The Road to Predictive Immunology:

Data, AI, and the Next Generation of Clinical Diagnostics for Rare Immune Disorders

Nazanin Fathi, Ph.D.   |   Immunology Visioneer, 20/15 Visioneers   |   July 2026

Summary: Among the most diagnostically complex conditions in clinical medicine are diseases in which the immune system fails—not because of infection or cancer, but because of underlying genetic defects that impair immune development, signalling, or function. Inborn errors of immunity (IEIs), a group of more than 550 genetically defined disorders, represent one of the most compelling examples of this challenge. Despite decades of research and major advances in genetic diagnostics, immunophenotyping technologies, international collaboration, and knowledge sharing, diagnostic workflows for IEIs remain fragmented, non-standardized, and heavily dependent on individual laboratory expertise. Artificial intelligence is now demonstrating significant potential to address this gap—but only where high-quality data, well-designed laboratory workflows, and scientific standardisation are already in place. This article explores the intersection of AI, immunophenotyping, and laboratory data strategy through the lens of rare immune diseases, arguing that the primary bottleneck is not the algorithm itself, but the data infrastructure behind it. 

A Framework: The Five Stages of AI-Ready Immunophenotyping

Before exploring the science and market landscape in depth, it is helpful to establish a shared roadmap. The path from a traditional clinical flow cytometry laboratory to a fully AI-enabled immunophenotyping environment involves five distinct stages — each building on the last, and each requiring specific investment in data infrastructure, workflow design, and organizational change. We refer to this framework as the Five Stages of AI-Ready Immunophenotyping—a pathway laboratories seeking to implement AI effectively will need to navigate to move from fragmented data capture to predictive diagnostic capability.

Many clinical immunology laboratories today operate at Stage 1 or early Stage 2. The AI tools currently being developed—and many commercial platforms already on the market—require organisations to reach Stage 3 before they can consistently deliver reliable value. The gap between where many labs are and where AI requires them to be is the central challenge — and the central opportunity — of the next decade.

Acknowledge “Accurate IEI diagnosis requires multiple layers of evidence — clinical manifestations, immunophenotyping, genetic testing, and functional assays — interpreted together. Each layer contributes something the others cannot. Genetic diagnostics have advanced significantly in recent years, with large, well-curated databases now supporting precision medicine approaches. Clinical assessment, however, remains inherently variable and is often fragmented across institutions, and functional assays are still in early stages of standardization.

Immunophenotyping — primarily performed through flow cytometry — occupies a unique position in this landscape. It sits at the intersection of instrument-generated data and human interpretation — partly machine, partly specialist judgment. This duality makes it both highly informative and highly variable. It is also the layer most amenable to AI standardisation right now: the data is structured, the parameters are measurable, and the patterns are learnable. For these reasons, this article focuses on immunophenotyping as the entry point for AI integration in IEI diagnostics — not because the other layers are less important, but because this is where the opportunity for immediate, scalable impact is greatest.”

The Problem Nobody Talks About

The most expensive mistake in clinical immunology is not a failed experiment. It is a delayed diagnosis caused by data that exists but cannot be compared. Imagine a nine-year-old child who has had six episodes of pneumonia over the past three years. She has been treated for each episode. She has recovered each time. But the underlying question remains unanswered: why does she keep getting sick?

This scenario is not rare. In IEI — a heterogeneous group of over 550 genetically defined immune disorders — the average diagnostic delay ranges from 4 to 9 years, depending on the condition and the healthcare system. During that time, patients accumulate organ damage, receive inappropriate treatments, and face avoidable mortality. For healthcare systems, these delays also translate into repeated investigations, avoidable hospitalisations, increased healthcare costs, and inefficient use of specialist resources. The most recent phenotypic classification now covers 559 IEI entities, up from just a few dozen in the 1980s — reflecting both continued gene discovery and an improved understanding of the clinical and genetic diversity of these disorders. 

The clinical presentation of IEI is extraordinarily heterogeneous. The same genetic defect can produce markedly different phenotypes across patients. A patient with CVID — the most common IEI in adults — may present with recurrent respiratory infections, autoimmune manifestations, lymphoproliferation, or granulomatous disease. Another patient with an NF-kB pathway disorder may look clinically identical but require completely different treatment. Getting the diagnosis right is not simply a matter of clinical intuition. It requires data—specifically, high-quality immunophenotyping data from flow cytometry, interpreted in the context of clinical history, genetic testing, and functional assays. At scale, consistently collected and standardized data enable AI to identify patterns across patient populations, supporting faster, more consistent, and data-driven diagnostic decision-making.

