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.