Autoimmune Diseases Are Commonly Misdiagnosed
According to a study by the American Autoimmune Related Disease Association, individuals see an average of (nearly) 5 different physicians over 4.6 years before receiving an accurate autoimmune disease diagnosis.
This comes as no surprise to members of our community. Many of whom, with neuroimmune axis conditions such as PANDAS/PANS, ME/CFS, and mast cell activation syndrome (MCAS), to name a few, experience similar delays in diagnosis, in addition to being gaslit by physicians who tell them their symptoms are “in their head.”
Neuroimmune axis conditions are those defined by a dysregulated communication between the immune system and the nervous system. These two systems are closely interconnected allowing our nervous system to respond to immune system cues and visa versa. The medical conditions listed blow are often associated with a range of psychiatric symptoms. Thus, patients are commonly told their symptoms are psychosomatic, implying they are “in their head.”
Could Misdiagnosing Autoimmunity Become A Thing of The Past?
What if there was a clinical diagnostic test that blended DNA sequencing with artificial intelligence to identify what autoimmune disease you might have with near 99% accuracy? This may be possible thanks to researchers at Stanford University who leveraged artificial intelligence to characterize what Eric Topol dubs, ‘the Immunome’, the vast repertoire of T cell and B cell receptor sequences unique to specific autoimmune conditions.
This blog post is designed to educate a non-technical audience about the recent immunome proof-of-principle manuscript and serve as a thought experiment, incorporating speculative insights into how comparable machine learning (AI) models might be trained by scientists to diagnose select neuroimmune axis and autoimmune conditions such as those listed below.
- Autoimmune forms of Psychosis
- Autoimmune forms of Arthritis
- Autoimmune forms of OCD
- Rheumatoid Arthritis (RA)
- Autoimmune Encephalitis
- Post Infectious ME/CFS
- Multiple sclerosis (MS)
- Sjögren’s Syndrome
- Ulcerative Colitis
- Crohn’s Disease
- PANDAS/PANS
- Celiac Disease
- Long COVID
- Psoriasis
- Cancer
While this study didn’t use the technology to diagnose the medical conditions from above, the same AI approach could theoretically learn their unique immune signatures, given enough patient data. Such a tool would not only reduce the rate of misdiagnosis but also improve patient outcomes for a range of conditions while reducing medical gaslighting by taking the guesswork out of certain “invisible” medical conditions.
But first, let’s start by understanding why many of the conditions listed above are commonly misdiagnosed as anxiety- or stress-related.
Are you ready? Let’s dive in.
The Difficulty Diagnosing Autoimmune Diseases and Infection Associated Chronic Conditions (IACC).
We at the Brain Inflammation Collaborative exist as a non-profit because many neuroimmune axis conditions are exceptionally difficult for medical professionals to diagnose. This traps patients in healthcare quicksand for years before they can find an accurate diagnosis and the proper treatment. Here are a few reasons why.
Symptom Overlap
First, the symptoms of these conditions often overlap, increasing the difficulty of diagnosis. For example, brain fog, fatigue, and neurological symptoms such as headache, numbness, or tingling are shared symptoms of autoimmune encephalitis, Long COVID, ME/CFS, SLE, and Sjögren’s syndrome. In addition, the high prevalence of psychiatric disorders experienced by those with the neuroimmune axis and autoimmune diseases also complicates the diagnosis, leading the medical professional to speculate the symptoms are psychosomatic, especially if preliminary testing is negative.
This happens quite often in patients suffering from autoimmune forms of psychosis, where symptoms like hallucinations or mood swings might be mistaken for primary mental health issues like schizophrenia or bipolar disorder. Without a clear biomarker to confirm the autoimmune root, these patients can face delays in proper treatment, as doctors may lean toward psychiatric explanations rather than digging deeper into an immune-based cause, leaving the true condition undiagnosed or dismissed. A neurologist can run a test to identify an autoantibody that might be attacking the brain. But even then, the tests aren’t foolproof. Some patients don’t show these antibodies despite having the condition, and not all doctors agree on which antibodies to look for or how to interpret the results, leaving many stuck in a diagnostic gray area with treatments that don’t fully address the underlying issue.
Incompletely Understood Disease Pathology
The mechanisms behind these conditions involve intricate interactions between the immune and nervous systems, which are not fully understood and can be challenging to detect with standard diagnostic tools. For instance, no single Long COVID test exists, and standard lab tests (i.e. CRP, complete blood count, etc.) often return normal.
Lack Of Awareness
There is a lack of widespread awareness and education about neuroimmune axis conditions, both among healthcare providers and the general public, further complicating timely and accurate diagnosis. For instance, many general practitioners, not to mention psychiatrists, are not taught that chronic inflammation can impact our thoughts, behavior, and mental health.