"The bottleneck in IEI diagnosis is rarely the absence of technology. It is the absence of standardized, AI-ready data to feed that technology."

And this is where the problem starts. Because in many clinical immunology laboratories worldwide, the data simply do not exist in a format that AI can use. 

For healthcare and life science organizations, this is more than a scientific challenge. It is an operational and strategic one. Without standardized, interoperable laboratory data, organizations cannot fully realize the value of their investments in digital health and AI.

What Flow Cytometry Generates — And Why It Is So Hard to Use

Flow cytometry is the cornerstone of IEI diagnostics. It measures the composition, proportion, and functional characteristics of immune cell populations — T cells, B cells, NK cells, monocytes, and their countless subsets — with extraordinary precision. Modern spectral flow cytometers can simultaneously measure more than 40 parameters across millions of cells within minutes. The data they generate are rich, nuanced, and clinically irreplaceable.

But this same richness is the source of the problem. A single flow cytometry panel for IEI evaluation can generate dozens of cell population percentages, absolute counts, ratios, and expression levels — each of which must be interpreted against age-matched reference ranges, clinical context, and the evolving classification of IEI. Multiply this across a laboratory running hundreds of samples per month, using panels that differ from those used in the next hospital, interpreted by specialists who may disagree on gating strategies, and stored in legacy LIMS systems that were never designed for computational analysis — and you begin to understand why IEI diagnosis remains so difficult to scale. 

The core challenge is not the instrument. It is everything around it.

· Panel designs are not standardized across institutions (despite initiatives such as EuroFlow PIDOT), making cross-laboratory comparisons difficult.

· Gating strategies vary by operator, introducing inter-observer variability that can change a clinical interpretation

· Reference ranges differ by centre, age cohort, and the platform used

· Data are typically stored in FCS files that require specialist software to access — not in formats that integrate with clinical data systems or AI pipelines

· The FAIR principles — Findable, Accessible, Interoperable, Reusable — are almost universally absent from clinical flow cytometry environments

Recently publications in Clinical Immunology have demonstrated that large language models can enhance efficiency and consistency in flow cytometry reporting for primary immunodeficiency diagnostics — but the authors were explicit: The findings reinforced a broader principle: AI performance depends heavily on the quality and consistency of the underlying data. Where the input was messy, ambiguous, or non-standardized, the AI produced unreliable outputs. The conclusion was straightforward: garbage in, garbage out — even with the most sophisticated model.

Where AI Is Already Making a Difference

Despite these challenges, the field is moving. And the early results are genuinely encouraging — not because AI has solved the standardization problem, but because it has demonstrated what becomes possible when that problem is addressed.

DeepFlow, a commercially available AI platform, demonstrated automated analysis of a 3-tube, 10-color panel for primary immunodeficiency diseases across 379 clinical cases — completing analysis in under 5 minutes per case with performance comparable to expert hematopathologist interpretation. The key to its success was not the algorithm alone. It was the fact that the input data was produced using a defined, consistent panel with standardized gating logic.

At the Mayo Clinic, AI-enhanced flow cytometry analysis for measurable residual disease detection in chronic lymphocytic leukemia demonstrated high sensitivity for atypical immunophenotypes, including cases where standard markers such as CD5 were absent. Again, the performance was enabled by a high-quality, structured data environment built over years of clinical practice.

Ozette Technologies and Hema.to have built AI platforms specifically for comprehensive immune profiling and automated population classification in clinical flow cytometry. Both are developing AI-enabled platforms for clinical flow cytometry applications. Both require, as a precondition, that laboratories produce consistent, well-annotated data.

"Every successful AI application in immunophenotyping has one thing in common: someone built the data infrastructure first."

The IEI Diagnostic Gap Is a Data Strategy Problem

Having spent several years in clinical immunology research — working directly with patients with IEIs, designing flow cytometry panels for antibody deficiencies and lymphocyte subset analysis, and publishing on conditions ranging from CVID to NF-kB pathway disorders — I have seen this problem from the inside.

The IEI field is not short of scientific knowledge. The diagnostic decision trees are detailed and evidence-based. The genetic tools — whole-exome sequencing and targeted gene panels — are increasingly accessible. And yet the average diagnostic delay remains measured in years, not months.

