoid-guided treatment for carefully selected patients.

Initially, this would probably be limited to:

• rare genetic syndromes,
• severe neurodevelopmental disorders,
• treatment-resistant patients,
• and prospective clinical research studies.

At this stage, throughput would still be relatively low, with perhaps hundreds rather than thousands of patients each year.

 

Beyond 2035: Scaling Precision Medicine

The biggest challenge is unlikely to be biology.

It is engineering.

Growing brain organoids, recording electrophysiological activity and testing dozens of drugs currently requires highly skilled scientists working in specialist laboratories.

For this technology to become routine, much of the process will need to become automated.

As robotics, artificial intelligence and laboratory automation continue to improve, it is easy to imagine integrated platforms that routinely perform:

• whole-genome sequencing,
• transcriptomics,
• metabolomics,
• patient-derived organoid generation,
• multi-electrode array recordings,
• and AI-assisted analysis.

At this stage, specialist hospitals might begin processing thousands of patients each year rather than just a handful.

 

The Long-Term Vision (2040 and beyond)

Perhaps the greatest irony is that the ultimate success of organoids may reduce the need to grow them.

As researchers accumulate data from tens of thousands of patients, artificial intelligence may begin recognising recurring biological patterns that predict treatment response.

Patient-derived organoids would then become the training ground for precision medicine.

Instead of growing an organoid for every patient, clinicians might increasingly rely on AI models trained using years of organoid data, reserving personalised organoid testing for unusual or particularly difficult cases.

For many patients, the entire process might eventually begin with nothing more than a blood or urine sample.

From that single sample it may become possible to perform genomic sequencing, analyse molecular pathways, generate induced pluripotent stem cells, grow patient-specific brain organoids, measure neuronal network activity and identify the treatments most likely to restore healthy brain function.

That vision remains ambitious, but many of the individual technologies already exist. The challenge now is bringing them together into a single, reliable clinical workflow.

 

My prediction

If I had to make one prediction, it is this:

The first routine clinical use of patient-derived brain organoids will probably not be to discover entirely new drugs.

It will be to choose more intelligently between the drugs we already have.

Many of the treatments currently being investigated for autism—including bumetanide, verapamil, pioglitazone, rapamycin, memantine and N-acetylcysteine—already exist.

The real challenge is identifying which patient is most likely to benefit from which treatment.

If brain organoids can answer that question, they will have transformed precision medicine long before they discover the next breakthrough drug.

 

Or the shoebox today?

General Medical Disclaimer

The "shoebox method" and the testing of unapproved or repurposed compounds carry significant health and toxicological risks. This concept represents an experimental paradigm and does not constitute actionable medical advice. Any drug screening or administration must be strictly supervised by a licensed pediatric neurologist or qualified clinical team.