Lack Of Diagnostic Tests
If you present with signs and symptoms of a viral infection, a medical professional will run a test to detect what virus is causing the symptoms. Yet, many neuroimmune axis conditions lack a test that is widely agreed upon by medical experts in the field. For instance, PANDAS/PANS lacks a test that is broadly agreed upon by other PANDAS/PANS experts. This forces pediatricians to exclude every other possible cause of the symptoms before they can ‘rule in’ the possibility of PANDAS/PANS.
Doctors face the same challenge with diagnosing Long COVID and ME/CFS. If you present with crushing fatigue or brain fog after a virus, there’s no quick test to determine if you have Long COVID or ME/CFS. Instead, they have to play detective by running blood tests, checking your thyroid, or looking for heart issues to rule out other culprits. Only after crossing off those possibilities can they start to say it might be Long COVID or ME/CFS, which can feel like a long, frustrating process for the patient since there’s no clear shortcut to the diagnosis.
The Results Explained
We are particularly excited about a new proof-of-concept framework published recently in Science that could theoretically be applied to numerous difficult-to-diagnose diseases, such as those many in our audience suffer from. However, before we begin, we want to pump the brakes and remind our readers that it’s easy to dream up future use cases beyond what was published in the manuscript. However, it’s difficult to apply these concepts to solve real-world problems at scale because diverse patient data and validation for rare conditions are needed to avoid missteps.
That being said, this is an exciting step in the right direction.
Without further ado, here is what they found.
Combining DNA sequencing with artificial intelligence, scientists identified with near 98.6% accuracy (in the specific experimental context) which of the 593 blood donors in their study were healthy controls, had HIV, COVID-19, Lupus, Type 1 Diabetes, or were recently vaccinated for the flu. Here is how they did it.
They isolated the white blood cells from blood samples and sequenced the DNA of every B cell receptor (BCR) and T cell receptor (TCR). To understand why this is so cool, we must learn some basic B cell and T cell receptor biology.
B and T cell Receptors Explained
Both of these receptors are custom-built to fight infections and cancer. Think of the building blocks of each receptor, amino acids, as “Lego” pieces. Each of the 20 different Lego pieces (there are 20 amino acids) has a different size, shape, and charge. Rearranging these Lego pieces in different combinations provides each receptor with the endless variability needed to recognize every pathogen or cancer cell that your immune system is exposed to.
Here is the key. The amino acid building blocks are not rearranged like “Lego” pieces that can be swapped around on the fully assembled protein receptor. Instead, the genes that encode each BCR and TCR are rearranged (and mutated). Here is an analogy that might help.
TCRs and BCRs are Similar to Assembled “Lego” Structures
Think of each TCR and BCR (pictured below) as a pre-built “Lego” structure straight from the factory. These pre-assembled “Lego” structures are not put together by hand. Instead of rearranging the “Lego” pieces by removing one and swapping it for another by hand, the genetic code given to the 3D printer is changed so a different “Lego” structure is manufactured.
This is how it works inside the cell. The genes that encode the BCRs and TCRs are rearranged. These DNA code modifications change which amino acids will be used to assemble the BCR and TCR, resulting in receptors with different biochemical properties.
Variable Regions, Not Constant Regions, Are Rearranged
It is not the entire B cell or T cell receptor that is rearranged. It’s only the variable region of the receptor (shown below). The constant regions in grey don’t change.
Each B cell and T cell Expresses Unique BCRs and TCRs
Each T cell and B cell expresses one variety of receptor. In other words, there are multiple Lego receptors on individual T cells and B cells. But each receptor belonging to the B and T cells is genetically identical. The reason is that each cell has a single set of DNA instructions that encode the BCR or TCR.
Figure Legend, Left Side: The upper right corner shows two types of T cells, naive and mature T cells (blue and orange respectively). Naive T cells have yet to rearrange their TCR. This is needed to recognize a specific foreign molecule from a pathogen or a cancer cell. That’s why the TCR is shown in all blue, indicating it has yet to rearrange. In contrast, the bottom left quadrant shows the same TCR after it’s been rearranged with different colored “Lego” pieces. To the right of the bottom left yellow box, you will see that each TCR specializes in recognizing a tiny piece (antigen) of a pathogen (or cancer cell). In this case, it’s a virus. Figure Legend, Right Side: The upper right corner shows two types of B cells, naive and mature B cells (blue and orange respectively). Naive B cells have yet to rearrange their BCR. This is needed to recognize a specific foreign molecule from a pathogen or a cancer cell. That’s why the BCR is shown in all orange, indicating it has yet to rearrange. In contrast, the bottom left quadrant shows that the same BCR has rearranged, as shown by the yellow and purple segments. The bottom far right diagram shows that T cells help naive B cells rearrange these “Lego” pieces, creating memory B cells that can differentiate into plasma cells and make large amounts of antibodies. APC= antigen-presenting cells such as macrophages, dendritic cells, B cells, etc.