When I examine why, the answer almost always comes back to the same place: the clinical data generated during those years of diagnostic wandering is not connected. A patient who has had immunological workups at three different hospitals may have three sets of flow cytometry results that cannot be compared because the panels were different, the gating was different, and the data format was different. AI cannot help if it cannot read the data. And the data cannot be read if it was never designed to be.

This is not a criticism of clinicians. It is a systemic challenge that requires a systemic solution.

Specifically:

· Standardized immunophenotyping panels for IEI — agreed upon at a network or national level — that produce comparable data regardless of which laboratory runs the sample

· FAIR data environments in clinical immunology laboratories — where flow cytometry data is not just stored but findable, accessible, interoperable, and reusable

· Workflow design that integrates the data capture step into the diagnostic process — not as an afterthought but as a core clinical output

· AI-readiness assessments for clinical immunology departments — understanding what data they have, what format it is in, and what infrastructure investment is needed before AI tools can be deployed

What This Means for Life Science Organizations

The implications extend well beyond clinical immunology departments. Healthcare organizations are increasingly investing in AI to improve diagnostic accuracy, operational efficiency, and clinical decision-making. However, the value of those investments depends on robust data foundations. Without standardized, interoperable laboratory data, even the most advanced AI systems cannot deliver reliable clinical or business outcomes. The same data infrastructure challenge that delays IEI diagnosis affects drug development, clinical trials, biomarker research, and commercial diagnostics.

Pharmaceutical companies conducting clinical trials in oncology, immunotherapy, autoimmune diseases, and rare immune disorders rely on immunophenotyping data as a key efficacy readout. The quality of that data directly affects the reliability of their clinical decisions. A lack of harmonization in a multi-centre trial can compromise data comparability, delay analysis, and ultimately slow regulatory submissions.

Biotech companies developing next-generation diagnostics for immune disorders face the same problem from the other direction: their AI models are only as good as the training data available. And that training data — clinical flow cytometry from real IEI patients — is currently locked in incompatible formats across hundreds of institutions worldwide.

The market opportunity is significant. More importantly, it is not primarily a technology challenge—it is a challenge of workflow design, data strategy, and scientific standardisation. This is precisely where 20/15 Visioneers helps life science organisations transform laboratory data into AI-ready infrastructure.

"The next breakthrough in immunology diagnostics will not come from a better algorithm. It will come from the organization that finally standardizes the data."

A Practical Framework: From Data Chaos to AI Readiness

For life science organizations — whether clinical laboratories, biotech companies, or research institutes — beginning the journey toward AI-enabled immunophenotyping requires a structured approach. Informed by current best practices in laboratory digital transformation and practical experience supporting organizations, the following framework provides a practical starting point.

Step 1: Audit your current data environment

Before any AI tool can be evaluated, an organization should first conduct an AI readiness assessment. This means understanding what data it actually has by mapping immunophenotyping workflows, identifying where data is stored, evaluating data quality, interoperability, workflow maturity, governance, and digital capabilities, and assessing the consistency of panel design, gating strategies, and data accessibility. Most organizations discover at this stage that their data is far less structured than they assumed. 

Step 2: Define your FAIR data strategy

A FAIR data strategy for immunophenotyping does not require replacing existing systems. It requires designing the data capture process — from panel design to results reporting — with interoperability in mind. This means standardizing the annotation of cell populations, establishing consistent reference ranges, and ensuring that data is stored in formats that can be computationally accessed and integrated with downstream analytical systems.

Step 3: Standardize before you automate

The most common mistake organizations make is attempting to deploy AI tools before their data environment is ready. The result is invariably disappointing — not because the AI does not work, but because the data they are fed do not consistently reflect clinical reality well enough for the models to learn from them. Standardization of panels, gating, and reporting must come first.

Step 4: Pilot in a controlled environment

AI tools for immunophenotyping should be piloted against a dataset that has been specifically curated for the purpose — with known clinical outcomes, standardized panel designs, and expert-annotated gating strategies. This provides a realistic assessment of performance and identifies the specific failure modes that need to be addressed before broader deployment.

Step 5: Build the human infrastructure alongside the technical

The most overlooked element of any AI implementation in clinical immunology is the change management required to make it sustainable. Scientists and clinicians must understand what the AI is doing and why — and must be equipped to identify when it is wrong. This requires training, clear governance, and a culture that recognizes data quality as a shared organizational responsibility rather than a technical afterthought.

Emerging Market Landscape: Who Is Building the Future

The commercial ecosystem around AI-enabled immunophenotyping and rare disease diagnostics is still early — but it is accelerating. Understanding who is building in this space, and where the gaps remain, is essential for any life science organization evaluating its own positioning.

Category 1 — AI-Assisted Flow Cytometry Platforms

The most mature commercial category. Tools like DeepFlow (DeepCyto), Ozette Assay-to-Insights, and Hema.to are already in clinical or near-clinical deployment. These platforms automate population identification, reduce analysis time from hours to minutes, and flag subtle anomalies that may be overlooked during routine expert review. The key differentiator is not the algorithm — it is the depth of the training dataset and the breadth of validated clinical panels.

Category 2 — Comprehensive Immune Profiling Platforms

A broader category that encompasses multi-omic immune characterization — combining flow cytometry with transcriptomics, proteomics, and genomic data to build integrated immune profiles. Companies like Immunai are building in this space, primarily for pharma R&D and clinical trial biomarker programs. The data requirements are even more demanding than single-modality flow cytometry — making FAIR data infrastructure an absolute prerequisite.

Category 3 — Clinical Decision Support for Rare Diseases

An emerging category specifically targeting the diagnostic journey for rare immune disorders. Tools here combine genetic, clinical, and immunophenotypic data to guide clinicians toward the correct IEI classification more rapidly. The International Union of Immunological Societies (IUIS) phenotypic classification—now available as a smartphone-accessible decision tree covering 559 IEI entities—provides an expert-curated framework that AI-assisted tools in this category aim to translate into scalable clinical decision support. The market opportunity is substantial, given that the average IEI diagnostic delay of 4–9 years imposes significant costs on healthcare systems and immeasurable burdens on patients.

Category 4 — Rare Disease Diagnostic AI for Drug Development

Pharmaceutical companies developing therapies for IEI — including selected signaling pathway disorders, primary immune regulatory disorders, and other genetically defined immune diseases— require validated immunophenotyping biomarkers to demonstrate efficacy in clinical trials. The companies building AI tools to standardize and analyze these biomarkers across multi-centre trials are addressing a significant unmet need. Investors in this space include specialist life science funds with strong interest in rare disease platforms.

"The organizations that build the data infrastructure now will own the AI advantage in immunology for the next decade."

What is notably absent from all four categories is a widely adopted, cross-institutional standard for IEI immunophenotyping data. The field is still waiting for the equivalent of what HL7 FHIR achieved for healthcare interoperability — a common language that allows AI models to learn across institutions rather than within them. This is the gap where scientific leadership, standardization expertise, and commercial strategy intersect. 

" At 20/15 Visioneers, we help life science organizations bridge the gap between scientific expertise and AI-ready data infrastructure. Our consulting services in workflow design, data strategy, and laboratory digitalization are built for organizations that have the science but need the foundation to make AI work."

Conclusion: The Opportunity Is in the Infrastructure

Artificial intelligence will transform immunophenotyping and IEI diagnostics. The evidence from early deployments is clear: when the data is right, the results are impressive. Diagnostic workflows compressed from days to minutes. Inter-observer variability eliminated. Rare immunophenotypes identified that may be overlooked during routine expert review.

But the data is not yet right. Not in a large proportion of clinical immunology laboratories. Not in many biotech companies conducting immunology trials. Not in most research institutes studying rare immune disorders.

Closing that gap — building the data infrastructure, standardizing the workflows, designing the FAIR data environments that AI requires to deliver on its promise — is the work of the next decade. It is scientific work, technical work, and change management work simultaneously.

It is also, I would argue, some of the most important work in medicine right now. Because behind every delayed IEI diagnosis is a child with six pneumonias and a question nobody thought to ask.

We have the tools to ask it. We need to build the infrastructure to hear the answer.

“At 20/15 Visioneers, we help life science organizations build the data foundations that make AI-ready diagnostics possible. If your organization is navigating this journey, we would welcome the conversation."

Author

Nazanin Fathi, Ph.D. is a Medical Immunologist and Scientific Consultant & Business Developer at 20/15 Visioneers, based in Barcelona, Spain. She specializes in inborn errors of immunity (IEIs) and works at the intersection of laboratory science, AI-enabled diagnostic innovation, and digital transformation in the life sciences. 

Contact: nazanin.fathi@20visioneers15.com

20/15 Visioneers | www.20visioneers15.com | info@20visioneers15.com