Why TCRs and BCRs Are Rearranged
The immune system relies on TCRs and BCRs to recognize antigens— or molecules targeted by the immune system. The variable regions of these receptors, which directly bind to antigens, must be rearranged and mutated because each pathogen is distinct, and pathogen targets constantly change. Therefore, the body does not pre-make receptors for every possible threat.
This rearrangement randomly alters the amino acid sequence of the receptor, giving it an equal chance of binding more tightly or weakly to the antigen. Only mutations that improve binding strength are kept through a selection process. For TCRs, this binding happens when antigens are presented on human leukocyte antigen (HLA) molecules, ensuring precise recognition (read our HLA deep dive for more information).
The Results Revisited
Now that we understand that T cell and B cell receptor genes are rearranged to tailor-make receptors that specialize in recognizing fragments of foreign proteins, let’s revisit the results.
The scientists reported that there were, “16.2 million [unique] BCR…clones and 23.5 million [unique] TCR … clones systematically collected from peripheral blood samples of 593 individuals, including patients diagnosed with Covid-19 (63 [people]), human immunodeficiency virus (HIV) infection (95 [people]) (13 [people]), SLE (86 [people]), and T1D (92 [people]), as well as influenza vaccination recipients (37 [people]) and healthy controls (220 [people]).
“This means there was an average of 39,629 T cell receptors and 27,318 B cell receptor genes per person. Let’s assume each gene encodes a unique receptor for simplicity. That means, on average, 66,947 different DNA sequences must be analyzed per person. Now we can appreciate why an AI model was used to learn which specific DNA sequences (corresponding to TCRs and BCRs) are enriched in known disease or immunological states. Specifically, they developed a machine learning framework to analyze the data, which they referred to as Mal-ID (machine learning for immunological diagnosis). The tens of thousands of DNA sequences per person are fed into the Mal-ID framework so it can learn which sequences are found in Lupus, T1D, HIV infection, COVID infection, or healthy controls.
AI Training
They used the same 593 people, whose disease states were known to the researchers (and the AI), and generated large datasets to train the machine learning framework. They next re-analyzed the blinded samples to test the efficacy of the Mal-ID framework. In this particular experimental context, they could identify the immunological state of each person with 97% accuracy (0.97, where 1 is perfect, B, C) by analyzing the T cell receptor and B cell receptor genes in combination.
Current Technology Status
Dr. Eric Topol reported that the primary authors agree that BCR and TCR sequencing can be done inexpensively at scale, in batches with a few days turnaround. While there is no start-up company yet, the authors have filed patent applications according to the same interview summary.
Conclusion
For too long, millions with autoimmune and neuroimmune conditions—like PANS, MCAS, or Long COVID—have wandered through years of misdiagnosis, dismissed as “all in their heads” while racking up costs and heartbreak. The Stanford team’s AI-driven leap into the immunome isn’t a magic fix yet, but it’s a beacon of what’s possible: a future where a simple blood test could slash those delays, save billions, and hand patients the proof they need to reclaim their lives. It won’t happen overnight—more data, research, and grit are needed to extend this to the full spectrum of diseases we care about. But for a community used to fighting in the dark, this is a spark worth nurturing. Together, patients, advocates, and scientists can push this forward, turning diagnostic guesswork into precision and doubt into dignity.
Dollars Saved Lives Reclaimed
Imagine cutting the misdiagnosis rate for autoimmune and neuroimmune conditions even by a fraction—say, from 25% to 15%. Economically, this could save tens of billions annually, trimming the $100 billion-plus burden of wasted healthcare costs like unnecessary tests, wrong treatments, and emergency visits while boosting productivity as patients return to work sooner. For individuals, the impact hits deeper: years shaved off the 4.6-year diagnostic odyssey mean less financial strain, potentially thousands saved per patient, and faster access to treatments that work, turning exhaustion and isolation into hope and healing. Though speculative until this AI tech scales, the potential to ease both wallets and suffering makes it a game-changer worth watching.
Unhide: Amplifying AI’s Potential With Patient-Driven Data
What if these researchers could tap into a wealth of real-world patient data to supercharge their AI models for conditions like PANDAS/PANS, ME/CFS, and Long COVID? This is one of many powerful ways our clinical research platform Unhide could come into play. This free tool empowers patients to voluntarily share their health records, symptom histories, and even tissue samples like blood, creating a rich, diverse dataset for scientists. By providing access to these samples and insights from the very community desperate for answers, Unhide could help researchers like Dr. Scott Boyd and his team train AI frameworks like Mal-ID to recognize the unique immune signatures of specific neuroimmune and autoimmune diseases beyond the study’s initial scope. It’s a potential game-changing bridge between patients and progress.
Written By
Matthew Menendez, Ph.D.
Director of Science Communication, Brain Inflammation Collaborative
Dr. Menendez has a passion for empowering patients navigating complex chronic diseases. Through innovative educational strategies, he bridges the gap between cutting-edge research and real-world understanding, helping individuals advocate for better care, access timely treatments, and reclaim their quality of life.
